Is compute power expansion endless? CoreWeave executive reveals: AI demand is "finding new ways to strengthen itself every day," and power supply data centers are now the key bottleneck for AI infrastructure.

The expansion momentum of AI computing power demand has yet to peak and is spreading to broader infrastructure layers.

CoreWeave co-founder and Chief Development Officer Brannin McBee and Vice President of Corporate Development and Investor Relations Nick Robbins recently told media that AI demand “is looking for new ways to strengthen itself every day,” with no signs of slowing down. The two executives pointed out that as the wave of agentic AI accelerates, market demand for ancillary resources like CPUs and storage is rising significantly relative to GPUs, leading to a fundamental rearchitecture of data center design.

Regarding the current principal bottleneck constraining the expansion of AI infrastructure, McBee clearly identified the “powered shell”—meaning the physical data center building with completed electrical infrastructure, not GPU chips or HBM memory. This assessment provides direct guidance for evaluating the pace of AI infrastructure investment.

Demand accelerates; agentic AI becomes the new driver

CoreWeave executives marked the starting point of this surge in AI demand as the fourth quarter of last year. McBee said the company, through deep engineering communication with customers, sensed early on that agentic AI products would be concentrated in market launches in the first quarter of this year. “From a product perspective in the AI market, the first quarter was a huge inflection point for inference and AI consumption, and acceleration is ongoing.”

Robbins described the current demand state even more directly: “It is looking for new ways to strengthen itself every day.”

This judgment comes from CoreWeave’s unique position in the AI ecosystem. According to Robbins, CoreWeave is currently the only independent cloud service provider serving Anthropic, OpenAI, Meta, Google, Microsoft, Nvidia, and other leading AI organizations, which gives it forward-looking insight into tech evolution and lets it plan infrastructure ahead of time.

Architectural overhaul: CPU and storage demand rises significantly relative to GPU

The emergence of agentic AI and inference models is changing the hardware mix logic of data centers.

McBee said CoreWeave has operated CPU resources since 2023, but the current trend is that demand for CPUs and storage relative to GPUs is obviously increasing. “As agents and inference truly take off in models, storage demand has risen significantly compared to prior generations, and I think this trend will persist.”

Robbins revealed CoreWeave fundamentally redesigned its standard data center solution last year to allow for more storage and CPU space.

He confirmed that there will be mass deployment of Nvidia Vera CPU racks alongside Vera Rubin GPU servers in the future.

Regarding CPU supplier choices, Robbins stated the current fleet mostly uses AMD, but as customers’ requirements evolve, Nvidia Vera CPU is expected to become an important early adoption direction, “There is great interest in Vera CPU.” McBee added, over 98% of CoreWeave’s revenue comes from contract-driven business, and customers specify the needed infrastructure configuration: “They are defining what we have to build.”

Powered shell — the most urgent expansion bottleneck right now

When asked about the biggest current constraint, McBee directly named “powered shells,” specifically mentioning a shortage of electricians as one complexity. He said CoreWeave now has 49 sites online, with plenty of practical experience in dealing with supply chain issues: “We know which suppliers to work with and which not to.”

Robbins responded to concerns about HBM memory cost and shortages, saying the company’s business model effectively hedges price risk—by locking in the client’s price when signing the GPU procurement order, preserving profit margin, and passing cost rises smoothly to clients. He added, “Parts acquisition isn’t the main bottleneck right now; powered shell is. But this answer might switch at some future point.”

Vera Rubin mass production: 2027 will be the big year

Robbins provided a clear timeline for Nvidia’s new Vera Rubin (VR) platform mass production.

He said CoreWeave is the first in the world to bring VR cabinets online and pass complete validation, with VR servers to start delivery in the second half of this year. Full-scale mass production ramp-up will last throughout 2027.

Robbins compared this pace to previous GB200/GB300 platform rollout—the GB series emerged in 2025, but true large-scale production was in 2026. “I expect VR will follow a very similar mode over the next 12 to 18 months.”

Building a moat with execution and ecosystem depth

Facing competition from ultra-large cloud providers (Azure, AWS, Google) and other emerging cloud players (SpaceX, Nebius, Oracle), CoreWeave executives cited three differentiators: execution speed, performance, and ecosystem depth.

McBee referenced third-party validation data, stating nine of the world’s top ten AI labs (excluding China) use CoreWeave’s platform. AI research firm SemiAnalysis awarded it a unique platinum rating. He believes Nvidia’s priority allocation of GPU resources to CoreWeave derives from high confidence in its engineering execution capabilities: “This is a supplier with deep faith in our execution record and engineering skill.”

Robbins illustrated competitive logic from customer segmentation: versus ultra-large clouds, CoreWeave wins with lightning-fast deployment and steady operation; for research institutions, with top performance and per-token efficiency; for enterprise clients, with the best inference and developer orchestration stack, helping them turn data into models and agents, thereby achieving cross-selling of CoreWeave cloud services.

Risk Warning and DisclaimerThe market carries risks. Investment should be cautious. This article does not constitute personal investment advice nor does it take into account individual user’s specific investment goals, financial situation, or needs. Users should consider whether any opinions, views, or conclusions herein fit their circumstances. Investing on this basis is at your own risk.