Editor's Note: This interview offers a window into the AI computing power cycle: demand has not cooled down following the previous wave of GPU procurement frenzy; instead, it is being continuously driven higher by agents, inference, and enterprise-grade AI applications.
This article features an interview with Brannin McBee, Co-founder and Chief Development Officer of CoreWeave, and Nick Robbins, Vice President of Corporate Development and Investor Relations, discussing the current state of AI demand and the neocloud market. The core message from CoreWeave executives is direct—AI demand seems to be intensifying daily in new ways, and the real bottleneck is shifting from "having GPUs" to more complex infrastructure issues: data center powered shells, CPUs, storage, electricians, supply chain execution capabilities, and how much customers are willing to pay for next-generation computing power.
CoreWeave's uniqueness lies in its position in the middle of the AI infrastructure chain: it serves leading clients like OpenAI, Anthropic, Meta, Google, Microsoft, and Nvidia, while also directly sensing demand changes from research labs, enterprise customers, and hyperscale cloud providers. Therefore, what it observes is not just "GPU shortages," but structural changes in AI workloads themselves. With the rise of agentic AI and reasoning models, computing power demand is no longer solely centered around GPUs; the importance of CPUs and storage is also increasing. New-generation data center designs must reserve space for Vera CPUs, Vera Rubin servers, and more storage.
This also explains why AI infrastructure competition is shifting from mere chip procurement to more comprehensive engineering delivery capabilities. Those who can secure powered data centers, deploy servers, streamline supply chains, and optimize cost per token faster are closer to the core of this round of AI capital expenditure cycles. CoreWeave repeatedly emphasizes being "customer-driven," which reflects a larger judgment: AI cloud providers are no longer just selling computing power; they are proactively reconstructing the next-generation AI factory based on the roadmaps of their most advanced clients.
For investors and industry observers, the most notable aspect of this interview is not any single data point, but the direction of change in AI infrastructure demand: GPUs remain important, but bottlenecks are spreading; Nvidia is still the core, but CPUs, HBM, storage, and data center power supply capabilities are becoming new variables; AI demand continues to grow, but future success may depend on who can deliver complex infrastructure consistently, stably, and at scale.
The following is the original text:
CoreWeave is regarded as an innovative early market leader in the neocloud (new type of cloud services) domain.
It is the only cloud service provider to have received the highest "Platinum Rating" from AI research firm SemiAnalysis. Founded in 2017, CoreWeave provides large-scale GPU computing power to startups and large enterprises.
Key Context recently interviewed Brannin McBee, Co-founder and Chief Development Officer of CoreWeave, and Nick Robbins, Vice President of Corporate Development and Investor Relations, to discuss the current state of AI demand and the neocloud market.
The following are edited highlights from the conversation:
AI Demand Continues to Intensify
Tae: When did the wave of demand for agentic AI begin to surge?
Brannin: We saw the real beginnings in the fourth quarter of last year. At that time, we were engaging in engineering-level discussions with clients about the products they expected to bring to market in the first quarter of this year.
This has always been a crucial perspective for how we view customer demand. We have a deeply interconnected engineering relationship with our customers. It is this relationship that allows us to see trends ahead of time, rather than reacting after changes occur.
If you look at the product landscape of the AI market, I'd say the first quarter was the moment of a massive inflection point for inference and AI consumption, and this acceleration is still continuing now.
Tae: What is the current state of AI demand? Compared to a few months ago, has there been absolutely no sign of slowing down in recent weeks?
Nick: It seems to be intensifying in new ways every day.
Tae: Could you discuss the rising trend of CPU demand relative to GPUs in the agentic AI wave? Would you deploy racks of Vera CPUs alongside Nvidia GPU servers?
Brannin: CoreWeave has been running CPUs since 2023. We've always had a full cloud product offering. So the question isn't whether we just started adding CPUs, but rather, what do customers actually need? And is this demand rising in relative terms? The answer is, very clearly, yes.
As agent and reasoning capabilities truly emerge within models, storage demands are also increasing compared to previous generations. I believe this trend will continue.
Nick: To your question, the answer is yes. You will absolutely see lots of Vera CPUs deployed alongside lots of Vera Rubin servers. Last year, we fundamentally redesigned our base data center solution to make room for more storage and more CPUs to be deployed alongside GPUs.
We did this because we are in a very unique position within the entire ecosystem. We are the only independent cloud provider serving all the most advanced technology users. No other independent AI cloud provider can say that Anthropic, OpenAI, Meta, Google, Microsoft, Nvidia, etc., are all their customers.
This creates a beneficial flywheel, or positive feedback loop, for our business: we understand where customers are taking the technology and plan accordingly.
The Bottleneck Is No Longer Just GPUs
Tae: Will you primarily use Nvidia Vera CPUs in the future?
Nick: That depends on the specific workload. Our actions are driven by customer demand. We do expect to be an early and significant adopter of Vera CPUs, which we have disclosed. Currently, our fleet is actually predominantly AMD, but this may change over time based on customer needs. There is very strong customer interest in Vera CPUs.
Brannin: This is also a good reminder to talk about how our contracts work. As you probably know, over 98% of our revenue is contract-driven. We are not guessing what infrastructure customers want. Customers tell us very explicitly what configurations they need. Everything is customer-driven. It is the customers defining what we build.
Tae: Let's talk about the competitive landscape. How do you enter the market and compete against neoclouds like SpaceX, Nebius, Oracle, and hyperscalers like Azure, AWS, and Google?
Brannin: In terms of differentiation, I prefer to look at it from a third-party validation standpoint. Nine out of the top ten AI labs globally, excluding China, are using our platform. SemiAnalysis has consistently ranked us alone at the highest level for performance. I don't think the GPU allocation we receive is due to any personal friendship with Jensen.
This indicates that suppliers have deep confidence in our execution track record and engineering capabilities, believing we can best showcase their product capabilities globally.
Nick: We win hyperscaler customers because we are exceptionally good at execution. We can build these systems incredibly fast, and they run exceptionally well. We win research lab customers because we provide the highest-performing versions of the technology and perform best in per-token efficiency.
We win enterprise customers because the infrastructure simply runs well, and we've built a very excellent, best-in-class orchestration layer, which is also recognized by things like the Platinum Rating.
But increasingly importantly, among AI cloud providers, we have built out the most mature layer of capabilities covering inference and development tools, helping enterprises actually put AI into production.
This means we are building and delivering products that ultimately help enterprises with relatively lower technical maturity to transform data into models, and models into agents that can run internally, while we can cross-sell CoreWeave cloud services in the process.
Tae: What is the current bottleneck? Is it data center powered shells? GPUs? Or electricians?
Brannin: It's powered shells. More specifically, the components inside these shells. You specifically mentioned electricians, which is absolutely correct. It's a complex area.
But importantly, we already have 49 of these sites live and operational. We are not pinning our hopes on one or two sites. We've done this 49 times.
This is a very deep execution track record.
It also means we have accumulated a vast amount of knowledge on how to handle supply chain issues, which suppliers in that supply chain are good to work with, and which ones aren't.
Tae: What can you share about the cost and shortage of HBM memory? How are you managing this? Do customers need to bear the increased costs?
Nick: The answer is yes. Our business model is designed to lock in the GPU price we charge customers at the same time we sign GPU purchase orders and determine our cost. More broadly, that's the server price, and the server price obviously includes HBM cost.
This is how we isolate ourselves from day-to-day price fluctuations.
If our component costs rise in the next transaction, we reflect that in the price we believe we can charge customers, thereby protecting our margins. We are very well protected in passing these costs on to customers. This is something we watch very closely.
Currently, sourcing components is not the biggest bottleneck. The biggest bottleneck is the powered shell. But that answer may shift back and forth at some point in the future.
Tae: How do you expect the deployment ramp for Vera Rubin to unfold? What will the second half of this year look like?
Nick: We are clearly the first company globally to have spun up and fully validated VR, the Vera Rubin rack. We did the same with GB200, GB300 last year. I expect VR to start appearing later this year.
I expect the really massive, very strong deployment ramp to run through all of 2027. This pace is similar to GB: GB started appearing in 2025, but the truly massive ramp was actually throughout 2026. That is to say, quite a bit was deployed by the end of last year, but this year is the year of truly massive GB deployment.
I expect VR to follow a very similar cadence over the next 12 to 18 months.
