Google, Blackstone plan new AI cloud company- WSJ

Google and Blackstone will launch a new AI cloud venture, with Blackstone investing $5 billion and taking majority ownership. The company aims to bring 500 megawatts of computing capacity online by 2027 using Google’s Tensor Processing Units, software, and services, potentially intensifying competition with Nvidia and CoreWeave. The announcement is supportive for Google’s AI commercialization efforts and signals continued capital inflows into AI infrastructure.

Analysis

This is less about one project and more about a potential re-pricing of the AI infra stack. A hyperscale-adjacent buyer validating non-NVIDIA silicon at scale weakens the “all roads lead to NVDA” narrative and creates a second sourcing channel for compute buyers that care more about capacity availability and unit economics than absolute peak performance. The immediate beneficiary is GOOGL: TPU utilization improves, software lock-in deepens, and the company moves one step closer to monetizing chips as an ecosystem rather than a captive internal tool.

BX is interesting because this looks like a private-markets style infrastructure play with contracted demand and long-duration optionality, not a normal corporate venture bet. If execution is credible, it can become a template for financing other power-hungry digital infrastructure assets, which should widen BX’s addressable opportunity set beyond traditional buyout capital. The second-order effect is that cloud/compute capacity increasingly behaves like a power-and-real-estate constrained asset class, which favors sponsors with balance sheet access and operating leverage.

CRWV is the clearest near-term loser because the market may start discounting a tougher funding and customer-acquisition environment if larger incumbents and sponsors can replicate its value proposition with better procurement terms. For NVDA, the risk is not volume collapse but multiple compression: if TPU-based capacity becomes a credible alternative, the market may assign a lower scarcity premium to every incremental AI server deployment. The key timeframe is 6-18 months, when initial capacity announcements start translating into purchase orders, developer migration, and benchmark evidence.

The contrarian read is that this is bullish for the AI buildout overall, but bearish for the most crowded expressions of that trade. Consensus still treats compute demand as uniform, yet buyers are increasingly bifurcating into “best performance” and “good-enough, cheaper, available now,” which is exactly the wedge that can cap pricing power at the margin. If power availability remains the bottleneck, the winners will be the firms that can finance megawatts, not just design chips.

Key Decisions for Investors

• Long GOOGL vs. short NVDA into the next 1-2 quarters: thesis is modest multiple expansion for GOOGL as TPU monetization becomes more visible, while NVDA faces scarcity-premium compression; target 10-15% relative outperformance if alternative silicon adoption broadens.
• Buy BX on pullbacks for a 6-12 month hold: the venture structure creates an infrastructure/credit-like stream with embedded optionality; risk/reward improves if the market starts valuing BX as an AI-capital allocator rather than just a traditional PE manager.
• Short CRWV or use put spreads for 3-6 months: increased competitive pressure from scaled incumbents should weigh on customer growth and pricing; best expression is a defined-risk bearish spread because the name can remain momentum-driven near term.
• Pair trade long infrastructure enablers vs. short pure-play AI compute beta: e.g., long GOOGL/BX, short a basket of the highest-multiple AI compute beneficiaries, to isolate the shift from scarcity premium toward capital access premium.
• For NVDA holders, trim upside exposure via covered calls over the next earnings cycle: the structural story remains intact, but this headline increases the odds of near-term multiple mean reversion before fundamentals catch down.