Blackstone and Google on May 18 announced a joint venture to create a new U.S.-based company that will offer data center capacity, operations, networking and Google Cloud TPUs as a compute-as-a-service. Blackstone said funds it manages will make an initial $5 billion equity commitment, a figure Google confirmed in its own blog post. The venture expects its first 500 megawatts of capacity to come online in 2027, the companies said, signaling that AI infrastructure is being turned into a financed, standalone platform business rather than remaining solely a hyperscaler service.
The structure pairs the world’s largest alternative asset manager—which already controls the QTS and AirTrunk data-center platforms and held more than $1.3 trillion in assets under management at the end of March—with Google’s custom-designed AI chips. Google will supply the hardware, software and services, while Blackstone-managed funds provide the initial equity. By creating a commercial channel outside Google Cloud, the venture gives enterprise and startup customers another way to access TPUs, chips that have primarily been available through Google’s own infrastructure or through large, bespoke agreements. In April, Anthropic announced that it would receive multiple gigawatts of next-generation TPU capacity starting in 2027, a deal with Google and Broadcom that underscored the chips’ appeal to external builders and demonstrated that large AI developers are willing to commit to TPU capacity at scale.
Benjamin Treynor Sloss, a Google vice president with more than two decades of experience in site reliability and infrastructure, has been appointed chief executive of the new venture. The appointment signals deep technical continuity with Google’s operations, even as the company stands apart from the cloud unit.
The 500-megawatt target represents a bet that AI developers will seek alternatives to Nvidia’s dominant GPU stack, but the venture’s first megawatts are not due until 2027, making it a forward-looking play in a market where Nvidia’s current-generation chips already command the bulk of AI training workloads. The new service enters a landscape filled with Nvidia-backed neoclouds such as CoreWeave. Analysts see the venture as a potential challenger, though any competitive advantage will depend on pricing and adoption—details that have not been made public. The companies described the offering as adding another option to the existing Google Cloud TPU service, not replacing it, and avoided drawing direct comparisons with GPU pricing.
Bloomberg and The Wall Street Journal reported, citing people familiar with the matter, that the total investment value could reach about $25 billion when leverage is included, and that Blackstone is expected to hold a majority stake. Blackstone and Google did not confirm those figures. The legal name of the new company remains undisclosed; internal materials reference “BXN1,” but that has not been confirmed as the official name. The companies also did not disclose data center locations, power sources, or whether the 500 megawatts will include accelerators other than TPUs. No customer agreements or startup-specific pricing have been revealed.
The joint venture arrives amid a wider surge in AI infrastructure investment. Reuters has reported that Big Tech AI outlays are expected to top $700 billion in 2026. Blackstone’s expanding data-center footprint through platforms like QTS and AirTrunk positions it to capitalize on that demand, while Google gains a capital-light path to scale TPU adoption. For AI developers, the promise is an additional source of compute—though its cost and availability will not be tested until the first capacity comes online next year.