Chasing a $400 Billion Prize: Google's New AI Cloud Company Is a Threat to Nvidia and CoreWeave

Alphabet's new AI cloud joint venture with Blackstone is targeting a market estimated at $400 billion by 2031 and could eventually pressure Nvidia and CoreWeave. The venture plans 500 megawatts of capacity by 2027 and is expected to rely heavily on Google's TPUs, potentially reducing Alphabet's reliance on Nvidia's chips. Analysts warn the move could compress CoreWeave's pricing power and margins on new contracts, though the competitive threat is still early-stage.

Analysis

This is less about an imminent revenue hit to NVDA and more about the probability-weighted erosion of its pricing power over the next 12-36 months. The strategic shift is that hyperscalers are now trying to internalize the most expensive layer of AI infrastructure economics: custom silicon plus captive cloud capacity. Even if Google’s venture captures only a modest share of incremental AI spend, the marginal buyer behavior changes first in the highest-margin training workloads, then in inference, which is where unit economics matter most and where substitution away from NVDA can compound quietly.

The bigger second-order effect is on the supply chain. If Google pushes TPU-based capacity, the bottleneck moves away from GPU availability toward networking, power delivery, liquid cooling, and data-center buildout execution. That tends to benefit infrastructure enablers while pressuring pure-play AI compute resellers whose business model depends on scarce GPU access and wide spreads between contracted pricing and replacement cost. In other words, the market may be underestimating how quickly “AI cloud” can become a commoditized layer once the largest buyers decide to verticalize.

CRWV is the cleaner near-term short than NVDA because its valuation is much more sensitive to pricing and utilization assumptions, and it carries more balance-sheet fragility if new capacity comes online into a less favorable contract environment. The risk to the short is that demand remains so strong that supply additions are absorbed faster than expected; that would push out any margin compression story by several quarters. For NVDA, the debate is slower and more structural: investors are likely fine until evidence accumulates that custom silicon starts taking a larger share of incremental capex, which is a 2026-2027 story rather than a near-term earnings issue.

Key Decisions for Investors

• Short CRWV into strength over the next 2-6 weeks; the asymmetric risk is multiple compression if new Google TPU capacity forces a re-rate in forward margins and contract pricing.
• Reduce or hedge NVDA exposure with 6-12 month downside protection rather than outright selling; the core thesis is slower share loss in incremental capex, not an immediate demand shock.
• Pair trade: long GOOGL / short CRWV for 3-9 months, expressing the view that Google can defensively monetize its own stack while pressuring third-party AI cloud economics.
• Prefer infrastructure beneficiaries over AI cloud resellers: consider long INTC only as a relative-value hedge if the market starts rewarding custom silicon optionality, but size small given execution risk.
• If holding NVDA, own downside via put spreads around next two earnings cycles; the catalyst is not this quarter but evidence of TPU adoption in 2026 capex budgets.