Google, Blackstone Launch $25B TPU Cloud Venture
Key insights
- Blackstone's $5B equity stake makes this one of the largest private-equity bets on AI infrastructure demand to date.
- The venture targets CoreWeave's dedicated-compute market directly, with 500 MW of TPU capacity planned by 2027.
- Google offloads capital risk to Blackstone while retaining control over TPU supply, design, and technology licensing.
Why this matters
Dedicated AI compute is splitting off from general-purpose cloud as a separate market with its own financing structure, and a $25 billion joint venture between Google and Blackstone is the clearest signal yet that hyperscalers are adapting their go-to-market rather than ceding ground to CoreWeave. For AI infrastructure buyers, this creates a credible TPU-native alternative to NVIDIA GPU clusters at scale, which materially affects hardware procurement decisions starting in 2027. For founders and technical leaders building on cloud compute, the emergence of Blackstone-style PE-backed dedicated compute as a third category alongside hyperscalers and GPU clouds will reshape pricing leverage and contract terms across the board.
Summary
Google and Blackstone are standing up a dedicated AI cloud company together, with Blackstone taking majority ownership through a $5 billion equity injection and the full deal valued at roughly $25 billion including leverage.
The venture gives enterprise customers direct, dedicated access to Google's custom TPUs rather than routing them through Google Cloud's shared infrastructure. First capacity, targeting 500 megawatts, is slated to come online in 2027. The structure mirrors CoreWeave's model closely: private-equity-backed, infrastructure-heavy, selling guaranteed compute access rather than on-demand cloud slots.
Essentially: (Google, Blackstone) are building a CoreWeave competitor using TPUs as the hardware wedge.
- Blackstone becomes majority owner, making this one of the largest private-equity commitments to AI infrastructure on record.
- The 500 MW initial target positions the venture as a hyperscale player from day one, not a boutique compute provider.
- Google retains TPU supply and design control while offloading capital expenditure and balance-sheet risk to Blackstone.
The deal signals that dedicated AI compute is now a distinct asset class, not just a feature of general-purpose cloud platforms.
Potential risks and opportunities
Risks
- CoreWeave, which filed for IPO earlier in 2026 on the strength of NVIDIA GPU dominance, could face valuation pressure if Google-Blackstone TPU capacity comes online at competitive pricing in 2027.
- Blackstone limited partners are now heavily exposed to AI infrastructure demand holding; a slowdown in enterprise AI capex between now and 2027 capacity delivery would leave the venture with stranded assets and leverage against declining revenue.
- Google risks channel conflict with existing Google Cloud customers who may delay or restructure Cloud commitments anticipating cheaper dedicated TPU access through the new entity.
Opportunities
- Data center operators and power infrastructure firms (Equinix, Digital Bridge, NRG Energy) are well-positioned to supply the 500 MW capacity the venture needs before 2027.
- MLOps and AI orchestration vendors (Weights & Biases, Anyscale, Modal) that add TPU support now gain first-mover advantage as the venture onboards enterprise customers seeking software tooling.
- Sovereign wealth funds and other institutional LPs looking for AI infrastructure exposure now have a Blackstone-structured vehicle as a template to demand similar co-investment terms from other hyperscalers.
What we don't know yet
- Whether enterprise customers will be locked into long-term capacity contracts similar to CoreWeave's take-or-pay structure, and on what terms.
- How Google will price TPU access through the venture relative to existing Google Cloud TPU pricing, and whether existing Cloud customers face any cannibalization.
- Which workloads the venture will target first in 2027 given that TPUs historically underperform on non-transformer inference tasks compared to NVIDIA GPUs.
Originally reported by bloomberg.com
Read the original article →Original headline: Google and Blackstone Form $25 Billion AI Cloud Joint Venture to Compete With CoreWeave Using Custom TPUs