How CoreWeave Sees the Market for Compute Right Now

The AI inference market is no longer a niche experiment—it's a corporate imperative driving explosive demand for compute, with companies like CoreWeave at the center of a global infrastructure boom. In this episode of Odd Lots, host Brandon McBee, co-founder and CDO of CoreWeave, reveals that enterprise adoption is accelerating beyond AI labs and hyperscalers, with financial institutions like Jane Street now running proprietary models directly on CoreWeave’s platform. Despite headlines of budget blowouts—Uber spent its entire 2026 AI budget in four months—McBee insists there’s no sign of pullback, only deeper, more strategic investment. The real bottleneck isn’t GPUs, but powered shells: fully energized, cooled data center spaces with the electrical infrastructure to support them. What’s more, the era of short-term GPU leases is over—AI labs are now signing five-year, take-or-pay contracts for access to the latest NVIDIA architectures like Blackwell and Vera Rubin, signaling a fundamental shift toward long-term, capital-intensive infrastructure planning. And while commoditization of compute seems inevitable in theory, McBee argues it’s unlikely in practice: performance varies wildly across identical GPUs due to software, cooling, and operational efficiency, making true fungibility nearly impossible—especially at scale.