We have spent the last several weeks working through the optical supply chain and the manufacturing steps that will set the pace of optical content inside AI systems. Demand is the easy part of the forecast everyone would adopt optical today if could be manufactured and was reliable at scale. The industry needs more bandwidth between accelerators, and copper hits a physical limit as link speeds go up and scale-up domains get bigger. The harder question, and the one we care most about, is how fast the supply chain can turn that demand into qualified optical engines that ship at repeatable yield.

In our inter interview Lightmatter CEO Nick Harris gave us a closer look at one of the more differentiated attempts to solve that problem. Most commercial lasers are still made on small indium phosphide wafers by a fairly small group of suppliers, and Nick makes the case that Lightmatter can move more of that work onto the 300-millimeter silicon base mature CMOS foundries already run. If that holds up at volume, it opens a different path to laser capacity. He walks through how the approach works and how Lightmatter is designing around the reliability problem that comes with thousands of optical links in one system.

He also gives his own timeline for when near-package and co-packaged optics ramp, which we found useful to hold up against our optical attach ladder, along with the reliability data Lightmatter has built up so far. As always, we would separate a product shipping from broad attach across deployed racks, and that gap is a big part of what we dig into with Nick Harris. The full video, audio and edited transcript are available to subscribers below.

Subscribers with CS Atlas access can deeper on Lightmatter and the wider optical transition. The Lightmatter packet breaks down Passage and Guide and places the company on our timing ladder from NPO through CPO and the longer-term photonic-interposer roadmap. A separate manufacturing section tests the 300-millimeter claim against the yield and qualification steps that still have to be proven.

Subscribers can then use our optical-solutions comparison and interconnect SWOT to see which companies control each part of the stack, how their products differ, and where they are focused across scale-up, scale-out, and scale-across networks. Each company entry maps the manufacturing unit it must scale and the evidence we need before moving its timing or competitive position higher.

For Subscribers

• Full video interview

• Why Lightmatter views interconnect as the problem that now sets the pace of compute.

• How Passage and Guide address optical links from NPO through a photonic interposer.

• How bonding indium phosphide onto 300-millimeter silicon photonics wafers changes the manufacturing approach.

• Why reliability, optical test capacity, and foundry process control still govern scale.

• Where Nick expects NPO, CPO, and photonic interposers to enter the deployment curve.

Read more

In our Bloom Energy report, we looked at behind-the-meter power as a way to get data centers running sooner while customers wait on utility service. We wanted to take that question to Billy and get a better sense of what it actually takes to make these projects work. We’ve made the point in prior reports that customers may be willing to pay more for earlier access to power, and we still think that’s true. But the developer still needs an agreement that covers the cost of building and operating the plant.

Billy spent a good chunk of our conversation on a problem we had given less attention to. The power plant is a small share of the total investment at a compute site, but an outage can put a much larger investment at risk. So the question becomes how much of that risk the power developer can afford to take on. Adding spare generation or batteries costs money, and the customer’s contract has to run long enough for the developer to earn it back. That gives us another way to look at the BTM opportunity from the Bloom report: what reliability the customer needs, what it costs to provide, and whether the two sides can agree on who pays when something goes wrong. It also adds a piece we were missing in our TCO analysis and in our Gigawattonomics model, as we track AI capex ROI all the way down to the power shell provider.

The conversation also helped us understand another challenge with running data centers on natural gas, which is getting enough gas to the site. The US has plenty of gas, but a data center may still need a pipeline extension or more compression before it can use that supply. The companies building that infrastructure need customer commitments to justify the investment, so it becomes one more project that has to be funded and built on a schedule that lines up with when the data center needs power. It’s the same pattern we keep seeing, with the constraint moving further down the stack.

Below, we get into how these projects come together, why the contracts are hard to structure, and where Billy sees the next constraints showing up.

Our companion report develops the investment implications through project examples, gas-demand scenarios, and the economics of dedicated generation and who is best positioned. Subscribers, with Atlas access can explore our full power landscape research, and BTM / natural-gas and compute research alongside the interview.

In the conversation:

• Why generation, power delivery, and the customer’s opening date are difficult to align.

• What happens when a power supplier cannot buy replacement electricity during an outage.

• How reliability requirements and contract length affect the price of power.

• Why pipeline companies need credible customer commitments before construction begins.

Read more

In Secret Agent CPU, Revisited, published May 26, we raised that forecast. Our conversations across the industry were strengthening our conviction that dedicated CPU capacity would become an essential part of running agentic systems at scale. We remain bullish on CPUs because of the clarity in inference workload demand and the clear role the CPU plays becoming a larger infrastructure requirement.

But while we talk a lot about the CPU-to-GPU ratio, it is very hard to isolate that number and turn it into a demand forecast. The mix depends on the infrastructure operator and the decisions it makes about serving AI. Key point here, each hyperscalers ratio will be different. There will be no standard CPU/GPU ratio configuration. The main key here is to understand those decisions, we need to understand the scale-up domain and why this is priority number one for serving inference at scale and monetizing compute infrastructure.

Why we add “domain” to scale-up

A lot gets talked about scale-up, which involves connecting accelerators so they can work closely together in a single compute fabric. We add “domain” firstly because its a helpful mental model to understand networked infrastructure topology and because we want to “connect” how much networked compute and memory can cooperate efficiently on the same workload.

The goal is to make more compute and memory work tightly together as one domain. Today, one way to think about that is a rack whose accelerators cooperate to serve a model to many customers at the same time. Depending on the model and workload, that can mean thousands of simultaneous users, although there is no fixed number of users per rack. Longer conversations require more memory, while more demanding reasoning can occupy the processors for longer. Operators want to expand that domain where doing so lets them serve larger models or more users economically. Across the broader datacenter, they also connect multiple domains, without requiring every processor to participate in one tightly coupled system.

There are two ways this creates demand for more racks. An operator can run additional copies of the model and distribute customers across them. Or a model can require more resources than one rack provides efficiently, which means multiple racks must cooperate on its workload execution.

Extending the scale-up domain across racks gives the operator more tightly connected resources to work with. Models can also run across separate domains through scale-out networking, although the communication cost changes. Connecting racks alone does not establish that they form one scale-up domain.

This is where model size and model context enter the conversation. The model’s weights take up memory, and serving active users requires additional working memory. For many models, the KV cache holds information from the tokens already processed so the model can continue generating its response. Longer context and more simultaneous sessions increase that requirement.

A larger domain gives an operator more room to place the model and its active state. Depending on the workload, that can make a larger model practical to serve or improve throughput at the response time customers expect. The operator still has to balance those benefits: a larger model or longer context can consume capacity that otherwise serves additional users.

And shared memory needs a qualification. Memory remains distributed across processors, with a cost to accessing it remotely.More connected compute does not, by itself, make the same model smarter or more capable. It gives operators more capacity to deliver the capabilities of the models they choose.

Interconnect determines how far the domain can grow

Our view is that hyperscalers and neoclouds have a strong incentive to expand these domains where the additional scale improves serving economics. Some enterprises will face the same decisions as they deploy larger AI systems on-premises.

The limit is how many accelerators can exchange data fast enough to work efficiently together. When a model runs across multiple accelerators, each needs results from the others to continue its work. If those results arrive too slowly, adding accelerators can leave more expensive hardware waiting for data.

Copper has limits around reach and signal integrity as speeds increase. The underlying boundary depends on the architecture, including its physical layout and power budget.

Optical interconnects expand the possibilities for reach and bandwidth density, which is why we have spent so much time on this part of the market. But the optical system still has to be manufactured, qualified, and serviced. In Optics Won’t Scale as Fast as the Market Expects, we explained why demand for optical connectivity can move faster than qualified supply.

The race is to build larger scale-up domains that can serve larger models and more users concurrently and economically. If the processors spend too much time waiting for data, operators are paying for compute they cannot fully use. How well each company solves that problem will help determine its margins from serving AI at scale.

Where the CPU changes the equation

The CPUs already present in GPU systems give us a baseline ratio. For example, NVIDIA’s GB200 NVL72 contains 36 Grace CPUs and 72 GPU packages, or one CPU for every two GPU packages. Counting the dies inside those packages would produce a different number, so we need to keep the denominator consistent.

Where the ratio conversation becomes more interesting is when we add dedicated agentic CPU capacity which will manifest itself as dedicated racks of CPUs in the scale up domain of GPU compute racks.

This is why we separate head-node CPUs from dedicated agentic CPUs in our models. While each customer building their own compute racks, like AWS, Azure, and Google, can optimize the number agentic CPUs for their own unique needs the template we have today is of NVIDIA’s Vera CPU Rack which supports up to 256 CPUs and is positioned alongside its accelerator systems. Arm’s AGI CPU also targets agentic infrastructure with AMD, Qualcomm, and Intel following suit.

We do not yet know how much of that capacity every operator will deploy or what their scale up domain mix of CPU/GPU/premium accelerators will be. It could be one CPU rack supporting several GPU racks, or more, with the allocation changing as the workload, or economics, changes.

Consider a useful, but hypothetical, example deployment of a scale up domain of ten GPU racks, each with 72 GPU packages and 36 host CPUs. That gives us 720 GPU packages and 360 CPUs. Adding a dedicated rack containing 256 CPUs takes the total to 616 CPUs, or about 0.86 CPUs per GPU package. Two dedicated CPU racks take it to 872 CPUs, or about 1.21 to one.

That is math using stated configuration assumptions, not a deployed-system claim or a forecast. We highlight this use case to show how varied potential configurations can be and why trying to estimate any CPU:GPU ratio is not the right analysis or measurement of the market.

Demand still has to fit what can be manufactured

Our CPU forecast (available in institutional tier of CS Atlas) separates general-purpose servers, accelerator head nodes, and standalone agentic systems along with key vendor share assumptions. We then constrain shipments using our manufacturing assumptions since our forecasts are grounded also in what’s manufacturable.

Our most current base case implies the following growth from a 2026 baseline:

• 2027: CPU silicon-value growth of 29%; CPU package shipment growth of 20%.

• 2028: CPU silicon-value growth of 33%; CPU package shipment growth of 22%.

• 2029: CPU silicon-value growth of 25%; CPU package shipment growth of 11%.

• 2030: CPU silicon-value growth of 35%; CPU package shipment growth of 29%.

These are conditional, and continually updated, Creative Strategies estimates, grounded in our most current assumptions and industry checks. Silicon value includes an imputed value for captive CPUs, so it is broader than merchant CPU revenue. Pricing and product mix explain why dollar growth can exceed shipment growth given a core assumption we are making on ASP for agentic CPUs specifically.

The model constrains CPU output by compatible foundry-node capacity and allocation after competing demand, then applies yield, assembly, and substrate assumptions. Its 2029 slowdown and 2030 acceleration depend heavily on node transitions and the capacity those transitions make available. For customer/clients with access to our data models we break down the full assumptions in base/bull/stretch scenarios.

We remain aggressive in our forecasts but also grounded in manufacturing reality. Wafer mix assumptions matter and we cannot assume every wafer is available to CPUs while GPUs and custom accelerators are competing for capacity. Nor does the model establish the industry’s absolute manufacturing ceiling. We will continue refining these assumptions as new information becomes available on customer allocations and system-level deployment constraints.

Any ratio estimate has to be grounded. We would therefore be careful about declaring that a two-to-one CPU-to-GPU ratio, or more, is either inevitable or impossible. It first needs a defined workload and counting convention, then a test against the supply model aligned with tangible tracking of hyper scaler scale up domain mix. Demand above our forecast and demand above feasible supply are different conclusions.

Each operator is making a hardware bet

Hyperscalers make distinct decisions about the infrastructure they want to build, working with system suppliers and manufacturing partners. They choose the balance of merchant and custom silicon, then determine the memory and networking required to make those processors productive within a TCO profile.

It is important to note how one operator may prioritize flexibility across many customer models. Another may have enough predictable internal demand to justify a more specialized accelerator. They can also assign different stages of inference to different systems when the benefit outweighs the cost of moving state between them. Every implementation is unique and many times bespoke which is why generalities harm more than hurt the analysis.

Mixing merchant and custom infrastructure across a fleet is different from making those processors cooperate inside one tightly connected domain. That requires compatible hardware and software. These decisions commit capital to assumptions about workloads that are still evolving, and those assumptions help determine the margin profile of serving AI at scale.

Our conviction is that enterprise and personal agents will create much more work around the model layer in the scale up domain. We will have many simultaneous sessions calling on CPUs and then returning to accelerators, with memory retaining the state needed to continue. The amount of dedicated new CPU infrastructure depends on how much existing capacity absorbs and how efficiently the software schedules that work. Another reason, we are bullish on the growth profile of AWS, GCP, and Azure.

For who benefits here in silicon, we have high conviction on both AMD and Intel, with Intel standing to doubly benefit with Intel Foundry well positioned to also manufacture CPUs as additional capacity is brought online. We maintain NVIDIA is also a primary beneficiary with their agentic CPU roadmap, and Arm and Qualcomm also coming online with solutions. This is not the server CPU TAM of old where share taking was the fundamental analytical baseline. This is a significant TAM expansion and the analysis now who is are the unit volume and price share gainers in a market that is growing in dollar value ~30% YoY in our supply constrained forecast.

What would make us less bullish is evidence that CPU work per completed task falls enough to offset usage growth, or that existing infrastructure absorbs much more demand than we expect. For now, we believe the expansion of agentic workloads supports more CPU capacity, with operators making different decisions about how much belongs around each accelerator deployment.

That is how we think about the CPU-to-GPU relationship: through the complete inference system and how its scale-up domains shape what operators can manufacture and deploy economically. Keeping that view grounded is our ongoing research goal as hardware choices evolve and we learn more about how these systems perform in practice.

Continue the research in Atlas

Subscribers with CS Atlas access can use Atlas to connect our CPU reports with the research on networking and optical supply. Ask how dedicated agentic CPU racks change the processor mix, which assumptions constrain growth, and what evidence would change the forecast.

Explore this research in Atlas

The link opens an editable question after sign-in. Answers depend on the research and model versions available with your access.

• CPO Summit Takeaways

• 800 VDC Transition and Timelines

• Learnings from AWS Meetings on Compute and Model Trends

We have spent time over the last month in conversations with stakeholders at different hyperscalers and have landed on a conviction about how infrastructure decisions are likely to play out during this AI buildout. We believe the future of compute is fungible. While we appreciate NVIDIA’s framing that the GPU is fungible, and that is true, it is clear to us that hyperscalers have a vested interest in building their data centers with workload fungibility in mind.

This viewpoint is an important baseline observation on the diversity of solutions they will use, spanning merchant compute, custom compute, copper, optical, liquid and air cooling, and power delivery from 800 VDC to older generations. We expect these technologies to coexist across their infrastructure, with software coordinating the work across that diversity as one compute fabric. Note, this is a particular advantage unique the hyperscalers and we believe will be a key part of their continued competitive advantage in designing AI infrastructure at scale and competing for enterprise workloads. Having this perspective raises a number of implications and questions that our report explores.

The value of this viewing infrastructure choices this way is in how it changes our assessment of the companies supplying this buildout. More choice gives customers another way to buy and build capacity, but the savings and supplier profits depend on which workloads move, what it costs to support them and who continues to supply the rest of the system. We work through six questions to understand where those economics are likely to improve and where the apparent benefit of greater choice breaks down.

1. How does greater choice change the customer’s purchasing power?

We expect hyperscalers to start by using a wider mix of systems for new workloads while keeping existing work on the hardware already running it. That includes buying from other suppliers and developing their own silicon and rack-scale designs, giving them more control over how the systems are built and which parts they buy. At first, these efforts may help them add capacity and better match hardware to their workloads. As the alternatives reach production at scale, customers gain more room to negotiate prices. We expect those benefits to develop before broad pricing pressure affects suppliers’ growth and margins.

2. Where does specialization offer better economics, and where does GPU flexibility remain more valuable?

Workload fungibility gives operators more choice over where they run their work. We expect custom silicon to gain ground where steady demand justifies the cost of designing and supporting it, while GPUs remain valuable for a changing mix of workloads. The benefit comes from matching work to the systems that can run it most economically. That requires enough software support to make those choices practical. A cheaper chip offers limited savings when it cannot run enough of the customer’s work or takes too long to get into service.

3. How much influence moves to the software that decides where the work runs?

The software deciding where work runs also influences which hardware the operator needs to buy or they design custom compute. We expect hyperscalers to keep considerable control over those decisions because better utilization and more purchasing options directly benefit their own businesses. Suppliers still have an opportunity where their software reduces the work required to deploy and operate a system. The distinction is who collects the economic benefit. More important orchestration software does not necessarily mean a larger revenue opportunity for an independent software company.

4. How does serving inference across larger compute domains change the network?

As hyperscalers connect more systems to serve inference at scale, the network helps determine how much of that capacity they can use together. Workload fungibility depends on being able to send work to available compute without adding too much delay or cost. This raises questions about where copper remains sufficient, where optical connections become necessary, and how much flexibility operators can gain while keeping tightly connected systems running efficiently. We examine what those choices mean for networking and optical suppliers as operators build larger compute domains.

This builds on our ongoing optical and networking research and our recent conversation with a senior AWS executive, covered in Atlas. Larger, sparse models need fast connections between the accelerators holding their different parts. The race is on for the larger scale up domain where optical links are necessary extend those connections across more racks (bigger compute clusters), giving operators more room to spread power and cooling demands. The return depends on whether the larger system serves more work at the response times customers need, after accounting for the cost of connecting it.

5. What is infrastructure flexibility worth over the life of a data center?

Power and cooling decisions made today will affect what an operator can install several equipment generations from now. We see a stronger case for spending against a defined upgrade path than for preparing every part of a facility for the highest possible density. The extra investment has to earn its way back through lower modification costs, less downtime or getting paid for new capacity sooner. Our facility analysis shows why the return changes substantially when earlier installation does not produce earlier income, and why the lease matters as much as the engineering.

6. Which companies retain the most value as operators gain more options?

We keep coming back to the work a supplier retains as the customer gains more options. Another company may supply the accelerator while NVIDIA continues to sell valuable infrastructure around it. A custom-silicon partner may win a design but retain less of the profitable work when it reaches production. This is why we assess implementation, proprietary IP, networking and support separately. It also explains how greater customer choice can pressure pricing and still support profit growth for suppliers that remain difficult to replace.

In the full report, we examine:

• When cheaper compute costs more. We test how a lower-cost system’s apparent advantage holds up once utilization, software compatibility and deployment delays enter the calculation—and where paying more for GPU flexibility or integration produces a better return.

• What supports our company preferences. Why Broadcom remains our largest modeled dollar opportunity in custom silicon, why MediaTek leads our incremental conviction ahead of Marvell, and what would change that order.

• How NVIDIA participates alongside custom compute. Which parts of the system can remain valuable to NVIDIA even when a customer chooses another accelerator.

• Which developments would change our view. The production orders, retained supplier work and completed facility upgrades we will assess over the coming quarters.

Take the analysis further in CS Atlas. Explore seven company assessments and three illustrative models, then connect the findings with our broader industry research. Ask which suppliers benefit as hyperscalers design more of their own infrastructure, compare company opportunities, and test what would change our views. We built Atlas so subscribers and clients can work through our research on their own terms and apply it to the decisions they face.

Not yet a paid subscriber? Join The Diligence Stack to access the full report and go deeper with our research in CS Atlas.

Want to engage more with our firm and our analysts, reach out here.

Read more

Our interconnect SWOT is also available in CS Atlas, where subscribers can explore how we assess Credo alongside its competitors. In our July scorecard, Credo received our highest score for current-cycle strength, its role in addressing key connectivity bottlenecks, and its opportunity in scale-up systems. This report builds on that work, including why we think owning more of the copper and optical design gives Credo room to grow and what we still need to see as those products reach customers.

The companion research is available to subscribers with CS Atlas access.

We will have full coverage of AI Infra Summit in Atlas as well this week so check CS Atlas later in the week for our event notes.

Why we are positive on Credo

We are big believers that vertically integrated companies have more sustainable competitive advantages than their less integrated peers. This is a key reason we have been positive on Credo from the start, and while management consistently articulates this, we still think it is not given enough weight in the analysis of Credo’s competitive position. While Credo still has a lot of runway in its AEC and ALC businesses, we think the market is undervaluing what DustPhotonics adds as interconnect designs move toward NPO, CPO and scale-in.

DustPhotonics gives Credo control of the photonic chip alongside its own SerDes, optical DSP and firmware. We think owning the SerDes is an important part of this advantage because Credo can optimize how the electrical signal is sent and received as it works through the optical design. When a customer needs a connection to use less power, its engineers have more choices about where to make that improvement and how to balance it against reach and reliability. They can work across more parts of the electrical and optical parts together, including how a change in one affects the other. We think that flexibility becomes more valuable as customers move to new architectures and look to integrate heterogenous compute diversification, giving Credo more ways to meet their requirements and compete for the next design.

We also think Credo’s position as a connectivity-focused company with both copper and optical products deserves more weight than is given. Several competitors are stronger on one side of that market, while broader platform suppliers approach connectivity as part of a larger system. Credo can work with customers on where copper still makes sense and where optics is needed, including deployments that use both. That gives it more ways to stay involved as the customer’s architecture changes, without needing every connection to move to optics for the business to grow.

The same flexibility applies when the customer needs a more reliable connection or encounters a problem during qualification. Credo has more of the design available to investigate and adjust, including how it works with the customer’s switches, network interface cards and software. That also gives it more options when working through a production issue with manufacturing partners. If the solution requires changing both the electrical and optical design, the work can be coordinated inside the company rather than across separate suppliers and development schedules.

Helping a customer get to market faster is one benefit we expect from that control and one regularly highlighted my management. The customer may also get a product better suited to its power and thermal requirements, or have less work to do resolving problems in qualification and operation. What the May DustPhotonics acquisition gives Credo is more ability to make those tradeoffs across the full connection, and we think that will matter as customer requirements change from one design to the next and vary across many bespoke rack scale designs, as is the industry trend. That is the broader competitive advantage we see in its vertical integration. Credo acquisition announcement

Winning more of the customer’s business

Credo’s current electrical/connectivity relationships give it a way to bring this broader set of capabilities to customers whose systems are already familiar. This is another element that favors an integrated approach in that it creates the environment for more customer co-design and less likely to be designed out. By covering many parts of the complete solution Credo can extend from the cable to optical components and complete transceivers, depending on what the customer needs help with.

There are good alternatives in both cases. Broadcom and Marvell have substantial electrical and optical capabilities, and established module makers have experience getting products through manufacturing and into customer deployments. Some buyers will prefer separate suppliers because that makes parts easier to change. Others may benefit from having more of the design and support with one company, particularly when power, reliability and the deployment schedule have to be worked through together. While we recognize the option of flexibility, our extensive work analyzing the supply chain consistently lands on the observation that customers prefer a simpler supply chain when possible and “one neck to choke,” meaning a single source with more of a total solution is preferred more often than not.

Exhibit 1. Credo can do more of the work inside the connection

Source: Creative Strategies analysis of Credo’s public product portfolio and optical architectures. The map describes product responsibility, not measured competitive performance. Platform participation remains subject to qualification.

Our expectation is that having more ways to meet the customer’s requirements makes it a stronger competitor for the connections that remain open to outside suppliers. Time to market is one reason a customer might choose it; the ability to keep improving the product with that customer is why we think the advantage can last.

How much growth follows?

Our proprietary accelerator and networking models help us better estimate this dynamic. More accelerators, and in particular higher density of accelerators, create more demand for connections, but the number of units is only part of the equation. As systems spread within the rack and across racks, the distance between parts of the system and the bandwidth they need also change. Copper may still make sense for the short links, or dedicated ones once optics intra rack is reliable enough. And its already established optics serves the longer distances. We expect opportunities for both, rather than one date when the industry moves everything to optics. More in our report on optical timing scenarios below.

We envision a scenario where Credo could grow faster than accelerator deployments as the totality of their solutions get higher attach in rack scale builds. Our models let us test how much those assumptions matter, while keeping bandwidth growth from being counted twice. We also have to distinguish selling a component from selling a complete module. The module brings more revenue, but Credo has to pay for more hardware, inventory and support, so a bigger optical business does not automatically mean a higher company margin.

We are less certain about the pace of that ramp than we are about Credo’s position going into it. Qualifications can take longer than expected, manufacturing output can be uneven and customer schedules can move. Over the next year, we want to see repeat orders turn into gross profit and cash. Our view is that Credo is well positioned to win more business, with a path to growth that could justify its valuation. How quickly it gets there, and how much it earns along the way, are the parts we outline in our house view of Credo.

Inside the full report

• What DustPhotonics adds. How joint electrical and optical design could improve power, reliability, qualification and delivery—and where manufacturing dependencies remain.

• Where Credo can win. Why customers might choose Credo over Broadcom, Marvell or established module suppliers, depending on what they’re buying.

• How optical changes affect the business. What Credo could gain or lose as customers adopt linear optics, NPO and CPO.

• The case for outgrowing deployments. Our accelerator and networking models test how content, share and delivery assumptions affect calendar-2028 optical revenue and gross profit.

• What would justify the valuation. Growth and cash-margin scenarios through FY32 show what could support the September 11 valuation—and what falls short.

• What changes our conviction. The orders, design wins, margins and cash conversion that distinguish a lumpy ramp from a weaker thesis.

• Go deeper in CS Atlas. The companion, Credo and AI Interconnects: Technology, Competition and Growth, includes technical detail, competitive comparisons, full optical assumptions and 20 investor questions. Readers with Atlas access can explore those assumptions and ask their own follow-ups.

Read more

Using agentic AI to defend the enterprise

The first leg of AI adoption in the enterprise centered on basic LLMs that employees could ask questions or use to draft content. Those early pilots did not raise anywhere near the security concerns enterprises face today as agents run wild inside their four walls. In less than a year, AI went from those early pilots to software that can interpret a goal, decide what to do next, and keep working around the clock. Enterprises have already recognized the potential, but agent use has moved ahead of governed production. They are now working through how much authority to give systems that can change files, operate software, and access company data. That means setting clear limits, requiring approval for sensitive work, and having a way to recover when something goes wrong. Our first cybersecurity report examined the controls enterprises need as AI gains access to their systems. This report looks at the next phase: using agentic AI to defend those systems. Agents can help investigate threats, find vulnerabilities, and prepare fixes. The next challenge is how much of that response the enterprise can safely let them carry out.

One example from the practitioner interviews helps illustrate the defensive use case: a security agent detects suspicious activity and disables an employee’s account. That could stop an attacker, but a mistaken decision could also interrupt legitimate work. The company needs to know why the agent acted, what it was allowed to do, and how to restore access safely. Those requirements determine how much of the response it can automate.

We measure progress in remediation through verified closure: confirming that a security problem has been resolved and the affected system still works. Disabling an account may contain an attack, but the team still needs to check whether the attacker retains another way in. The opportunity is to help security teams complete more defensive work safely. This report examines how that work divides among frontier AI developers, security platforms, and enterprise teams, and where it could create new revenue.

Exhibit 1. AI Makes Discovery Faster; Enterprise Change Control Becomes the Limiting Step

Frontier labs are becoming part of the defense stack

We believe an under appreciated aspect of frontier AI models is how frontier model labs are becoming suppliers of cyber-defense capability. OpenAI’s Daybreak program gives approved defenders access to models with fewer restrictions on authorized security work, with separate approval for its purpose-trained cybersecurity models. Anthropic’s Project Glasswing gives selected partners access to an unreleased frontier model for defensive work. Both programs make advanced cyber capabilities available through controlled access rather than unrestricted public use.

Cybersecurity creates a distribution problem because the same model can help a defender validate an exploit or help an attacker develop one. Public services therefore block some advanced requests, including legitimate defensive work. A trusted enterprise program can allow more capability because access is tied to a verified user and a defined defensive purpose, with the provider monitoring each session. OpenAI’s Daybreak program is the clearest current example, while Anthropic’s Project Glasswing uses the same basic approach with controlled access to an unreleased model.

Frontier labs are likely to keep investing, making the case for the frontier strong, because gains in coding and reasoning already apply to cyber work. A model has the capabilities to stay with a long technical investigation or work across a large code base may create more defensive value than a public assistant that has to block advanced cyber tasks. That supports a restricted tier for vetted enterprises and governments, often through security-vendor partnerships. We are ready to treat trust-gated cyber access as an emerging product pattern. Private model weights and on-premises deployment remain unproven. A sovereign model is also a reasonable scenario for customers that require tighter isolation around sensitive telemetry, although provider and customer disclosures still have to show which delivery forms become real products.

Continue this question in Atlas: What evidence would move restricted frontier-model monetization from Medium to High conviction?

Model intelligence still needs enterprise authority

A defensive agent needs both model reasoning and the customer’s operating context. The model can help investigate a threat and prepare a response; identity and security platforms enforce what it is allowed to do. The enterprise owns the approval and recovery decisions. Partnerships are the practical near-term path because model developers and control platforms each supply part of the work needed to defend a live system.

The first financial proof we are watching for should appear in higher security-platform consumption or broader paid identity and recovery coverage as customers use AI for more defensive work. The real test is whether better reasoning helps teams resolve more exposure safely. Products that stop at recommendations leave the customer with the work of carrying out the response and the risk if it fails.

Inside the Full Report

• How agentic AI can help investigate threats, contain attacks, and complete fixes.

• The delivery paths and value-capture scenarios beyond today’s controlled-access programs.

• Why the model layer and the security control layer remain complementary.

• A threat-to-control map for the authority and tool paths agentic systems create.

• A safe-remediation framework built around skill-level authority and recovery.

• An auditable completeness scorecard for the leading positions.

• A 2030 market model that separates gross spend from incremental revenue.

• The pricing units and buyer evidence that would confirm or break the thesis.

• Integrated companion research in CS Atlas for subscribers to engage more with the research.

Read more

More suppliers can participate in the same program

The custom ASIC market has become more diverse and competitive as customers mix and match suppliers across their programs. Program wins once gave us a clearer basis for assigning beneficiaries. Customers now have a wider menu of compute, I/O, networking, and front-side and back-side services, and can divide that work among vendors. Supplier revenue and margins depend on the IP, products and design services retained as customers develop more of their own tooling.

Our April reports, Google’s TPU Strategy Offers a Clearer View of the Next AI Bottleneck and Custom ASIC Is No Longer One Market, connected workload specialization in a growing market with different supplier roles and margin profiles. Our September model more clearly separates compute from networking and allows for more sourcing combinations. As Google and Amazon continue to deepen their own tooling, they gain more control over which responsibilities they retain internally and which they assign to suppliers. We expect the wider availability of supplier IP to extend this mix-and-match approach across custom-silicon customers, including those with less in-house IP or tooling.

Our conviction in GPU flexibility remains intact because models are still changing, and inference workloads have very different memory and latency requirements. For enterprises managing that diversity, we continue to see GPUs as the strongest general-purpose choice on total cost of ownership (TCO): the ability to redeploy capacity as demand changes protects utilization. Specialized ASICs can retain an advantage where customers control a sufficiently large, predictable workload.

Broadcom leads in dollars, while share becomes a layer-by-layer question

Broadcom remains the largest modeled dollar beneficiary and the best-proven integrated supplier with a SerDes moat. The revenue and profit it retains depend on the mix of proprietary IP, products and design services customers continue to buy.

MediaTek is our highest-conviction incremental beneficiary. We see its TPU business expanding within a growing market, while its NVIDIA relationship gives it a route to customers developing custom accelerators for NVLink-connected infrastructure. Reaching that broader customer base still requires production volume and MediaTek retaining its implementation role through subsequent generations and widening customer base.

The broader vendor map includes Marvell, Qualcomm, Intel, AMD, GUC and Alchip. Their different combinations of owned IP, implementation experience and production responsibility make a single ranking of technical capabilities inadequate for judging the business opportunity.

How to read the visual: the customer assigns responsibilities across the program. Each box can go to a different supplier; one vendor may also fill several roles. This is a sourcing example, not a disclosed customer design or an allocation of market value.

A larger opportunity still requires program-level diligence

Our 2028 base case puts the value of custom chips and the networking used with them above $350 billion, with a stretch case above $400 billion. Those figures include the estimated value of chip development done inside companies such as Google and Amazon, so they are larger than the revenue outside suppliers would report. We model supplier revenue separately. We also exclude networking used with off-the-shelf GPU systems. The full report explains how different growth and networking assumptions affect these estimates.

Suppliers can earn more gross profit from a growing market even if they receive a smaller share of each customer’s spending. Profit could exceed our estimates if customers buy more of their higher-margin products. Revenue growth from buying components on a customer’s behalf adds less profit because much of that money goes to other suppliers.

As customers choose suppliers for their next designs, we will look at who supplies the compute, interfaces, packaging and networking. Accelerator sales can meet expectations while supplier earnings differ considerably, depending on which parts of the system each company provides.

Inside the Full Report

• The five wallet layers that show where front-side, compute, back-side, I/O, and network value can accrue.

• A vendor capability map covering Broadcom, Marvell, MediaTek, Qualcomm, Intel, AMD, GUC, and Alchip.

• The base, high-attach, and stretch cases, with the assumptions that bridge custom logic and networking.

• Broadcom’s modeled logic allocation and the sensitivity of its total-program share to networking participation.

• The proof milestones and customer sourcing behavior that would change our view.

• What has changed since April, why MediaTek leads our incremental conviction, and how program mix affects valuation.

• Full companion deep research pack in CS Atlas for subscribers to dive deeper with this report.

Read more

Operating View: AI demand continues to grow through 2027, but the industry cannot convert all of that demand into shipments and deployed compute because critical manufacturing capacity is still being installed, qualified, and ramped. The capacity investments initiated in 2026 begin contributing during 2027, but many do not provide a full-year or mature-yield benefit until 2028. This creates the potential for a visible growth caps in 2027 followed by greater unit-growth capacity in 2028–30.

The most recent round of earnings has only reinforced a fundamental point about the AI infrastructure supply chain. The industry only grows as fast as the most constrained layer. We have been tracking as many of the critical supply chain constraints that have been continually moving down deeper and deeper into the supply chain. We have maintained that demand sits around 115-120% above annual capacity. The semiconductor supply chain, nor the power/utility supply chain moves very quickly, and only recently in late 2025 and into 2026 has the wider LTA agreements from top vendors with scale, and balance sheet, provided capacity buildout clarity for key supply chain providers who are now building out.

It is our view by early 2028, much of the capacity funded in 2026 and installed through 2027 should begin to provide some relief. This is a point both for foundry and component capacity and we believe data-center projects should also be further along, giving the industry more room to turn higher component output into working systems and sustain faster unit growth through 2030. Demand should continue to run ahead of supply as the buyer base broadens and each new system generation requires more infrastructure around the accelerator. Our view is that the gap persists in 2028 and starts to bring better supply to demand imbalance (not solve) as more of the stack comes online together potentially leading to a re-acceleration into the end of the decade.

What the latest hardware confirm in this view

Explore the earnings read-through across Dell, HPE, NVIDIA, Broadcom, AMD and Intel in Atlas

Across the latest earnings calls, the same constraints are continually emphasized. Dell and HPE see orders running ahead of the complete systems they can deliver. NVIDIA, Broadcom and AMD show demand spreading from accelerators into the network and rack around them. Intel is adding clean-room and substrate capacity to support the same build. The signal keeps going further down the stack, into MLCCs, analog power components and the electrical equipment needed to bring a data center online.

This is why we see 2027 as the peak year of constraint. More capacity will come online, although the additions will not arrive or qualify on the same schedule. A system can have the accelerator and still wait on a substrate, a capacitor, a power stage or an energized site. The investment underway today should begin to provide broader relief in 2028, when more of these layers have had time to reach volume together. Supply should remain tight, with fewer pieces holding the whole system back.

Take this one level deeper in Atlas: Where are the 2027 bottlenecks, and what begins to ease in 2028?

For Subscribers:

• Why 2027 can be the point when the gap between AI infrastructure demand and deployable supply is most visible.

• How constraints move across leading-edge wafers, HBM, advanced packaging, networking, rack integration, and data-center power.

• Why several capacity additions could begin providing much more relief in 2028 without fully closing the supply gap.

• What the latest results from Broadcom, Intel, TSMC, Micron, SK hynix, Dell, HPE, NVIDIA, and AMD tell us about the timing.

• The operating signals that would strengthen or weaken our view, including qualification progress, backlog conversion, customer readiness, and utilization.

• How subscribers can use the deeper Atlas companion research to test the thesis by company or capacity layer.
  

Read more

The GPU Tsunami framing recap: The race to accelerated computing, driven by AI workloads and the GPU, is pulling forward the entire semiconductor ecosystem as the industry races to meet demand, develop significantly more advanced chips, and solve new engineering challenges which emerge.

For most of the semiconductor industry’s history, test followed a familiar flow around a mostly monolithic chip. The angstrom era is changing that as chiplets, HBM, and other disaggregated designs turn the finished device into a system assembled from several pieces. Every additional die and connection creates another place a defect can appear, while more value is committed before final test. This is forcing manufacturers to screen earlier and add checks after assembly, often under harder power and thermal conditions.

A year ago, all major semiconductor companies told us testing would play a larger role as design complexity increased. We sat on this report because we were unsure how far into the manufacturing flow to take readers. Recent earnings make the change visible across the test market and we think tracking the category shows some early design signals worth playing attention. We are tracking the main test suppliers because each gives us a signal from a different part of the market. Aehr shows where wafer-level burn-in is moving into production. Teradyne and Advantest show how functional-test time and equipment demand are changing across advanced compute and memory. Cohu shows where hotter, more complex packages require new handling, thermal control, and inspection. Together, they reveal where rising design complexity is turning into additional factory work and test capacity.

The GPU Tsunami is reaching semiconductor test

Semiconductor manufacturers use test machines, commonly called testers, to apply power and signals to a chip and confirm that it works. Each machine has a fixed number of hours available. Larger accelerators can remain on the machine longer because they draw more power and require deeper test patterns, reducing the number of chips each tester can process. Advanced packaging adds cost around every die, while optical I/O introduces checks that can only happen after more of the device has been assembled.

We therefore focus, and have framed an economic model, on test content per sellable AI system: the total factory work required to prove that the components and the completed system will operate as intended. Chip volume remains important, although it is only one source of demand. Test capacity also grows when a device occupies the machine longer, requires checks at additional manufacturing stages, or consumes enough power that fewer devices can be tested at once.

Required test capacity = unit volume × test insertions × test seconds per insertion ÷ effective parallelism

Each part of the model has a direct meaning. More devices, additional checks, and longer tests increase the equipment requirement. Testing more devices at the same time reduces it.

In the full report, we use this model to estimate where the added test value can accrue. Functional-test suppliers benefit when processors require more machine time. Handling and optical-test companies participate when advanced packages create new checks. Wafer burn-in depends on how many wafers receive sustained stress and how long the process takes. We then map that capacity to equipment content and supplier position to estimate the commercial opportunity.

Wafer burn-in finds weak die before they become expensive failures

An individual chip, called a die, can pass a short electrical test and still carry a weakness that appears only after sustained heat or power. An AI accelerator die, as an example, can pass a short electrical test while still carrying a weakness that appears only after sustained power and heat. If the failure is discovered after the processor has been packaged with HBM, an interposer, and other components, the manufacturer can lose the value of the entire assembly. Wafer burn-in applies that stress before the expensive package is built.

Finding the weakness at this stage lets the manufacturer remove the die before packaging. A failure discovered later can destroy the package, the good die inside it, and the work already used to assemble everything. Earlier reliability screening protects more value as the package becomes more expensive.

To show how the risk compounds, assume a package requires eight to twelve die and that each selected die has a 98% to 99.5% probability of being good after screening. If failures are independent, the probability that every required die is good ranges from approximately 78% to 96% before assembly losses. These are illustrative assumptions, not measured industry yields. The point is the multiplication: as packages combine more required die, even a small improvement in die-level confidence can protect substantially more downstream value.

Test orders show where advanced silicon is entering production

Recent earnings put numbers around this functional change for testing. Our blended measure of broad automated-test revenue grew roughly 35% to 45% in the latest full-year periods. More recent disclosures show compute-related test revenue and orders rising roughly 100% to 150% year over year. A separate test-cell utilization series moved from the mid-70% range to about 80%.

Wafer burn-in adds a specialized production indicator. Effective backlog at one supplier is roughly twice its recent annual revenue, supporting a forward revenue range of about 2.5 to 3 times that base. These figures measure different jobs and periods, so we treat them as independent markers of customers committing more factory time and equipment to advanced silicon.

Test equipment can reveal where a new chip architecture sits in the manufacturing cycle. An engineering system shows that a chipmaker has reached a new test problem. Qualification shows that the equipment can find a failure or performance limit the customer cares about. A production installation means the test has cleared the cost and throughput hurdle required to become part of manufacturing. Repeat orders show that the program is scaling or that the same test is spreading to more devices.

We highlight Aehr here because our conviction is it gives us a tangible view of that progression because its system configuration and order pattern reveal how customers are using wafer-level burn-in. Silicon carbide established the production case, while recent repeat capacity for AI processors and silicon photonics shows the process moving into additional high-value manufacturing flows. We apply the same framework across the broader test market to interpret functional-test, handling, optical-test, and reliability-equipment demand.

Inside the full report

• An original capacity model showing how more test steps, longer test times, and parallelism translate into equipment demand across the test market.

• A manufacturing-flow map showing where test enters accelerators, HBM, advanced packaging, and optical I/O.

• A wafer-burn-in sensitivity model that converts screening coverage and stress time into required production capacity.

• An application map separating current production evidence from the next proof point across SiC, silicon photonics, GaN, silicon power, advanced packaging, and memory.

• A value-capture map showing which suppliers participate in each test job, how the spending reaches them, and what evidence to track.

• A monitoring ladder that follows new test processes from engineering and qualification through factory transfer, repeat capacity, and adoption across independent programs.
  
  Yesterday we introduced CS Atlas, a new research platform from our firm Creative Strategies, now available to all current subscribers. Use it alongside this report to go deeper into the work behind our view—from the signals across Aehr, Teradyne, Advantest and Cohu to the links between semiconductor test, advanced packaging, HBM, optical I/O and power semiconductors. Atlas can help you compare suppliers, work through our capacity model and identify when a new test process is moving from engineering into production.

  Some suggestion prompts for CS Atlas on this subject:

  • “Compare the demand signals provided by Aehr, Teradyne, Advantest, Cohu and FormFactor.”

  • “How do chiplets, HBM and advanced packaging increase test content per sellable AI system?”

  • “What milestones show that a test process has moved from engineering to qualification and production?”

  • “Where does wafer-level burn-in have the strongest technical and economic fit across SiC, GaN, silicon photonics, advanced packaging and memory?”

  • “How does optical I/O change the number and type of tests required during manufacturing?”

  • “Which assumptions have the greatest effect on required test capacity: units, test insertions, test time, parallelism or screening coverage?”

Read more

Research built to inform the decisions in front of you

Today we are introducing the next step in that work. We are opening CS Atlas to current subscribers as a soft launch included with every active subscription.

Atlas is a research environment built around the full body of work Creative Strategies produces, including research that extends deeper and wider than what we publish on Substack. Subscribers can ask Atlas a question and receive an answer grounded only in our research and analysis. Each response shows the Creative Strategies work behind the answer and provides direct access to the supporting reports, data, or models included with the subscription.

Atlas is designed for the work that follows reading. A subscriber can pressure-test a market assumption or examine how our view of a company has changed through several earnings cycles. The same research path can follow a technical change through the companies and markets it affects. This shortens the distance between a current question and an evidence-based judgment while preserving the research trail.

Why Creative Strategies

A research agent inherits the quality of the research behind it. That is where CS Atlas begins.

Creative Strategies has worked inside the technology industry since 1969. That history gives us a long record of how platform shifts develop and which patterns tend to repeat. It provides the foundation for how we map markets and evaluate the companies positioned within them. Our analysts combine that historical perspective and deep industry experience with direct exposure to the decisions shaping the industry today. Having worked through every major technology cycle since our founding, Creative Strategies is now relied on by many of the world’s leading technology companies for independent research and market insight.

Our analysts are in the rooms where companies explain product roadmaps and defend the assumptions behind them. We attend launches and analyst briefings, then spend time with management in smaller sessions. The work continues after the event as we track those claims through subsequent earnings cycles and test them against reported results and our broader technical and competitive analysis. Over time, this creates a record of how priorities shift and whether management’s confidence is supported by results. It also shows where a vendor has sustained an advantage. Atlas makes the publishable portion of that work searchable and source-linked, with the latest research and commentary carried into every answer.

That company-level work becomes more valuable when placed inside our broader view of the industry. Creative Strategies covers the technology stack from silicon through adoption. We can follow a change in semiconductor architecture through the systems that use it, into the software economics it reshapes, and out to how customers adopt it. Atlas connects those layers so subscribers can test what management says against the underlying technology and determine whether the economics support adoption. The coverage map below shows the research areas that feed that analysis.

How subscribers can use the research in CS Atlas

• Earnings notes: Our earnings work focuses on what each quarter tells us about the core companies we cover. We compare reported results with our models, explain where our conviction changed or held, and identify the evidence we will watch through the next earnings cycle. We provide research rather than stock ratings or stock picks.

• Timely news analysis: When an announcement changes a company’s strategy or market opportunity, we place it inside our longer-running research and explain which assumptions should change.

• Key event takeaways: Our analysts attend product launches and briefings with company leadership. The coverage connects each announcement to the company’s technology roadmap and competitive position, then identifies the adoption evidence that should follow.

• Commentary from management meetings: When company leadership adds context we can publish, we explain how it changes our reading of the company and add that evidence to the Atlas research record.

• The Creative Strategies research archive: Atlas connects current questions to our accumulated company and market work. Subscribers can trace how our analysis developed over time and, where maintained models are included with their access, inspect the assumptions behind the answer.

Every Atlas answer is grounded exclusively in Creative Strategies research and shows the supporting sources. Subscribers can see how the conclusion was formed and open the underlying work to continue the analysis from the evidence.

Atlas becomes more valuable as our research record grows. New reports add current evidence, while the accumulated archive makes it possible to compare today’s claims with what companies said and did in prior cycles. We will continue expanding those connections and the ways subscribers can work with them.

How to start

All subscribers with a currently active subscription can log in to CS Atlas here using the email tied to their subscription. Start with a current question and follow the source links into the supporting research. The soft launch also gives subscribers access to our expanded fall event coverage as our analysts move through the industry calendar.

All current subscribers will remain grandfathered into the Landmark research tier at their existing price for as long as their subscription stays active. After the soft launch, Landmark will be priced at $500 per month or $5,000 per year. Technology vendors and institutional clients can also access Meridian, which includes expanded data and model dashboards, or Horizon, which includes our currently maintained core models and new models as they are released. Contact us to discuss those tiers or multi-seat institutional access.

We designed Atlas to make Creative Strategies research easier to use when a decision depends on understanding what changed and how the effects move across the technology stack. The supporting evidence remains attached to the judgment so subscribers can inspect the reasoning and continue the work.

Access CS Atlas

We appreciate your interest in our research.

Subscribe now

Why This Looks Like an Era of Ferment

In their seminal 1990 paper, Technological Discontinuities and Dominant Designs: A Cyclical Model of Technological Change, management scholars Philip Anderson and Michael Tushman formalized the idea of an “era of ferment.” They used the term for the period after a major technical break, when an industry tests competing approaches before one becomes the common architecture. Their research found that the design that wins often, though not necessarily, trails the technical frontier because adoption also depends on cost, manufacturing scale, software support and ease of use. (Anderson and Tushman, 1990)

Hot Chips gave us the clearest evidence yet for applying that framework to inference. NVIDIA has already set the merchant standard for training. Inference is producing a much wider range of technical bets, allowing for some competitive pressure, because vendors disagree about where data should live and which processor should handle each part of a request. Those choices change how much work software must carry. More specialized hardware can improve speed or cost, although the gain has to justify the added work. We highlight a few examples below before examining each bet in the full report.

Memory Is Becoming Part of Compute Design

Memory gave us some of the clearest examples of these different views. Micron and SK hynix are extending standard HBM through taller stacks and better bonding. Samsung is turning the HBM base die into a control point, a move we believe will become table stakes for next-generation HBM, and eventually wants to place DRAM over logic with zHBM. d-Matrix bonds custom DRAM under its accelerator. OXMIQ’s HBF uses flash as a cheaper home for model state that stays cold, while XCENA and Samsung use CXL memory to hold older KV pages and reduce them before sending a result back to the GPU. HBM supports a broad range of workloads at higher package cost. HBF and CXL ask software to identify state that can sit farther away, while d-Matrix accepts a smaller local memory pool to keep the useful data close.

Ethernet Still Provides the Common Network Base

The networking talks showed more agreement at one layer. Ethernet is becoming the common protocol for scale-out networks. Copper remains practical over the shortest links, while optics—and the variety of approaches showing up— takes over as distance and speed increase. The scale-up fabric connecting accelerators into one tightly coupled system remains less settled. Broadcom provides merchant NICs and gives customers more choice over cables and optics. NVIDIA controls the DPU, switch and software, and is bringing optics into the switch. Microsoft changes how work moves over Ethernet through software. Google uses optical circuit switching to shape the network around the workload. These systems share a protocol while putting the physical links and traffic control in different places.

Accelerators Carry the Widest Disagreement

Accelerators showed the widest range of views. We saw it in the public presentations and in meetings with six private AI accelerator companies, where each team had a firm view of why its architecture maps better to inference. The public presentations gave us several examples of how those views lead to very different hardware.

Google has landed on an ASIC roadmap that separates TPU designs for training and inference, while Meta is building one MTIA platform for recommendation and generative AI. Microsoft uses software to place data explicitly across Maia. OpenAI keeps KV state local and changes the active compute mix inside Jalapeño as a request moves through its phases. NVIDIA is bringing GPU throughput and Groq’s low-latency LPU into one platform. SambaNova maps the model into a persistent dataflow system, while Cerebras makes the wafer the compute unit. Each architecture also carries a forecast about how models will behave years from now. A design built around sparse access, local KV state or deterministic execution keeps its advantage only while that workload assumption holds.

NVIDIA Is Closest to the Merchant Standard

Our conviction is that the dominant design will form at the system level because useful inference depends on software coordinating compute and memory to produce the required output at an acceptable speed and cost. As specialized processors are added, that coordination becomes part of the architecture itself. The system has to make different processors and memory designs work as one pool of capacity, which pushes vendors toward deeper co-design around the workload.

NVIDIA is the clearest current example of that thesis. It has the most complete in-house control of the merchant AI system, from accelerators and CPUs through networking and fleet software. CUDA ties those parts together. Bringing Groq’s deterministic LPU alongside the GPU shows how NVIDIA can add a specialized inference engine without asking customers to operate a second system.

AMD’s Helios follows the same direction through an open reference design that partners turn into products. Google and Amazon already integrate custom chips with their cloud infrastructure around workloads they control. Hyperscalers have enough internal demand to support those systems, while NVIDIA is building for customers that need to buy the full stack. That gives NVIDIA the clearest path to becoming the broadly used standard outside the largest clouds.

How We Will Know the Market Is Settling

This period could last longer than earlier hardware shifts because AI models can change several times during one chip-development cycle, while inference includes workloads with very different speed and memory needs. A design that fits today’s models may lose its advantage before the next chip reaches volume. Specialized hardware only improves the economics when its performance gain exceeds the added software work and operating cost. If we use the era of ferment framework we would expect this part of the cycle to end when competition among rival designs produces a dominant design, after which the industry shifts toward incremental improvement around that common architecture. We detail in the full report why some nuance and variations will still exist but it will not be nearly as diverse as we see in the market today.

What intrigues us about this moment is what that long testing period could produce. And a follow on question to who ultimately benefits the most in longevity of uncertainty in models and inference architecture standards. Several 800-pound gorillas already control much of the market, yet the range of architectures gives the industry a way to test very different answers against real workloads. Some will fall short once their software and operating costs are included. Others could create a larger opening for a smaller company or become part of an incumbent’s broader roadmap. In the rest of this report, we examine the technical bets from vendors large and small that stood out at Hot Chips 2026, along with the workload assumptions that have to hold for each one to work.

Paid subscribers get the full technical breakdown

• How HBM scaling, Samsung zHBM, d-Matrix 3D DRAM, HBF and CXL processing near memory each relocate the memory bottleneck, including the workload behavior required for their economics to hold

• The architecture choices that remain open above Ethernet, including the handoff from copper to optics and the division of fabric control between hardware and software

• The model and software forecasts embedded in Google TPU8, Meta MTIA, Microsoft Maia, OpenAI Jalapeño, NVIDIA/Groq, SambaNova and Cerebras, along with the assumptions that would strengthen or weaken each technical bet

• Why NVIDIA is currently closest to the merchant dominant design and how hyperscalers or specialized accelerators could still sustain alternative systems

• Three original Creative Strategies exhibits mapping each architecture’s technical bet, required workload behavior, main constraint and the proof points needed to separate benchmark gains from durable inference economics

Read more

For those of us who have seen many different technology adoption cycles, the patterns that make the AI buildout still look like an immature market are familiar. Some parts of the market begin to show consistency, while the technology, standards, processes, or protocols remain far from settled. This is particularly true in early enterprise adoption. Even among the businesses moving most aggressively, more questions than answers remain. This report updates our E/AI Index CIO/CTO survey through our most recent channel checks across the areas that report highlighted as open questions, along with our most up-to-date observations from recent enterprise AI fieldwork conversations with decision makers on AI deployment.

The May E/AI Index showed where budgets were forming. This update follows them into production

That customer work formed the basis of our May E/AI Index CIO/CTO report. Our survey across a cohort of CIO/CTOs showed that AI had become a staple of enterprise IT budgeting. The survey revealed the clearest ROI was showing up in what we articulated as bounded workflows. These are repeated tasks with a clear result and an existing baseline, which lets the company compare cost and quality before and after AI. Agent deployment was also running well ahead of broad production and still is today. Budget formation is also a key metric we are tracking, as we want to continue to see AI become budget additive vs. take away from other areas scoped dollars.

The main change since the May E/AI Index is the move into what we would call governed production. This is the point where an AI workflow leaves the relative safety of a pilot and starts operating inside the business under defined rules. Companies are now learning whether the value they saw in a controlled test holds when the workflow reaches more employees and touches more company systems. That wider use is exposing the upgrades required across the technology stack and internal processes before AI can move into more of the enterprise.

Governed production also gives management a much clearer view of the cost and risk attached to each workflow. Usage begins to shape which model is worth paying for, while company data has to reach the agent in a form it can use without losing the access rules around that information. The authority given to the agent then determines how much of the workflow it can complete on its own. Our fieldwork suggests this is where most enterprises sit today: the workflow is live, while companies are still building the cost controls and governance required to expand it safely.

Governed production is happening workflow by workflow

The same company can be at very different stages of adoption depending on the work being done. Coding may already be running at scale because the work is measurable and review is built into the process. A finance agent may remain under tight human control because an error can reach a system of record or affect an important business decision. This is why we look at enterprise adoption at the workflow level. Both the risk and the economics change with the work, and those conditions determine how much authority the company is willing to give the agent.

Exhibit 1. Enterprise AI has reached governed production, while scaled operating change remains early. Note: Qualitative Creative Strategies adoption-cycle synthesis. Stages describe operating maturity; the exhibit does not estimate a pooled survey distribution. Source: Creative Strategies analysis of enterprise CIO conversations and field checks through August 2026.

Our checks show that the companies furthest along are beginning to put a clear operating structure around each production workflow. A named owner sets the permission rules and determines how problems are escalated, which is what moves the workflow into governed production. The next proof point is whether the economics continue to hold as more employees use the workflow and the system completes a larger share of the work. Few companies have reached the point where AI-created capacity changes how jobs are designed or which software and outside services they continue to pay for. Those decisions would be stronger evidence that AI is beginning to change how the company operates and spends.

Companies are now managing the cost of production use

Our second-half checks also show that production is forcing CIOs to measure what it costs to complete a workflow. IT decision makers are counting model calls and asking where a premium frontier model creates enough economic benefit to justify the token cost. Our favorite quote is, “You don’t need to pay frontier model costs to summarize a Teams meeting.” Other areas may have clear value, such as coding and customer service, which have moved faster because the work already has a measurable baseline. Workflows without one will face more scrutiny, increasing demand for systems that route each request to the right model.

The economics also depend on whether the agent can reach the right company information while work is happening. Most enterprise data was organized for storage or later reporting. A production agent needs current context inside the workflow and has to preserve the access rules attached to it. Our checks keep coming back to this as one of the clearest ways production is exposing how much work remains across enterprise data systems. We detail that coming storage challenge in the report below.

Data sovereignty is also shaping model choice. Most workloads still run through closed frontier models in the cloud because they produce the best results on difficult work. Open-weight models are entering through cost-sensitive workflows, although interest still runs ahead of production use. Enterprises want to keep company context and workflow records under their control so they can change models without rebuilding the process. A credible open option gives them more choice when closed-model costs rise. We think that cloud-led pattern also creates a premium inference tier (good for Cerebras and NVIDIA Groq), particularly in workflows where lower latency increases completed work enough to justify paying for faster tokens on top of frontier-model pricing.

CIOs and CISOs are tightening control over who can create an agent and how much authority it receives. Companies still rely on human review across most production workflows because the controls around agent actions remain incomplete. Wider deployment now depends on whether companies can hold quality while improving the cost per completed task. Human review also has to grow more slowly than the amount of work the system completes.

Inside the Full Update

• Why governed production now best describes enterprise AI adoption and what changed once approved budgets reached live workflows

• Where companies sit in the five-stage adoption cycle and what still separates production from scale

• How CIOs are measuring the full cost of completing a workflow instead of relying on seats or token growth

• Why control of company context gives enterprises more flexibility across closed and open-weight models

• How permission rules and auditability determine the authority companies are willing to give an agent

• Why cloud-led deployment can create a premium inference tier even as some workloads move back on premises

Read more

Exhibit 1 groups the models into five classes based on the work they can complete reliably inside an agent system.

The Diligence Stack has covered the AI market broadly. This is our first full assessment of the models themselves. Something we will keep updating at different points in time. From talking with large enterprises, we believe we are going through the same exercise as it relates to both models that are acceptable for a wide range of knowledge work use cases and costs associated and tokens used. We share our learnings from this exercise and the key takeaways relevant to compute infra and model economics.

We built CS Bench to compare models inside our knowledge base agent that is the basis of our research for the Diligence Stack. The published benchmark measures the quality and estimated cost of first-draft of a financial analysis which is the anchor use case we test. We also use the models every day for research, software work, and the production of finished files. This report uses all tasks tested as its base for analyzing each model.

From a model evaluation standpoint, buyers pay model providers by the token and judge the result by the completed work. A low token price does not help if the output has to be checked for an hour or rebuilt. We include that review time when we compare model cost.

Financial modeling shows the problems more often than other workflows. The most common failure is a workbook that looks finished before its logic has been checked. It may have several tabs and a working scenario switch. The formatting looks credible on a first review. A deeper check then finds revenue pulled from the wrong fiscal period or driver rows hard-coded as values made to look like formulas. Finding those errors can take an hour or more.

We therefore score the work and the presentation separately. Correct work with poor formatting creates cleanup. Work with obvious errors is usually rejected quickly. The costly case is a polished file with errors buried inside it because the presentation makes the work look ready to use.

Exhibit 2 shows the difference. CS Bench measures the first draft before an expert takes over. We are now adding the time needed to review that work, the errors that repeat, and the time needed to find them. Those checks give us a better estimate of the cost of usable work.

Parameter count says little about whether a model will finish the job. We focus on whether it understands an incomplete task, uses tools until the work is done, and recovers when its first approach fails. The software around the model also affects the result. Retrieval, file handling, and execution can make the same model much more or less reliable. Buyers are paying for that full system.

Public API rates describe only one way buyers pay. Small teams often use fixed-price subscriptions. Enterprises negotiate lower rates in exchange for spending commitments that they may not fully use. The real cost includes the model, retries, and human review. A low token price can still lead to a high cost for completed work.

For investors, the amount of review is one way to see which models still earn a premium as token prices fall. Smaller models can handle large volumes of clearly defined work. Larger models remain better suited to work that requires more judgment. Software companies can also earn a share of the value when their products make the models more reliable. The market will support several classes of models because the work and the cost of failure vary widely.

Inside the Full Report

• How we group the models and our assessment of each class.

• Our scorecard for work quality, presentation, tool use, instruction retention, judgment, and cost.

• Workload-level recommendations for research, software engineering, financial modeling, and everyday assistant use.

• How subscriptions, enterprise contracts, usage, and review change the cost of completed work.

• Why switching models during a long conversation can raise the total cost.

• The evidence that would change our rankings and what CS Bench will test next.

Read more

In March we published our full primer on the transition from copper to fiber. We were clear then, that this transition takes time and NPO would be a primary driver before CPO with LPO/pluggables still the most mature segment and not going away any time soon. Today’s report goes deeper on the manufacturing challenges that will continue to more accurately inform the timeline of adoption of higher optical content per rack specific to NPO and CPO.

We have spent the last few weeks catching up with friendlies across the optical supply chain. The move toward NPO, CPO, and optical I/O still feels like the Wild West. Architectures remain unsettled, and each vendor still requires highly specialized co-design work. That lack of standardization is the clearest sign of an immature manufacturing ecosystem and makes timing and likely winners much harder to predict.

One constant thread has persisted through every conversation. Manufacturing will set the pace of optical attach through at least 2028, and potentially longer. Network architecture and compute roadmaps already point to more optics per unit of compute. The industry will try to adopt these architectures faster than suppliers can produce qualified systems, so qualified output anchors our timing and exposes forecasts that convert demand directly into shipments.

Absent a major manufacturing breakthrough, the constraint should move through the supply chain in a clear sequence. New InP capacity raises the upstream ceiling first. As good laser die increases, precision packaging becomes the next manufacturing challenge. Most AI racks already contain optical links, so optical presence tells us little about increased adoption. We track how much connectivity per deployed rack moves beyond pluggable modules into NPO or CPO. Pluggables remain the more mature, lower-risk path and may persist longer than many want to admit. NPO and CPO can lower power per bit and raise bandwidth density, though their system-cost advantage, and reliability, still has to be proven at scale.

Our model first estimates the incremental NPO and CPO links supported by 2028 network demand. Manufacturing constraints allow about 60% of those links to ship in our base case. This means, for every 100 incremental NPO/CPO links the architecture could support, roughly 60 are realized in 2028. The balance remains on pluggables or moves into later years. We see the larger conversion step in 2029, after new laser capacity has had time to qualify and package output becomes repeatable.

Capacity arrives before qualified output

Our work across public disclosures and supply-chain conversations kept turning up the same gap between installed capacity and usable supply. Capacity announcements usually measure wafer starts or installed equipment. Customers, however, buy qualified optical products. Performance depends on maintaining a low-loss light path across materials that expand differently as temperature changes and the package ages. The substrate has to meet the laser maker’s specification before the device fab can produce good die at repeatable yield. Those die still have to survive packaging and system qualification. Each stage runs on its own timeline, so finished output can remain constrained as physical capacity rises.

The move to 6-inch InP wafers is one attempt to raise the upstream supply ceiling. Larger wafers can produce more die per run once yields settle. Crystal uniformity, process transfer, and customer qualification take time, so we treat 6-inch capacity as the start of the supply ramp. Finished supply comes later.

Coherent has started 6-inch production. Lumentum’s Greensboro facility is scheduled to begin ramping in mid-2028, leaving limited time for qualified shipments that year. We treat 2029 as the first year these additions can support broader attach and will update that assumption as shipment evidence improves.

Packaging becomes the next challenge

More substrate and good laser die can move the manufacturing challenge downstream into packaging without producing the same increase in finished units. Lasers must be aligned to a lens or waveguide with very little loss. Some optical assembly flows still rely on manual or operator-assisted steps, particularly where package designs remain custom. Higher-volume lines can automate active alignment, though automation does not remove the cycle time. Precision equipment still has to measure optical power, move the component into position, and fix it in place for every package. More good die can therefore expose a packaging bottleneck instead of translating directly into finished units.

Manufacturers test again after assembly because bonding and thermal stress can shift the optical path. If the completed package fails, the working components inside it may become unsellable or require costly rework. NPO and CPO concentrate more component value in each assembly, making finished-package yield the better measure of usable optical supply.

Optical attach also moves at different speeds across the network hierarchy. Scale-across already requires optics, while scale-out uses pluggables as the volume base today. The larger incremental content opportunity sits in scale-up as optics moves closer to the accelerator. Our base case moves inter-rack links first because bandwidth is concentrated into fewer optical connections at the rack boundary. Broad intra-rack adoption comes later because it spreads optics across many more endpoints, multiplying the package output that must reach repeatable yield and making failures harder to service inside the rack.

Public disclosures point to early inter-rack CPO activity in 2027 and new laser capacity beginning to ramp in mid-2028. We use those milestones as timing anchors. Our model concentrates incremental NPO/CPO attach in lead inter-rack scale-up systems through 2028. Broader intra-rack attach comes later, after package output becomes repeatable and the manufacturing ecosystem matures.

Inside the full report

• A low, base, and high scenario for how quickly NPO and CPO can grow as a share of modeled AI compute-fabric links.

• A year-by-year map of the primary manufacturing constraint limiting qualified output.

• A pluggable-persistence sensitivity showing how the 2030 mix changes even if manufacturing improves.

• Why a laser fab starting in mid-2028 has limited impact until 2029.

• A public market map showing who controls each step of the optical transition, including the operating proof required at each layer.

• The manufacturing and rack-level proof that would cause us to raise or lower the attach curve.

• Company read-throughs for AXT, Broadcom, Coherent, Lumentum, and AAOI, tied to the manufacturing step each controls.

Read more

One of our favorite things to do at vendor events or broader trade shows is get up close and personal with compute trays and inspect the semiconductor content and adjacencies that show up across different boards. We know this labels us as geeks. We prefer technologists at heart, and we are ok with either label. There is much to learn in this process. Tracking whose names show up, how much content is present, and which functions are being added can signal trends we are watching for. This exercise led us to our report on power semis, and now to a vastly misunderstood and perhaps under appreciated bit of silicon called FPGAs.

We have followed Lattice Semiconductor for some time, spoken with management regularly, and attended numerous customer and developer events. The company had been outlining a datacenter opportunity, and the story was plausible yet unproven. Then we started seeing Lattice on AI accelerator boards, networking boards, and AI CPU boards. You get the point. Lattice has talked about FPGA ratios growing to multiples on compute boards, and observationally we can see it happening. The interesting part is why, and what increasing programmable content tells us about AI infrastructure design.

The rack has become one brain

We think this is well known and obvious, but modern day AI compute requires rack-scale architecture. The full AI compute rack system connects through a common fabric and functions as one brain.

As rack density increases, control work rises around the main processors and lets them focus on their core job. More variables have to be managed, monitored, or secured, which increases semiconductor attach per rack. The GPU tsunami is also pulling more FPGAs into the AI rack system.

All compute boards have a BMC (baseboard management controller) which is a dedicated processor that monitors and manages a server independently of the main CPU and operating system. While crucial, its I/O is fixed. A more complex board creates signals and interfaces the BMC may not cover cleanly. A small FPGA can sit beside it as a companion and handle a wider range of work. The BMC relationship with an FPGA is why we so often see an ASPEED processor (BMC) in close proximity to a Lattice FPGA.

We have spent time in years past trying to explain why FPGAs are so valuable. Engineers understand the value proposition. The broader market has a harder time seeing it because an FPGA looks like a blank slate to a degree. We think it is helpful to frame FPGAs as silicon that enables flexible specialization. Its programmable fabric becomes specific in its job once the customer designs it into a board. That flexibility has real value as AI infrastructure becomes more customized.

Understanding the rack as a compute system requires viewing it as planes in which silicon has specific roles. Accelerators and CPUs form the compute plane. Networking provides the fabric that lets the rack operate as one system. Power and cooling keep it within operating limits. The control plane sits across all of it, sequencing power, configuring boards and interfaces, monitoring system state, enforcing security policy, and isolating failures.

As AI infrastructure becomes more purpose-built, that control work becomes specific to each system. Programmable fabric makes those design variations practical while the architecture is still changing. The deeper this co-design goes, the stronger the need for flexible specialization.

Source: Creative Strategies analysis of public rack documentation, OCP specifications, company technical materials, and primary research.

Counting packages misses the point

That creates a value-capture problem. FPGA units and ASPs measure how much standalone silicon was sold, while our thesis follows programmable control content across the rack. Today, both are rising as more boards and control work increase FPGA attach. Some functions could eventually move inside combined devices, which would leave a chip count missing programmable content and the firmware that makes it reliable. We have not seen evidence that integration is reducing attach, so we treat it as a downside risk rather than the base case.

We created Programmable Infrastructure Content, or PIC, to measure programmable control wherever it sits. The model counts each function once and includes firmware only where a supplier can charge for it. PIC per rack, megawatt, or gigawatt lets us compare deployments of different sizes without losing the system-level view.

Our view today is that AI infrastructure is creating more places for programmable logic across the board and rack. Liquid cooling, 800 VDC, and more optical content should add new control work. Faster product cycles and greater customization favor a device that can solve a specific control problem without forcing a larger system redesign. The board-level evidence is still developing, which is why the full report builds the case from the architecture up and tests how far FPGA content can rise.

Inside the Full Report

• The PIC model from individual control domains through rack, megawatt, and gigawatt values

• A disaggregation sensitivity that tests new board boundaries against BMC, SMC, and eFPGA consolidation

• Lattice’s verified server attach history and the mechanism behind the increase

• A use-case map showing how flexible logic becomes specialized control on each board

• A TCO bridge from faster design cycles and lower board-support burden to rack manageability

• A four-layer software map separating design tools from board firmware and rack orchestration

• A role-based market map covering established FPGA leaders, specialist challengers, Chinese suppliers, and fixed-function substitutes

• A downside and monitoring framework with thresholds that can confirm or break the thesis

Read more

We are introducing a new feature called The Changelog. For the time being, we will update it quarterly because this industry changes enough over three months to justify a formal review. It will cover our core theses and the companies we track. It will also cover the micro and macro trends shaping our work. We will explain what changed and why, then identify new theses or areas where our conviction increased.

This regular review will help keep us honest and make industry changes easier to follow. Our research covers the full technology stack, so The Changelog will also give readers a consistent set of mental models they can use across our work.

What this first review found

This first edition covers six months of published research across our broad technology industry coverage. Our conviction in the broader AI thesis remains high and increased in several areas as we studied enterprise customers, their early deployments, and what they learned. We also maintain that demand for compute, and the supply chain around it, will exceed supply for the foreseeable future. At the same time, new evidence changed several views under the broader trend and produced eight new theses.

The broad AI demand thesis held across the six months. The evidence changed several views underneath it. Inference raised our conviction in agentic CPU demand, bringing along revived winners as a part of it, and the tension building across memory and networking. Our enterprise work increased our confidence that current adopters are seeing enough ROI to keep spending. Security also emerged as a new thesis as companies focused more on protecting their data and controlling what agents can access.

The thesis map below gives a high-level view of how each thesis moved during the period. Raised conviction means the evidence strengthened an existing thesis, while hold means our view remains intact at roughly the same level of conviction. Changed means the evidence altered the thesis or how we measure it. New means the thesis first emerged during this period, so we start with lower confidence and test it again in future editions.

Inside the full report

• The full raised conviction, hold, changed, and new thesis ledger.

• Why inference turned the merchant-versus-custom accelerator market into a TCO competition and raised our CPU forecast.

• Eight new theses, with the first report, current confidence, and evidence still needed.

• How chiplets deepened the advanced-packaging constraint and made Intel’s opportunity clearer.

• What enterprise ROI, token budgets, and security requirements now tell us about adoption.

• Why bitcoin miners may become better capacity partners for hyperscalers than pure AI neoclouds.

• A next-quarter monitoring table showing what would strengthen or weaken each view.

Read more

We spent the last few days at FMS (Future of Memory and Storage) and had meetings with all the key players in the memory, storage, and now also interconnect/networking, ecosystem. A clear takeaway was better line of sight to the CXL standard to start to solidify. If you are not familiar with CXL it is a standard interface that gives the industry a common way to attach and share memory over a PCIe-based physical link. Up to this point the deployment model around that interface is still being developed, which is why the ecosystem can support several architectures, from a memory box inside a rack to pooled or optical memory systems.

From our conversations we believe deployments are more likely to start next year and build into 2028, but the ecosystem is maturing enough for CXL to move from a standard into something that can be deployed with enough customers to meaningfully start to deploy in their AI compute infrastructure. The first and most logical customer base is the hyperscalers, which can put memory and inference workloads into custom compute clusters and use the surrounding ecosystem to qualify the architecture.

It is our conviction that solving the memory wall will take many different shapes by many different players but the same problem statement remains. We need more memory, and designers are up against how much memory can go on CPU/XPU/GPU package or near the package. Having ways to expand the available memory pool while keeping latency low enough for selected near-memory workloads is the promise of CXL.

Memory is becoming a fleet problem

The old server model ties memory capacity to one processor and its local channels. AI workloads make that boundary more expensive because context, KV cache, and orchestration state can grow faster than the memory attached to one compute device. A fleet can have plenty of memory in total while individual CPUs or accelerators still run short.

The practical distinction is important: “hot” describes the role memory plays in the workload, while local, off-die, off-board, and rack-level describe where that memory sits. HBM can remain the hot tier in an attached appliance when the fabric preserves the bandwidth and latency the workload requires. DDR can be split by role in the same way. Local DDR can serve the CPU’s most active data, while DDR4 or DDR5 behind a CXL controller can provide a larger attached tier and eventually a shared pool. That remote memory will not have the same latency as on-package HBM, but it can still be the highest-performance tier available outside the package.

CXL gives the system those placement choices through a coherent interface. The first commercial use is likely to be a card, module, or box that adds memory to one host. Later designs can attach several hosts to a pool, with software deciding which data belongs close to compute and which data can move into the attached memory. The value comes from expanding the working set and using capacity more fully without buying a full server for every increment of memory.

The graphic below frames the opportunity by workload rather than by device. HBM remains closest to compute for the hottest accesses, while local DDR/SOCAMM and CXL-attached memory support larger working sets, overflow KV cache, retrieval buffers, and shared state. CXL is not a fixed warm tier; its role depends on the memory type, topology, and workload.

Exhibit 1. Inference moves the memory problem from model weights to active state. CXL expands the set of places where that state can live.

Deployment flexibility

It is noteworthy how early we are in the rack scale compute era. By our estimates, using our accelerator installed base model, we believe rack scale accelerators in the range of 9-11% of the total AI accelerator installed base. The shift to rack scale solutions are the necessary catalyst that will enable CXL, and other solutions even if custom, to be designed by the customer. Knowing the data center customer is scaling their rack scale infrastructure helps as a catalyst for CXL whether that deployment is north south or east to west in its location. Enough of the ecosystem was present, through announcements and demonstrations at FMS, to make the deployment path more visible. Memory suppliers, interconnect companies, switch vendors, custom silicon providers, and networking ASIC companies were all discussing a path toward viability with much different language than earlier in the year. As of now, we expect the larger deployments are more likely to arrive in 2027 and build into 2028, yet the ecosystem is now moving from evaluation toward qualification, driven by a key set of customers.

We believe hyperscalers are the first catalysts for several reasons. They control rack specifications, work with ODMs, and can make system-level TCO decisions that are difficult in standardized OEM deployments. They also have access to a large legacy memory resource that is not always useful in its original server configuration, but could become valuable again in a CXL-based system. We explain the size of that resource and the qualification path in the full report. That flexibility gives hyperscalers room to qualify a CXL memory box, attach memory beside a CPU or custom ASIC, or build a dedicated memory system. The economic benefit is that they can add controller and system content around capacity they already own.

Our key take here is: We have increased confidence that the CXL standard will emerge as a preferred additional approach to infrastructure build out. Its economic and TCO benefits, detailed in the full report, along with its flexibility in implementation give it many advantages over other solutions. While we understand the tradeoffs, we believe the biggest customers in the world are positioned to drive the adoption of CXL and have distinct advantages over others in the market. The competitive advantage available to those customers through CXL could become evident.

Inside the Full Report

• Why hyperscalers are likely to become the first large CXL customers, and how their ODM model helps them qualify custom systems faster.

• How much existing memory could become reusable, what that changes for deployment timing, and the potential TCO benefit.

• Where CXL fits alongside HBM, local DDR and MRDIMM, HBF, proprietary memory attach, and networked-memory approaches.

• Our CXL market model through 2030, including the path from single-host expansion to rack-level pooling.

• The beneficiary map across controllers, memory suppliers, interconnect, systems, and software, plus the production signals needed to validate each opportunity.

Read more

In our prior report, The Hyperscaler Capacity Partner Hierarchy, we talked about the importance of partners for hyperscalers, particularly those with a specific type of business model that allowed them to keep a preferred margin profile when they could not secure infrastructure fast enough to convert their backlog. We think, as the market adopts behind-the-meter power and standards emerge, this can accelerate time to power and perhaps shift the speed at which hyperscalers can get land, shells, and other physical infrastructure in place (all things easier to secure and build), then work with a partner like Bloom, and others, to bring capacity online while they still work to get broader grid connectivity.

Grid connectivity is the hardest part of this equation. The hyperscalers that needed to monetize their compute were doing deals with those who had secured power via the grid, often on less favorable terms that impacted their margins, because they were desperate for power. As BTM scales and becomes more viable, it gives them another choice since it can operate without going through all the hoops needed to secure a grid connection. We are not saying they will stop pursuing grid connections. By adopting BTM, they can scale the easier part of the project, if anything in this process is actually easy: getting a shell built and all the surrounding infrastructure in place. That gives them first-party ownership instead of forcing them to go through partners simply because those partners have grid contracts done and ready to go. The maturation of BTM means the option now exists to add the grid later. That is why BTM can cost more and still be the better choice.

Inference changes the equation for BTM

From conversations a year ago, when BTM seemed more like a theoretical with potential but needed to be proved out, a few things have changed. First, the AI infrastructure mix moving from training toward more inference is a big driver. In training, the workloads are bursty and thus prone to massive power spikes versus more nominal and consistent power draw. Inference is less bursty and more manageable as a workload, meaning that having a range of redundancies in place to handle power spikes, which was a challenge for pure BTM, is less of an issue with inference. But even with training workloads now, the latest GPUs and rack-scale systems are putting more power control into their systems to help regulate power spikes and make them smoother. One data center operator we spoke with told us anything around 25% or lower for this kind of power spike was low enough that BTM can suffice, and that the latest rack-scale systems can keep those spikes at 25% or lower.

Public data supports the view from our conversations with those in the power industry. Microsoft production data measured a maximum two-second change of 37.5% of provisioned power for training, compared with 9% for interactive inference. NVIDIA’s GB200 documentation describes programmable power smoothing, while its newer Vera Rubin rack architecture adds more local energy buffering. We would not treat 25% as a universal engineering cutoff. It is one operator’s practical marker, and the broader point is that better rack-level power management gives the onsite generation system a smoother load to follow. Smoother workloads are one enabler of the BTM shift. Grid delays and the economics of bringing compute online earlier remain the larger forces.

Customers have moved from “can we use BTM?” to “how will we use it?”

We are not saying BTM becomes the standard or only source of power, only that BTM was not in a place where it was as viable as it appears now. That waas clear from Bloom’s call where they disclosed every major customer is now engaged to use BTM as a way to get to revenue faster. Customers have moved from BTM looks good on paper to all systems go to ramp BTM. We are in a cycle where the supply chain needs to scale and start to ramp to meet demand so customers can use BTM either to bridge the time to grid service or offset some grid needs and work toward more favorable power economics by using a hybrid architecture. BTM seems to be inflecting, for all the reasons we include, and now the main players need to start ramping their supply chains to meet demand.

From talking to operators in the field, we had consistently heard about the challenge that is electrical integration for the full site. While the equipment that goes into a BTM solution is one piece of the puzzle, the customer still needs to secure labor and testing, then bring together switchgear, transformers, UPS capacity, microgrid controls and a host of other things. The best-positioned companies are those that have already secured the fuel, have utility relationships and have sites ready to go.

Power is only part of the equation when analyzing who captures most of the value. This is why the ability to control the complete delivery schedule and stand behind the contract matters, because any delay has a direct impact on the customer’s time to revenue. The guarantee must cover an operating outcome. That shifts the risk of turning a power plan into operating compute away from the hyperscaler and onto the power or capacity partner.

Bloom is the clearest public proof point for now

Bloom went from essentially entering this market nine months ago, per the call, with one customer, to management saying its technology is now validated and approved by all major U.S. hyperscalers and more than a dozen other AI infrastructure customers. Those relationships span live deployments, booked and shipped systems under construction, and definitive agreements. Customer engagement has clearly inflected. The next proof is whether those approvals become repeat deployments and accepted, revenue-producing critical IT MW.

The reported results now sit behind that claim. Revenue reached $1.065 billion in Q2, up 165.5% from a year earlier, while GAAP gross margin increased to 33.4%. The next proof is repeat orders and accepted, revenue-producing critical IT MW.

A handful of large projects can make the market appear broader than it is. Developers may reserve several generation options for the same campus, while different suppliers count the same prospective demand. Repeat deployments provide stronger evidence. They show that the first installation worked well enough for the customer to use the architecture again. Accepted, revenue-producing critical IT MW then confirms that the project cleared every gate: fuel, permits, generation, electrical integration, commissioning, and usable compute. That would establish BTM as a repeatable procurement model rather than a collection of emergency power projects.

Inside the Full Report

• How BTM gives hyperscalers a faster path to owned compute

• The six gates between a power plan and usable capacity

• Where Bloom leads and how engines and turbines compete

• What Bloom’s Q2 confirms and what still needs proof

• Who captures value across the BTM delivery stack

• How BTM can scale and what would weaken the thesis

Read more

We have spent a lot of time tracking power and utilities because we continue to believe power will set the pace of the AI infrastructure buildout. Our recent conversations with industry sources keep coming back to the same problem. We have known about the power bottleneck for some time. What we are hearing now is that it has not gotten any better and, if anything, is getting worse. Data center demand continues to move faster than local utilities can support it. Grid studies can take six to twelve months, and those timelines are getting longer. A new substation can add another two to three years. A company can sometimes move faster if it pays for the substation work itself, but that makes the project more expensive. Even then, the equipment and skilled labor needed to finish the work are getting harder to secure on schedule. Permitting and local resistance can push the timeline out again. Based on what we are hearing, we think the power constraint is likely to stay with the industry through at least 2030 and probably longer.

As we have heard directly from hyperscalers, this timing problem has been shaping how they think about capacity for some time. Demand is coming in faster than utilities and the normal data center build cycle can support. That leaves them using outside partners that already have firm power or can bring a site online sooner.

Calling a company an power/shell/ landlord, neocloud or data center developer only tells us so much. What we care about is whether the site can actually get power and be delivered on time. Where we may differ from many consensus is our belief that the best fit is often a partner that can bring the power and the building online, then stop there. The hyperscaler still controls the compute and the customer, which keeps more of the economics inside its own business.

Amazon and Microsoft will put the capacity gap back in focus

This report will publish during the same week Microsoft reports fiscal Q4 results on July 29 and Amazon reports Q2 results on July 30. We expect both companies to raise the capital-spending bar and show another increase in contracted cloud demand. The labels differ between backlog and remaining performance obligations, but the economic signal is the same: Google (reported), AWS, and Azure are booking demand faster than internal infrastructure can be delivered.

The chart below puts the imbalance on one scale. Across the companies shown (full universe including meta and neoclouds), backlog and RPO have grown faster than cost-adjusted capex. That pulled the ratio from 44% in 2024 to about 37% in 2026. We would not treat this as a measure of physical capacity coverage because capex is an annual flow and backlog is a point-in-time balance. Still, the direction is clear. Contracted demand has grown faster than the capital response, which is why we do not expect capex to slow anytime soon. Key point to remember - revenue always lags capex.

That leaves the economics of the partner as the main question. Hyperscalers will keep using outside capacity while they build more of their own. On Alphabet’s latest earnings call, management called that capacity a bridge and said it would create modest near-term margin pressure. Google is willing to pay the premium because the revenue is available now. Waiting for its own data centers would mean leaving some of that demand unserved. As Google brings more controlled capacity online, part of the premium should go away. The type of partner, and how much of the stack that partner owns, will determine how expensive the bridge becomes. Alphabet Q2 2026 call transcript

Our prior reports looked at the AI cloud from the supplier side. We mapped where hyperscalers and neoclouds compete, then separated neoclouds by the layers they own.

This report looks at the same issue from the buyer’s side. We think AWS and Azure face the same choice Google described. They can, and will, buy outside capacity now and accept some pressure on unit economics, then move more of the workload into controlled infrastructure as their own supply arrives. The question is who is best suited to the hyperscalers are a foundational capacity partner.

That is why we are spending more time on the partner type. A company that secures firm power and delivers the physical layer can remain an attractive partner after the scarcity premium fades. The hyperscaler still controls the server system and the customer. From here, we want to know which providers can deliver the same kind of project more than once.

External capacity is not one product

A complete compute service and a leased data center shell can both deliver capacity, but the buyer is paying for very different things. A managed compute provider supplies the building and the server system, then operates it. The price has to cover the cost of the hardware and the risk that the provider cannot keep it fully used. It also has to cover the chance that the equipment loses value faster than expected. All of that adds another supplier margin to the cost.

A third-party-owned shell stops at the physical layer. The landlord develops the site and delivers a building that can support the required density. The hyperscaler can still own the accelerators and decide how the network is designed. It also keeps the workload inside its own control plane, which is what the hyperscalers generally prefer. More of the cloud margin stays with the buyer, while someone else owns the slow, long-lived physical asset.

This is the capacity hierarchy we are trying to describe. When there is enough time, hyperscalers prefer to own and operate strategic capacity themselves. A leased shell can get them there sooner without giving up control of the compute. Managed third-party compute still makes sense when the value of serving demand now is greater than the premium being paid.

Exhibit 1. The Hyperscaler Capacity Hierarchy. Relative profiles are Creative Strategies analytical judgments, not reported provider margins. Source: Creative Strategies analysis of Alphabet public disclosures and public infrastructure agreements.

The best landlord owns more than land

Power delivery is the known hook making these partners attractive. The latest disclosures help answer the next question: what turns that power into a valuable contract? Applied Digital delivered another 75 MW on schedule, while Core Scientific had 437 MW billing by mid-July against 1.1 GW leased. More important for our thesis, Applied Digital said direct hyperscaler leases should lower its financing cost over the full term. Core Scientific’s mix of direct AMD leases and neocloud leases with AMD protections shows how the risk can change by project. The value sits in converting power into accepted capacity under a contract the market can finance with a specific margin profile. Applied Digital Q4 2026 earnings call Core Scientific Q2 2026 results Core Scientific and AMD partnership

Hut 8 may add another version. The company has leased 704 MW at Beacon Point to an unnamed investment-grade tenant. The Financial Times reported that NVIDIA is the tenant and may sublease the capacity to neocloud partners, although neither company has confirmed it. If accurate, NVIDIA would be securing physical capacity itself rather than waiting for partners to find it. Hut 8 Beacon Point announcement Reuters summary of the FT reporting

Chip vendors may be starting to secure physical capacity themselves. AMD is doing it directly with Core Scientific, while NVIDIA is reported to be leasing Hut 8 capacity that it may place with neocloud partners. The obvious reason is speed, but we think there may be a competitive angle as well. Hyperscalers are putting more custom ASICs into the infrastructure they control. Securing outside sites could help NVIDIA keep scarce power tied to NVIDIA systems before those sites are absorbed into hyperscaler builds. That is our interpretation, not something NVIDIA or Hut 8 has said, but it would expand the buyer pool for the physical layer.

For now, our perhaps out of consensus view, is we still think the best long-term position may sit with the company that controls the physical asset and stops before the compute layer. Hyperscalers will use neoclouds when speed is worth the premium. Chip vendors may now compete for the same sites. All of them still need the landlord to deliver.

Inside the Full Report For Clients and Subscribers

• An original model for comparing the cost of delay with the premium paid for managed compute.

• A buyer-side margin and capital sensitivity across owned, leased-shell, and managed-compute structures.

• Google and Meta case studies showing how the same buyer uses different structures for different time horizons.

• A contract-quality ladder separating direct hyperscaler leases from backstopped neocloud tenancy and uncontracted pipeline.

• A selected capacity map showing where supply sits, followed by public-company market, beneficiary, and partner-fit maps that separate economic role from announced MW.

• A deployable-MW proof ladder and monitoring framework for testing whether announced capacity can become revenue-producing infrastructure.

Read more

This retrospective draws on a wide range of research from third party sources as well our own. Much of that work saw AI coming. The largest misses came from underestimating how quickly the pieces around the accelerator would have to scale together. Something we now refer to as the GPU Tsunami.

How much the market moved

• Four years early: A 2022 forecast put the semiconductor market at $1 trillion in 2030 (historical framework). Our current model reaches $1.79 trillion in 2026 and $3.49 trillion in 2030. Adding adjacent AI infrastructure brings the combined 2030 silicon-systems pool to $3.95 trillion.

• 5.8 times: Our 2027 accelerator model is now $720 billion, compared with an old $125 billion AI-compute endpoint.

• 60%: One quarter of NVIDIA Data Center revenue reached about 60% of the old full-year 2027 AI-compute endpoint.

• 2.7 to 3.2 times: The 2025 CoWoS capacity estimate was revised from 25,000 to 30,000 wafers per month to more than 80,000.

• 79%: The 2027 AI-switching endpoint was revised higher in less than a year.

• 3.0 to 3.6 times: Current frontier rack power surpassed an old 2030 expectation four years early.

• 2.4 times: Our seven-company 2027 capex model is already 2.4 times an old global data-center endpoint.

We think its important to look back, as a post mortem, to understand where the analysis went wrong and what the collective industry (us analysts) underestimated, so we can recognize those patterns when they appear again. Some earlier calls were right, while others moved too quickly on adoption timing. We also missed how much the unit of analysis was changing as the market developed. We were particularly too conservative on the ASP expansion across the semiconductor supply chain. We fully modeled semiconductor output based on our foundry capacity model, but we did not fully appreciate the pricing leverage created by the AI infrastructure cycle. Our midpoint scenario implies the industry’s blended semiconductor ASP rises from approximately $0.75 today to more than $2.10 by 2028, nearly tripling in just three years. Between 2025 and 2028, we estimate semiconductor unit shipments increase only 25–30%, while the average semiconductor ASP increases nearly 185%, driving approximately 260% industry revenue growth. This represents a structural shift in semiconductor industry economics as value per device compounds substantially faster than industry unit shipments. We map every constraint in the below report leading to this ASP increase, in the era we now call the era of margin expansion.

What the archive saw, and where its boundary failed

We reviewed a broad set of historical forecasts to understand how expectations changed over time. Licensed historical work is described as third-party forecasts or archive estimates, while current comparisons use public outcomes and Creative Strategies models.

The scorecard separates outcomes, run rates, and model revisions. A newer forecast shows expectations moving; only an outcome proves the old forecast wrong.

Each row uses a different type of evidence, but the pattern is consistent. The largest forecast revisions appeared once the market moved beyond the accelerator and began modeling the complete AI system. Compute only becomes usable capacity when the full rack can be commissioned. The early forecasts treated the infrastructure around the accelerator as a supporting input. It turned out to be part of the market itself.

Compute demand escaped the original boundary

In March 2023, a third-party forecast assumed that 40 to 60 large-model builds over the following 12 to 24 months would create approximately $10 billion to $15 billion of incremental GPU TAM. A June 2023 custom-silicon model put AI computing semiconductors at roughly $43 billion in 2023, growing to $125 billion in 2027, with custom ASICs reaching as much as 30% of cloud AI semiconductor spending.

Those estimates were aggressive at the time, but their boundary was still too conservative. The model counted likely builds and the accelerators required, then extended the curve as custom silicon gained share. It did not capture the wider inference market that would form around installed compute. Ultimately, what most got wrong was underestimating the compute intensity of the workload that is agentic AI. We think many still under appreciate this today as well.

NVIDIA reported $75.2 billion of Data Center revenue in Q1 FY2027. Data Center includes systems and networking, so it is not like-for-like with semiconductor TAM. Even so, that quarter reached 60% of the old 2027 AI-compute endpoint. In one quarter, NVIDIA generated revenue equal to 60% of what the earlier model expected from the entire AI-compute market for all of 2027. Looking forward, our current model puts merchant GPU and XPU logic plus custom AI accelerators at $770 billion in 2027 and $1.36 trillion in 2030.

The models underestimated how much demand each new accelerator would create. Training built the installed base, while better performance expanded the market for inference. That pulled custom silicon and rack-scale systems into what had started as a GPU market. As supply grew, the TAM grew with it.

Packaging was the clearest under call

In August 2023, a third-party advanced-packaging forecast put industry CoWoS capacity at roughly 15,000 wafers per month, rising to 20,000 to 25,000 in the second half of 2024 and 25,000 to 30,000 in 2025. The forecast saw little likelihood that CoWoS would remain a meaningful bottleneck beyond 2024.

By January 2025, a later archive model expected TSMC CoWoS capacity to exceed 80,000 wafers per month by the fourth quarter of 2025. That was 2.7 to 3.2 times the 2023 estimate for the same year. In July 2026, TSMC still described packaging as tight enough to limit customer growth. TSMC Q2 2026 materials.

In retrospect, what was missed was the timing, pull-in, of chiplet designs as AI accelerators were the first evidence of the broad shift from monolithic to systems based chip design.

Packaging became part of the capacity boundary, which is the central idea in our report The GPU Tsunami: TSMC, Intel, and Samsung Foundry. A leading-edge wafer becomes useful AI capacity only after advanced packaging brings it together with HBM and the rest of the rack.

Memory became a much larger market than the models allowed (and structural)

Memory was not missing from the early forecasts. It was treated as a cyclical recovery. In late 2023, the public market forecast called for roughly $130 billion of memory revenue in 2024. Actual sales reached $165 billion, 27% above that estimate.

The longer-range forecasts were even more conservative. One public model published in 2024 put combined DRAM and NAND revenue at approximately $206 billion in 2025 and just $214 billion in both 2026 and 2027. The assumption was that the recovery would level off once pricing normalized.

Our current model looks very different. We estimate combined DRAM and NAND revenue of $230–240 billion in 2025, $550–570 billion in 2026, and $800–850 billion in 2027. Against the old forecast, the 2026 market is now modeled at 2.6 times the prior estimate. By 2027, the difference reaches approximately four times.

This is a forecast revision, not yet a realized outcome, yet trending in that direction. But it shows what the earlier models missed. AI did not create demand only for HBM. It increased the amount of server DRAM and enterprise NAND required around each accelerator. At the same time, HBM consumed capacity that would otherwise have served conventional memory. The result was a much broader shortage and a much larger market than the original forecasts allowed.

The bottleneck kept moving

The early work saw the constraint moving beyond the accelerator, but underestimated how quickly the network and site would control the deployment schedule.

A June 2023 forecast put 2027 AI-switching revenue at $8.5 billion. Eleven months later, that endpoint had risen 79% to $15.2 billion. Our broader networking-silicon model now reaches $105 billion in 2027. The categories differ, but reported revenue confirms the direction: Broadcom’s $10.8 billion of quarterly AI semiconductor revenue reached 77% of its old full-year FY2025 forecast. Broadcom Q2 FY2026 results.

Power moved faster still. A July 2024 note expected average rack density to reach 40 kilowatts by 2030. GB200 and GB300 systems reached 120 to 142 kilowatts four years early, or 3.0 to 3.6 times the old endpoint. These are frontier systems rather than fleet averages, but they set the next building standard. GB200 specification and GB300 reference architecture.

Capex followed. A May 2024 model put global data-center capex at $500 billion in 2027. Our seven-company cost-adjusted base now reaches $1.18 trillion, or 2.4 times that endpoint.

The bottleneck moved from the chip into the network and then the site. Each fix exposed the next constraint. That is why Counting Real AI Capacity and Gigawattonomics focus on commissioned systems and economic output per watt. Until the full system is commissioned, announced capacity remains intent.

The top line hid the infrastructure reallocation

A January 2024 semiconductor model forecast sales of $645 billion in 2024 and $718 billion in 2025. WSTS reported $630.5 billion and $795.6 billion, respectively. The model was 2.3% high for 2024 and 10.8% low for 2025. WSTS 2025 result. It caught the recovery but missed the next acceleration in memory pricing and AI-infrastructure mix.

Public cloud was also broadly on track. A June 2024 model forecast spending rising from roughly $675 billion in 2024 to $1.38 trillion in 2028. The surprise occurred inside the total as AI infrastructure grew faster and carried more capital intensity than conventional workloads.

A top-line forecast can land within normal error while the internal demand map changes enough to redirect capital, capacity, and profit. The semiconductor and cloud totals did not reveal how much of the next dollar would be pulled toward accelerators and the infrastructure required to deploy them.

Enterprise AI had two opposite forecast errors

Infrastructure estimates were generally too small. Enterprise adoption produced two errors in opposite directions. Spending and usage intensity ran above the early markers, while scaled-production timing was sometimes too aggressive.

Two 2023 third-party forecasts framed an $820 billion enterprise-software TAM and approximately $150 billion of GenAI software spending within three years.

Our current working range puts enterprise GenAI spending across software, services, and inference compute at $175 billion to $200 billion in 2026, rising to more than $600 billion by 2030. The boundary is broader than software alone, so the comparison is directional. Even with that caveat, the 2026 midpoint is about 25% above the old three-year marker, and the decade-end range is more than four times as large.

The survey archive explains the other error. A 2024 IT survey found 52% of respondents live with at least one AI use case, but only 10% in production at scale. Later CIO surveys measured production at 30% in 2024 and 39% in 2025, while an earlier CTO survey expected nearly universal use by year-end 2024.

The error was treating a company with one live use case as equivalent to a company that had reorganized a production workflow. An enterprise can pay for AI and increase token consumption while only a small number of workflows reach scaled production. In our report, From AI Usage to AI Earnings Power picks up there. The test is whether repeated use changes a measurable operating baseline enough to earn a durable budget.

Where the current Diligence Stack work goes next

The current research starts from this revised unit of analysis. Counting Real AI Capacity establishes what has actually been commissioned. Gigawattonomics then asks whether that capacity can earn an acceptable return from the power it consumes.

Below the site, The GPU Tsunami applies the same capacity discipline to foundries. Where AI Constraints Become Pricing Leverage asks which shortages can support durable economics rather than temporary scarcity pricing. The memory and enterprise reports carry the method into capacity allocation and workflow monetization.

Bottom line

As a firm that has been doing market models, forecasting, sizing, and more for over 40 years, we understand being conservative, and how forecasts are almost always wrong. A forecast is only as good as the underlying assumptions we stay in a state of constant learning, observing, and looking for the right past and present patterns in order to continually strengthen our assumptions on every part of the industry we research.

We shared a lot of our internal forecasts and models in the look back, knowing full well they may be wrong and also pose the scenario, within the historical view, that what if even today we are not bullish enough.

For subscribers

Below the paywall, we share our notes from attending AMD Advancing AI and having time with management to discuss products and strategy and Intel earnings with feedback from our calls with IR. The Intel note explains why the earnings recovery may arrive before an external foundry win, with custom ASICs emerging as another business to watch. The AMD note looks at Helios and whether its open rack-scale design can turn strong component technology into a coherent system.

Read more