The discussion
The hosts hold a broadly mixed-to-constructive view on CoreWeave, with the debate centering on two key risks: GPU useful life and competitive displacement by hyperscalers. Friedberg is the clearest bear, warning that hyperscalers deploying $80B+ in annual CapEx could render CoreWeave's offering redundant — analogous to transitory arbitrage businesses that disappeared when infrastructure caught up. Chamath has evolved over time: he initially flagged the GPU useful-life assumption as an existential variable that could leave the business "deeply underwater," later warmed to the bull case alongside guest Gavin Baker (who argues the operational complexity of running large GPU clusters is underappreciated and may constitute a real moat), but more recently identified structural pricing friction — over-reliance on guaranteed contracts and unworkable spot pricing — as a limitation for enterprise inference at scale. The bear thesis on useful life was substantially addressed by May 2026, when Baker cited Nvidia's earnings as evidence that GPUs have a 10–15 year useful life, enabling long-duration asset-backed financing and a far more profitable model than critics projected, a development that shifted the balance of the debate toward the bulls.