Google and Blackstone launch $25B AI infrastructure joint venture

The Google–Blackstone JV pairs the world's largest alternative-asset manager with Google's TPU and data-center expertise to underwrite AI compute capacity at scale. Blackstone takes the majority equity position with a $5B initial commitment, and the headline $25B figure reflects total committed value including debt financing and asset contributions. The structure mirrors the Stargate-style approach OpenAI/Oracle/SoftBank have pioneered, but applies it to Google's existing accelerator stack rather than Nvidia's.

Mechanically, the JV will operate as a compute-as-a-service provider — selling capacity to enterprise customers (including, by implication, third-party AI labs) under long-term contracts that anchor the buildout. That structure lets Google monetize TPU capacity at a scale beyond its own internal needs while transferring construction and operating risk to Blackstone's infrastructure-finance machine.

Competitive context: this is the biggest counter yet to the Microsoft/OpenAI/Stargate axis and the Anthropic/AWS multi-billion compute deal disclosed earlier this year. It also lands the same week as Armada's $230M raise at a $2B valuation for portable AI data centers and Zyphra raising $500M to train models on AMD silicon — capital is flooding into anything that disintermediates Nvidia or unlocks new compute supply.

Skeptical takes from the wires focus on demand realism: $25B presumes a sustained scarcity of frontier compute through 2028, and Google's recent shift to consumption-based pricing across the Gemini product line suggests management believes the demand curve is intact. The JV also has political implications — TPU-based compute capacity offered to Anthropic, Meta, and AI startups would meaningfully reshape the Nvidia-anchored AI infrastructure map.