Original Title: An Interview with CoreWeave Executives: AI Demand Seems to 'Intensify' Every Day

Original Author: Tae Kim

Original Translation: Peggy, BlockBeats

Editor's Note: This interview provides a window into observing the AI computing cycle: demand has not cooled due to the previous GPU buying frenzy, but is being elevated further by agents, reasoning, and enterprise-level AI applications.

This article interviews CoreWeave co-founder and Chief Development Officer Brannin McBee, as well as Vice President of Business Development and Investor Relations Nick Robbins, discussing the current state of AI demand and the neocloud market. The core statements from CoreWeave executives are straightforward—AI demand seems to intensify in new ways every day, and the real bottleneck is shifting from "Are there GPUs?" to more complex infrastructure issues: data center power enclosures, CPUs, storage, electrical work, supply chain execution capabilities, and how much customers are willing to pay for the next generation of computing power.

CoreWeave's uniqueness lies in its intermediate position within the AI infrastructure chain: serving leading customers like OpenAI, Anthropic, Meta, Google, Microsoft, Nvidia, while also directly sensing the demand changes from research labs, enterprise customers, and hyperscale cloud providers. Therefore, what it observes is not merely "Are GPUs in short supply?" but rather that the AI workload itself is undergoing structural changes. As agentic AI and reasoning models emerge, computing power demand is no longer solely focused on GPUs; the importance of CPUs and storage is also increasing, and new data center designs must allocate space for Vera CPUs, Vera Rubin servers, and more storage.

This also explains why the competition in AI infrastructure is shifting from merely chip procurement to more comprehensive engineering delivery capabilities. Whoever can more quickly establish powered data centers, deploy servers, streamline supply chains, and optimize per token costs will be closer to the core of this round of AI capital expenditure cycles. CoreWeave repeatedly emphasizes "customer-driven," which actually reflects a larger judgment: AI cloud providers are no longer just selling computing power but are restructuring the next generation of AI factories in advance based on the roadmaps of cutting-edge customers.

For investors and industry observers, the most noteworthy aspect of this interview isn't a single data point but rather the directional changes in AI infrastructure demand: GPUs remain important, but the bottlenecks are spreading; Nvidia remains central, but CPUs, HBM, storage, and data center power capabilities are becoming new variables; AI demand continues to grow, but the future outcomes may depend on who can deliver complex infrastructure sustainably, stably, and at scale.

Following is the original text:

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

It is the only cloud service provider to receive the highest "Platinum Rating" from AI research institution SemiAnalysis. CoreWeave was founded in 2017 to provide large-scale GPU computing power for startups and large enterprises.

Key Context recently interviewed CoreWeave co-founder and Chief Development Officer Brannin McBee, as well as Vice President of Business Development and Investor Relations Nick Robbins, discussing the current state of AI demand and the neocloud market.

Here are the edited highlights from this conversation:

AI Demand Continues to Intensify

Tae: When did the demand wave for agent AI begin to surge?

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

This perspective has always been very important as 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 passively responding after changes occur.

If we look at the AI market from a product perspective, I would say the first quarter marked a crucial turning point for inference and AI consumption, and this acceleration has continued to this day.

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

Nick: It seems to intensify in new ways every day.

Tae: Please discuss the rising trend of CPU demand relative to GPUs in the agent AI wave. Will you deploy rows of Vera CPU racks alongside Nvidia GPU servers?

Brannin: CoreWeave has been operating CPUs since 2023. We have always had a complete cloud product. So the question is not whether we just started increasing CPUs, but what exactly do customers need? Is this demand increasing in a relative sense? The answer is very clear, it indeed is.

As agentic AI and reasoning capabilities truly rise in models, storage demand is also increasing compared to previous generations. I believe this trend will continue.

Nick: To answer your question, absolutely. You will definitely see a lot of Vera CPUs being deployed next to numerous Vera Rubin servers. Last year, we fundamentally redesigned the base data center solution to allocate space for more storage and more CPUs so they can be deployed next to GPUs.

We are doing this because we are in a very unique position within the entire ecosystem. We are the only independent cloud provider that serves all the leading technology users. No other independent AI cloud provider can say that Anthropic, OpenAI, Meta, Google, Microsoft, Nvidia, etc., are all our customers.

This creates a beneficial flywheel for our business or a positive feedback loop: we can understand where customers are taking technology and plan accordingly.

Bottlenecks Are No Longer Just GPUs

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

Nick: It depends on the specific workloads. We are driven by customer demand. We do expect to be early and important adopters of Vera CPUs, as we have disclosed. Currently, our cluster is still primarily AMD, but over time, this may change based on customer requirements. Customers have a very strong interest in Vera CPUs.

Brannin: This also nicely reminds us to 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 customers want. Customers will clearly communicate what configurations they need. Everything is customer-driven. It is the customers defining what we build.

Tae: Let's discuss the competitive landscape. How do you enter the market and compete against neocloud players like SpaceX, Nebius, Oracle, as well as hyperscale cloud providers like Azure, AWS, Google?

Brannin: In terms of differentiation, I prefer to look at it from the perspective of third-party validation. Nine of the top ten AI labs globally, excluding China, are using our platform. SemiAnalysis has continually recognized us as the highest level in terms of performance. I don’t believe we achieved our GPU allocations due to a personal relationship with Jensen.

This indicates a deep confidence from suppliers in our execution record and engineering capabilities, believing we can best exemplify their products globally.

Nick: We are able to win hyperscale cloud provider customers because we are very good at execution. We can set these systems up at a very fast speed, and they run extremely well. We can win research lab customers because we offer the strongest performance versions of technology and excel in per token efficiency.

We are able to win enterprise customers because the infrastructure indeed runs well, and we have built an excellent, best-in-class orchestration layer that is a source of recognition such as the Platinum Rating.

But increasingly importantly, among AI cloud providers, we have built the most mature level of capability, covering inference and development tools to help enterprises truly bring AI into production.

This means we are building and delivering products that ultimately help enterprises with relatively low technological maturity convert data into models and then convert those into agents that can run internally, while we can also cross-sell CoreWeave cloud services in this process.

Tae: What is the current bottleneck? Is it powered shells that have established power conditions? GPUs? Or electrical work?

Brannin: It is powered shells, meaning data center enclosures that meet power conditions. More precisely, it is the components inside these enclosures. You specifically mentioned electrical work, and that is absolutely correct. This is a complex area.

But importantly, we have already launched and are operating 49 such sites. We are not pinning our hopes on one or two sites. We have done it 49 times.

This is a very deep execution record.

This also means we have accumulated a lot of knowledge about how to handle supply chain issues, knowing which suppliers are suitable for collaboration and which are not in this supply chain.

Editor's Note: Powered shells refer to the actual data center buildings, excluding the actual computing server hardware.

Tae: Can you share anything about the costs and shortages of HBM memory? How are you coping? Do customers need to bear the cost of price increases?

Nick: The answer is yes. Our business model is designed to secure the GPU price we charge customers at the same time we sign GPU procurement orders and determine how much we will pay for the components. More broadly, this refers to the server prices, which obviously include HBM costs.

This is how we isolate the impact of daily price fluctuations.

If our component costs rise in the next deal, we will reflect that portion of the cost in the price we think we can charge customers, thereby protecting our margins. We are well protected in passing these costs to customers. This is something we are very closely monitoring.

Currently, acquiring components is not the biggest bottleneck. The biggest bottleneck is powered shells. But at some point in the future, this answer may change back and forth.

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

Nick: We are obviously the first company globally to launch and fully validate VR, the Vera Rubin cabinets. We did the same with GB200 and GB300 last year. I expect VR to start appearing later this year.

I anticipate that a truly large-scale, robust deployment ramp will span throughout 2027. This rhythm is similar to GB: GB began to appear in 2025, but the true large-scale ramp actually spanned all of 2026. That is, a significant amount was deployed by the end of last year, but this year is the true year for large-scale deployment of GB.

I expect that in the next 12 to 18 months, VR will exhibit a very similar rhythm.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。