CoreWeave Shifts Strategy as AI Inference Economics Reshape Infrastructure Demands

The efficiency of converting computational resources into useful intelligence is fast becoming the central metric for evaluating AI infrastructure. With the industry's focus shifting away from training and toward inference, the financial calculus driving infrastructure investment is being rewritten. CoreWeave is positioning itself at the center of this transition, moving beyond its traditional GPU compute offerings to encompass networking, storage and software components as inference workloads accelerate faster than training capacity. According to Dave Vellante, chief analyst at theCUBE Research, this fundamental reorientation is altering how organizations think about capacity planning, resource utilization and the per-token cost of operating at scale.

Speaking during a keynote analysis at the Fully Connected event, broadcast live on theCUBE, Vellante highlighted the dramatic shift in workload distribution. "Today they're at 50/50, and they expect to be 10/90 by next year," he said. "Both curves are growing. If you look at CoreWeave's numbers, they're off the charts." Vellante's remarks came during a conversation with Executive Analyst John Furrier, examining how inference expansion and system-wide optimization are reshaping the financial dynamics of AI infrastructure.

Token economics shift the infrastructure equation

The movement toward inference-heavy workloads demands greater attention to how effectively an entire AI system can transform raw compute into actionable output. This reality favors system designs that optimize across silicon, networking, storage and software in concert, rather than concentrating on isolated components. Such a shift represents movement toward evaluating performance at the system level rather than the component level, according to Furrier.

"If that shift happens, these data centers will look a lot different with the game still the same," Furrier explained. "Pump out as many tokens per watt as possible. Get the intelligence shipping. You got the perfect storm on the supplier side with Dell, ecosystems booming and CoreWeave in pole position."

Beyond raw performance metrics, the commercial picture extends into purchasing patterns. CoreWeave's customer base is increasingly demanding shorter contract terms, spot pricing mechanisms and on-demand consumption models, Vellante noted. These flexible arrangements command premium pricing relative to traditional long-term commitments with guaranteed minimum purchases, indicating that operational flexibility itself carries measurable economic value.

"The prices for those types of structures are much, much higher, and people are willing to pay," Vellante said. "It's going to be really interesting to see as that 98% starts to go down toward 50/50, how that's going to sort of affect the market."

CoreWeave builds beyond GPU capacity

While GPU scarcity may initially attract customers to CoreWeave's platform, the company is deliberately constructing a broader technology stack designed to deepen customer retention. The expanded offering now encompasses networking capabilities, storage infrastructure and a software layer centered on observability, security and iterative optimization, providing CoreWeave with greater leverage over the complete infrastructure ecosystem surrounding AI operations, Vellante noted.

"What you're seeing CoreWeave do strategically is they're expanding out beyond compute," he said. "They've got networking; they've got storage. They announced [CoreWeave] Forge today … they've got this software layer, which is observability. They've got security in there. They've got this closed-loop system between evaluation, observation, runtime curating … that's sort of how they're saying their software layer is now intact."

This comprehensive systems-oriented approach simultaneously strengthens CoreWeave's partnership with Nvidia Corp. While Nvidia's silicon remains foundational to the operation, competitive advantage increasingly derives from how efficiently the complete infrastructure ecosystem translates those components into productive tokens, Furrier explained. Consequently, silicon availability represents only one dimension of a multifaceted performance calculation.

https://www.youtube.com/embed/hjV3pKJe8YM?feature=oembed

"This is becoming increasingly a systems game; it's a systems race," he said. "Silicon matters enormously. What's useful commercially is how quickly the entire system turns into useful tokens. At the end of the day, that's the key."