Blackstone and Google Launch TPU‑Powered Cloud JV

Blackstone announced a joint venture with Google to build a U.S.-based cloud service that will sell access to Google’s Tensor Processing Units (TPUs) and associated data-center capacity to institutional and enterprise customers.

Under the plan Blackstone will make an initial $5 billion equity commitment and the new company expects to bring about 500 megawatts of AI compute capacity online in 2027, with an intention to scale further over time.

Google will supply hardware — including TPUs — plus software and services, while Blackstone provides capital, data-center operations, and networking expertise to run the service as a compute-as-a-service option outside of Google Cloud’s core offering.

Blackstone named Benjamin Treynor Sloss, a long-time Google infrastructure executive, as CEO of the new company, underscoring that the venture will lean on Google technical talent as it spins up capacity and operations.

The announcement frames the venture as a new channel for TPUs — Google’s custom AI accelerators that power Gemini and other large-scale workloads — giving enterprises another way to run TPU-optimized training and inference outside traditional public-cloud contracts.

Observers say the deal is part of a broader wave of private-equity and asset-manager partnerships aimed at accelerating enterprise adoption of AI by underwriting huge, capital-intensive data-center builds. Recent joint ventures involving Anthropic and OpenAI with private-equity partners illustrate that same push to pair model makers with outside capital.

Industry outlets note the deal fits a pattern in 2026: cloud and AI players are seeking third-party financing and specialized partners to expand capacity quickly without shouldering all the balance-sheet risk. That dynamic is driving “neocloud” plays and bespoke infrastructure firms focused on AI workloads.

For enterprise customers, the venture promises choice and potentially lower latency or specialized pricing for TPU workloads, because TPUs are engineered specifically for many AI training and inference tasks and can be more efficient than general-purpose GPUs for some model types. The companies say the offering will give customers more options beyond running TPUs directly inside Google Cloud.

The announcement leaves several commercial details open. Blackstone’s press release specifies the initial capital and capacity target, but it does not publish pricing, detailed customer contracts, or the exact ownership split of the new company in public materials. Independent coverage suggests Blackstone will be a controlling investor via its equity commitment, but reporting on final governance and leverage is still emerging.

Analysts say advantages for the JV include access to Google’s TPU roadmap and software stack while using Blackstone’s scale to site and operate facilities where energy and networking economics are favorable. Critics caution that the model depends on long-term demand for TPU-specific services and on smooth commercial terms between Google, the JV, and enterprise customers.

The broader market reaction will hinge on how quickly the new company can sign anchor customers and whether enterprises prefer third-party TPU capacity versus the big three public clouds’ GPU offerings. The move also adds another supplier to a market already seeing large private-equity-backed deployment plays aimed at enterprise AI adoption.

Timeline and oversight are concrete early markers to watch: Blackstone expects initial capacity in 2027, and Google says it will supply the chips and software. Regulators and large customers will likely scrutinize contractual terms, data residency provisions, and interoperability with existing cloud stacks as the JV moves from announcement to operations.