Back to stories
Industry

Blackstone Commits $5B to Google TPU Cloud Joint Venture, Targeting 500 MW Online in 2027

Michael Ouroumis2 min read
Blackstone Commits $5B to Google TPU Cloud Joint Venture, Targeting 500 MW Online in 2027

Blackstone and Google announced a U.S.-based joint venture on Tuesday to stand up a new TPU cloud, with Blackstone making an initial $5 billion equity commitment and the partners targeting 500 MW of data center capacity online in 2027. The venture will sell Google Cloud's Tensor Processing Units as a compute-as-a-service offering, packaged with data center capacity, operations, and networking.

A second front door to TPUs

Google will supply the hardware — including TPUs — along with software and services to the new company, which will operate independently of Google Cloud's direct sales channel. The pitch is straightforward: enterprises, AI labs, and frontier developers that want access to TPUs without buying directly from Google Cloud now have a balance-sheet-backed alternative built around the same silicon. Google Cloud CEO Thomas Kurian said the joint venture "helps meet growing demand for TPUs, which are optimized specifically for efficiency."

Benjamin Treynor Sloss, a 20-plus-year Google infrastructure veteran, will lead the new company as CEO — a signal that operational discipline, not just capital deployment, is the bet. Sloss spent years running Google's site reliability engineering and data center operations, the muscle the venture will need to bring 500 MW into service on schedule.

Blackstone's AI infrastructure stack

The deal extends a pattern Blackstone has been accelerating in 2026. The firm, which manages more than $1.3 trillion in assets, said earlier this month it had stood up a similar vehicle with Anthropic. Jon Gray, Blackstone's president and COO, framed the rationale plainly: "We see a generational opportunity to invest capital at scale building AI infrastructure." Jas Khaira, who runs Blackstone's N1 AI investing platform, called TPUs "a decade in the making and foundational to the AI economy."

By fronting equity at the data-center-plus-silicon layer rather than at the model layer, Blackstone is positioning itself as the financial counterparty for capacity that hyperscalers themselves are increasingly capital-constrained to build alone.

What changes for builders and buyers

For practitioners running production workloads, the headline is supply: a new 500 MW TPU pool — not adjacent to Google Cloud's existing fleet but additive to it — is being financed and built explicitly to be sold as third-party compute. That matters for teams whose training runs are gated on TPU pod availability, for inference shops looking for an alternative to GPU-bound providers, and for procurement leaders who want a non-Google counterparty on a TPU contract.

Two open questions will determine how much this actually loosens the squeeze. First, pricing and contract structure relative to direct Google Cloud TPU SKUs — the venture has not disclosed rate cards or commitment terms. Second, the equity split and governance between Blackstone and Google, which were not detailed in the announcement and will shape whether the company can credibly sell against its own supplier. Both will matter more than the headline $5 billion once 2027 arrives.

Learn AI for Free — FreeAcademy.ai

Take "AI for Business: Practical Implementation" — a free course with certificate to master the skills behind this story.

More in Industry

Google Ships Antigravity 2.0: A Standalone Agent Platform That Retires the Gemini CLI
Industry

Google Ships Antigravity 2.0: A Standalone Agent Platform That Retires the Gemini CLI

At I/O 2026, Google relaunched Antigravity as a standalone agent-orchestration platform with a CLI that replaces the Gemini CLI, an SDK, Managed Agents in the Gemini API, and an enterprise tier — all defaulting to Gemini 3.5 Flash.

10 min ago2 min read
Alibaba's Zhenwu M890 Claims 3x Nvidia's H20, Ships With Qwen 3.7-Max
Industry

Alibaba's Zhenwu M890 Claims 3x Nvidia's H20, Ships With Qwen 3.7-Max

Alibaba's chip unit T-Head unveiled the Zhenwu M890 AI accelerator, claiming 3x the agentic-inference performance of Nvidia's H20 and 144GB of HBM3, paired with the Qwen 3.7-Max model — a full domestic stack as US export controls tighten.

1 hours ago2 min read
Cohere Buys Reliant AI to Build 'North for Pharma' Sovereign Biopharma Agents
Industry

Cohere Buys Reliant AI to Build 'North for Pharma' Sovereign Biopharma Agents

Cohere has acquired Montréal- and Berlin-based biopharma startup Reliant AI, folding its automated research workbench into the North platform to launch 'North for Pharma' — its second acquisition in a month as it doubles down on sovereign, regulated-industry AI.

3 hours ago2 min read