Google-Blackstone $5B AI data center venture

Google and Blackstone said this week they are creating a new U.S.-based company to sell access to Google Cloud’s Tensor Processing Units through a dedicated “TPU cloud” structure. Blackstone said it will make an initial $5 billion equity commitment, and the partners said they expect 500 megawatts of capacity to come online in 2027. Google said it will provide the TPUs, software and services to the venture, while Blackstone will provide capital for the underlying data center buildout. ### What exactly did Google and Blackstone announce? Blackstone said on May 18 it had agreed a joint venture with Google to create a new company that will offer data center capacity, operations, networking and Google Cloud TPUs as a compute-as-a-service product. Google described the move on May 19 as a way to give customers “more choice and flexibility” in how they access cloud TPUs. (blackstone.com) The new company is U.S.-based, according to Blackstone’s release. The firms said the platform is designed around TPU capacity rather than a general-purpose colocation or wholesale data center model. ### Where does the $5 billion go? Blackstone said the $5 billion is an initial equity commitment tied to bringing the first 500 megawatts of capacity online in 2027. (blackstone.com) The company also said the platform is intended to “scale significantly over time,” though neither side disclosed a total long-term capital target, customer list or site locations in the initial announcement. Google said Blackstone will create the TPU cloud, while Google will supply the chips, software and services. That splits the venture into two clear roles: Blackstone finances and develops the infrastructure, and Google provides the computing stack customers will use. ### Why are TPUs the focus instead of generic AI capacity? (blackstone.com) Google introduced new eighth-generation TPU products at its Cloud Next conference in April, including TPU 8t for training and TPU 8i for inference. The company said those chips were built for more demanding AI workloads, including agentic systems, and positioned them as part of a broader push to expand AI infrastructure on Google Cloud. (blog.google) The Blackstone venture extends that push by creating a separate channel for TPU supply. In its announcement, Google said the arrangement is meant to broaden access to TPUs beyond Google’s existing cloud delivery model. ### How does this fit into Blackstone’s existing AI infrastructure strategy? (blog.google) Blackstone has been building out data center and power exposure through its infrastructure investing, and on its website it says AI demand is increasing the need for data centers, energy, hardware, software and related services. The firm says it has built positions in digital infrastructure through investments including QTS and AirTrunk. (blog.google) The Google joint venture adds a chip-specific AI cloud vehicle to that broader strategy. Blackstone’s press release framed the TPU cloud as an infrastructure business built around efficient capacity, operations and networking for AI workloads. ### What comes next, and what is still missing? (blackstone.com) The clearest next milestone is 2027, when the partners say the first 500 megawatts of TPU capacity are expected to be online. Google’s blog says it will provide TPUs, software and services to the new company, while Blackstone’s release says the platform will be based in the United States. (blackstone.com) The announcement did not identify customers, facility sites, power procurement arrangements or pricing. Those details will determine how quickly the venture converts Blackstone’s $5 billion commitment into contracted TPU capacity and whether the platform expands beyond its initial 500-megawatt target. (blackstone.com)