Google’s TPUs Get a Blackstone-Funded Cloud of Their Own

Google has spent years keeping its custom AI chips, the Tensor Processing Units, mostly to itself, rentable only through Google Cloud. The TPUs are Google’s own alternative to the Nvidia processors most of the AI industry rents to train and run its models. That is ending. In a joint venture announced on May 18, 2026, Google and Blackstone set out to create a separate company that builds data centers and sells TPU capacity directly to outside customers. Six weeks on, the clearest sign of how serious it is has arrived: the new business, still publicly unnamed, has started hiring the executives who will run it. For the wider hosting and cloud market, the more telling part is the shape of the deal. It puts a new merchant supplier of AI compute into the market, funded not from Google’s balance sheet but from private equity, and the way it is financed may matter as much as the chips it sells.

A Separate Company, Majority-Owned by Blackstone

Blackstone is committing an initial $5 billion in equity and taking majority ownership, while bringing the scale of what it calls the world’s largest data-center provider. Google supplies the TPU hardware, software and services. The equity is only the seed: layered with debt secured against the underlying assets, reporting puts the full build at around $25 billion. The new U.S. company will be led by Benjamin Treynor Sloss, the Google veteran who founded its Site Reliability Engineering practice and most recently ran TPU commercialization as chief programs officer. The first target is 500 megawatts of capacity online in 2027, with plans to scale significantly from there.

A Merchant Channel for Google’s Silicon

The point of a standalone company is to open a sales channel Google Cloud cannot. Today, renting a TPU means renting it from Google Cloud. The venture turns TPUs into capacity sold on their own terms, available to any buyer rather than only to Google Cloud customers. The Stack reports the target list runs from enterprises to hyperscalers, frontier AI labs and sovereign customers, buyers who want Google’s silicon without committing to the rest of Google’s cloud. It lets Google meet external demand without cannibalizing its own cloud capacity during a period of chip and power constraints, and it drops a credible non-Nvidia option into a compute-rental market that Nvidia GPUs have dominated. Google is opening its chips up in parallel, with TPU hardware sales to third parties expected to ramp through 2027.

Private Capital Is Funding the AI Buildout

This is where the deal gets its wider significance. Hyperscale AI infrastructure has grown too capital-hungry to fund from operating cash alone, and the answer increasingly is outside money. Blackstone’s equity is the seed and the debt does the rest, secured on infrastructure-grade assets. Google expands its chip footprint without carrying all the cost, and Blackstone gets a long-dated AI-compute asset of a kind few private investors can build directly. For a hosting and cloud industry watching the same financing playbook spread, it marks how the next wave of capacity is being paid for: off the hyperscaler’s balance sheet, onto a private-equity one.

Hiring a Cloud Into Existence

That is what makes the late-June job postings the real news. A chief operating officer role went up on June 23, with a chief technical officer brief following days later, spanning site reliability, hardware operations and the software platform. These are the hires of a company being assembled from nothing against a hard deadline, the first 500 megawatts due in 2027. In May this was a press release and a capital commitment. It is now becoming an operating business with a leadership team, built for a single purpose: selling Google’s chips to every AI buyer Google Cloud does not reach. Google’s TPUs are getting their own company, their own Blackstone-backed balance sheet, and now their own management.