Blackstone bets $5 Billion on Google’s AI chips in joint venture that skips the cloud

Blackstone and Google are building something that does not fit neatly into either company’s existing playbook. The two announced a joint venture this week to create a U.S.-based company that will sell access to Google’s Tensor Processing Units directly to customers, outside of the traditional Google Cloud platform.

The structure is worth paying attention to. Rather than simply expanding Google Cloud’s capacity, the deal creates an entirely separate company with its own leadership. Blackstone is committing an initial $5 billion in equity capital, with the first 500 megawatts of data center capacity expected to come online in 2027. The ambition, at least on paper, is to scale well beyond that.

Google’s TPUs are the engine behind the arrangement. The chips have been in active production use for over a decade and currently power Gemini along with workloads for some of the world’s largest AI research labs and financial institutions. Google will supply the hardware, software, and technical support to the new company, while Blackstone brings the real estate, energy infrastructure, and capital.

Benjamin Treynor Sloss, a Google veteran with more than 20 years building and running the company’s global infrastructure, will serve as CEO of the new entity. That appointment signals this is not a passive investment vehicle.

What makes this arrangement different from a typical hyperscaler expansion is the intent to give customers a separate path to TPU access. Organizations that prefer not to work through Google Cloud directly now have another option, and that kind of optionality tends to matter in enterprise procurement.

Blackstone president Jon Gray described it as a generational investment opportunity in AI infrastructure. The framing is familiar at this point, but the dollar figures and the involvement of Google’s proprietary silicon give it more weight than most announcements in this space.

The broader context is straightforward: demand for AI compute keeps outpacing supply, and conventional cloud infrastructure is struggling to absorb it fast enough. This joint venture is one attempt to close that gap by pairing financial scale with hardware that was built specifically for the task.