Google Just Shared Bad News for Nvidia, and Even Worse News For CoreWeave and Nebius

Alphabet is expanding its custom AI accelerator push by selling TPUs to third parties and launching a Blackstone-backed neocloud joint venture with $5 billion of capital support and an expected $25 billion of total spending power. The venture plans to deploy 500 MW of TPU capacity next year, which could strengthen Google’s TPU ecosystem but intensify pricing and competitive pressure on Nvidia, CoreWeave, and Nebius. Overall, the article is constructive for Alphabet’s long-term AI strategy but introduces meaningful competitive risk for neocloud peers.

Analysis

The immediate market read is not that Alphabet becomes a better cloud vendor; it becomes a more credible industrial-scale buyer of constrained AI infrastructure inputs. That matters because the scarcity regime is shifting from GPUs alone to a bundle of power, land, interconnect, and advanced packaging capacity, and a vertically integrated TPU stack gives Google more control over unit economics than a rent-a-GPU model. The second-order winner is Blackstone: it is effectively underwriting a capital-intensive AI real estate/energy platform with a strategic anchor tenant and a differentiated compute asset, which should improve financing terms for future neocloud-style ventures.

For Nvidia, the real risk is not immediate revenue loss but pricing power compression in the next 12-24 months if TPU adoption crosses from experimentation into procurement normalization. The threat is most acute in inference-heavy workloads where customers care more about cost per token and supply assurance than about absolute software lock-in; if that mix shift happens, Nvidia’s moat narrows at the margin even if absolute demand remains strong. The supply-chain implication is also non-obvious: a larger TPU order book improves Alphabet’s bargaining position at TSMC, which can deepen the capacity gap versus smaller AI compute operators that cannot pre-commit enough volume to secure priority.

CoreWeave and Nebius are not being killed by one announcement, but their equity cases become more fragile because their valuations depend on future refinancing and contract renewals staying benign. A well-capitalized entrant with owned silicon and lower cost of capital can force a downward reset in forward pricing just as these firms are still levering up to build capacity, which is the wrong part of the cycle to face a price war. The consensus is probably underestimating how quickly enterprise buyers will bifurcate between premium training clusters and cheaper inference fleets, which means the first casualties may be margins, not utilization.

The contrarian angle is that this may be more bullish for Google’s ecosystem than bearish for Nvidia near term: the JV can expand the addressable market for AI compute by making capacity cheaper and easier to finance. If that happens, there is enough demand elasticity to keep the whole stack growing, but the surplus accrues to the vertically integrated owners rather than the merchant providers. The key test is not launch timing; it is whether the JV can lock in multi-year customer commitments fast enough to prove that TPUs are a credible procurement alternative before the next wave of capacity comes online.

Key Decisions for Investors

• Long GOOGL / short CRWV as a 6-12 month relative-value trade: buy the vertically integrated capacity owner against the most levered neocloud incumbent; thesis works if TPU adoption and JV pricing pressure compress CRWV’s forward multiple faster than Alphabet’s capex is discounted.
• Long BX against a basket short of capital-intensive AI infrastructure names (CRWV, NBIS) over 3-9 months: Blackstone monetizes capital formation and real assets while the operating risk sits with the compute provider.
• Maintain or add to NVDA only on weakness, but hedge with downside calls or a GOOGL call spread: the near-term setup is not a demand collapse but a gradual share/mix leak in inference workloads over the next 12-24 months.
• Avoid chasing NBIS on capacity headlines for now; wait 1-2 quarters for evidence of pricing resilience before adding. The asymmetric risk is that the market is valuing capacity buildout as a moat when it may be a balance-sheet race to the bottom.
• For higher-risk expression, buy 6-12 month GOOGL call spreads financed by selling out-of-the-money NVDA calls if position limits allow; this captures the optionality of TPU commercialization while partially funding the hedge against broader AI capex disappointment.