Original Title: An Interview with CoreWeave Executives: AI Demand Seems to 'Intensify' Every Day
Original Author: Tae Kim
Translation: Peggy, BlockBeats

Editor's Note: This interview provides a window into the AI compute cycle: demand has not cooled off due to the previous GPU frenzy but is instead being further driven by smart agents, reasoning, and enterprise-grade AI applications.

This article interviewed CoreWeave co-founder and Chief Development Officer Brannin McBee, and Vice President of Corporate Development and Investor Relations Nick Robbins, discussing AI demand and the neocloud market. CoreWeave executives' core statement is very straightforward—AI demand seems to intensify in new ways every day, and the real bottleneck is shifting from "having GPUs" to more complex infrastructure issues: data center power shells, CPUs, storage, electricians, supply chain execution capabilities, and how much customers are willing to pay for next-gen compute power.

CoreWeave's uniqueness lies in its position in the AI infrastructure chain: serving both top clients like OpenAI, Anthropic, Meta, Google, Microsoft, Nvidia, and directly sensing the demand shifts of research labs, enterprise clients, and hyperscale cloud providers. Therefore, it sees not only "GPU scarcity" but also a structural change in AI workloads themselves. With the rise of agentic AI and reasoning models, compute demands are no longer just about GPUs; the importance of CPUs and storage is also rising, and next-gen data center designs must accommodate Vera CPUs, Vera Rubin servers, and more storage.

This also explains why AI infrastructure competition is shifting from mere chip procurement to a more comprehensive engineering delivery capability. Whoever can quickly get power-supplied data centers, deploy servers, streamline the supply chain, optimize the cost per token, is closer to the core of this round of AI capital expenditure. CoreWeave repeatedly emphasizes being "customer-driven," but behind this is a larger judgment: AI cloud providers are no longer just selling compute power but are, based on the cutting-edge customer roadmap, pre-constructing the next generation AI factory.

For investors and industry observers, the most noteworthy aspect of this interview is not a specific single-point number but the changing direction of AI infrastructure demand: GPUs are still important, but the bottleneck is spreading; Nvidia remains core, but CPUs, HBM, storage, and data center power capacity are becoming new variables; AI demand continues to grow, but the future outcome may depend on who can consistently, stably, and at scale deliver complex infrastructure.

The following is the original text:

CoreWeave is seen as an innovative early market leader in the neocloud space.

It is the only cloud provider to receive the highest level of "Platinum Rating" from AI research firm SemiAnalysis. CoreWeave was founded in 2017, offering large-scale GPU compute to startups and enterprises.

Key Context recently interviewed CoreWeave Co-Founder and Chief Development Officer Brannin McBee, and Vice President of Corporate Development and Investor Relations Nick Robbins to discuss AI demand and the state of the neocloud market.

The following are the edited highlights from the conversation:

AI Demand Continues to Intensify

Tae: When did the wave of demand for AI intelligence start to surge?

Brannin: We saw the real beginning in the fourth quarter of last year. At that time, we were engaged with customers at an engineering level, discussing their products expected to hit the market in the first quarter of this year.

This has always been a critical perspective for us in understanding customer demand. We have a deep, intertwined engineering relationship with our customers. It's this relationship that allows us to see trends ahead of time rather than reactively.

From a product perspective in the AI market, I would say the first quarter was a significant inflection point for inference and AI consumption, and this acceleration continues to this day.

Tae: What is the current state of AI demand? Is there no sign of slowing down in recent weeks compared to a few months ago?

Nick: It seems to be intensifying in new ways every day.

Tae: Can you talk about the trend of CPU demand versus GPU in the wave of AI intelligence? Will you deploy rows of Vera CPU racks next to Nvidia GPU servers?

Brannin: CoreWeave has been running CPUs since 2023. We have always had a full suite of cloud products. So, it's not about whether we are just starting to add CPUs, but what do customers really need? Is there a relative increase in this demand? The answer is a resounding yes, indeed.

With the rise of intelligent agents and reasoning capabilities in models, the storage requirements have also increased compared to the previous generations. I believe this trend will continue.

Nick: In response to your question, the answer is yes. You will certainly see a large number of Vera CPUs being deployed alongside a large number of Vera Rubin servers. Last year, we actually fundamentally redesigned the underlying data center scheme to accommodate more storage and more CPUs, allowing them 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 cutting-edge tech users. No other independent AI cloud provider can say that companies like Anthropic, OpenAI, Meta, Google, Microsoft, Nvidia, etc., are all their customers.

This has created a beneficial flywheel for our business, or a virtuous cycle: we can understand where customers are taking the technology and plan accordingly.

The Bottleneck is No Longer Just GPU

Tae: Will you mainly use Nvidia Vera CPUs in the future?

Nick: It depends on the specific workloads. We operate based on customer demand. We do anticipate being early and significant adopters of Vera CPUs, as we have disclosed. Currently, our cluster is actually mainly AMD, but over time, this may change based on customer demand. There is a lot of customer interest in Vera CPUs.

Brannin: This also reminds us that we can talk about how our contracts work. As you know, over 98% of our revenue is contract-driven. We are not guessing what kind of infrastructure the customer wants. The customer will tell us very clearly what configuration they need. Everything is customer-driven. It is the customer defining what we are going to build.

Tae: Let's talk about the competitive landscape. Faced with neocloud players like SpaceX, Nebius, Oracle, and hyperscale cloud providers like Azure, AWS, Google, how did you enter the market and compete?

Brannin: In terms of differentiation, I prefer to look at it from a third-party validation perspective. Excluding China, nine out of the top ten AI labs globally are using our platform. SemiAnalysis has consistently singled us out for having peak performance. I don't believe we are getting such GPU allocations because of a personal friendship with Jensen.

This demonstrates that the supplier has deep confidence in our track record of execution and engineering capability, believing that we can best showcase their product capabilities on a global scale.

Nick: Our ability to win over hyperscale cloud provider customers is because we excel at execution. We can rapidly deploy these systems and they run very well. We can win over research lab customers because we provide the highest-performing tech stack and excel in efficiency per token.

We can win over enterprise customers because the infrastructure indeed runs well, and we have built a very strong, best-in-class orchestration layer, also recognized by accolades like platinum ratings.

But increasingly importantly, in the AI cloud provider space, we have built out the most mature layer of capabilities covering inference and development tools, helping enterprises truly operationalize AI.

This means we are building and delivering products that ultimately assist those relatively less mature in technology enterprises in turning data into models, further transformed into internally deployable intelligent agents, and we can cross-sell CoreWeave cloud services through this process.

Tae: What is the current bottleneck? Is it the data center shells already equipped with power? GPUs? Or electricians?

Brannin: It's the powered shells, which are essentially data center shells already equipped with power. More specifically, it's the components inside these shells. You specifically mentioned electricians, and that is absolutely correct. It's a complex field.

But importantly, we have already brought online and operationalized 49 such sites. We are not placing our hopes on one or two sites. We have done it 49 times.

This is a very strong track record of execution.

It also means we have accumulated a wealth of knowledge, knowing how to address supply chain issues, knowing which suppliers along this chain are suitable for collaboration and which are not.

Editor's Note: Powered shells refer to the data center building itself, excluding actual compute server hardware.

Tae: Can you reveal anything about the cost and shortage of HBM memory? How are you addressing it? Will customers need to bear the cost increase?

Nick: The answer is yes. Our business model is designed to sign GPU purchase orders, determine how much we need to pay for the cost, while locking in the GPU price we charge to customers. More broadly, this is also the server price, and the server price evidently includes the HBM cost.

This is how we isolate ourselves from daily price fluctuations.

If in the next transaction our component costs go up, we will reflect this cost increase in the price we believe we can charge customers, thus protecting our profit margin. We are very well protected in passing on these costs to the customer. This is something we pay very close attention to.

Currently, acquiring components is not the biggest bottleneck. The biggest bottleneck is the powered shell. However, at some point in the future, this answer may fluctuate.

Tae: How do you expect the deployment ramp of Vera Rubin to unfold? What will the situation be in the second half of this year?

Nick: We are evidently the world's first company to launch and fully validate VR, which is the Vera Rubin cabinet. We were the same last year with the GB200 and GB300. I expect VR to start appearing later this year.

I anticipate that a truly large-scale, very strong ramp-up deployment will span the entire year of 2027. This pace is similar to GB: GB started to appear in 2025, but the truly large-scale ramp-up actually spanned the entire year of 2026. In other words, a good amount was deployed by the end of last year, but this year is the real year of large-scale GB deployment.

I expect VR to follow a very similar pace over the next 12 to 18 months.

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia