Google opens TPUs to enterprises beyond its own cloud via Blackstone JV
Google Cloud and Blackstone have unveiled a new joint venture aimed at building a large-scale standalone cloud platform powered by Google’s Tensor Processing Units (TPUs), marking one of the company’s clearest moves yet to expand its AI infrastructure beyond the traditional boundaries of Google Cloud. The new company will offer “efficient data center capacity, operations, networking, and Google Cloud’s Tensor Processing Units (TPUs) as a compute-as-a-service offering,” Blackstone said in a statement . Under the agreement, Blackstone will commit an initial $5 billion in equity funding to the venture, with Google supplying hardware, software, and services. The companies said the new platform will provide enterprises “another option to access cloud TPUs in addition to using them through Google Cloud,” signaling a broader shift in how Google plans to commercialize its proprietary AI chips. The project is expected to deliver roughly 500 megawatts of data center capacity by 2027, the statement added. The deal signals AI infrastructure is beginning to separate from the traditional hyperscaler cloud bundle and become its own economic layer, with accelerator access, power and data centre capacity now behaving like infrastructure constraints rather than software ones, said Sanchit Vir Gogia, chief analyst at Greyhound Research. “Google is not giving up control. It is changing the wrapper,” Gogia said. Google expands TPU distribution strategy Google’s TPUs have historically been tightly linked to Google Cloud services, giving enterprises access to the company’s custom AI accelerators primarily through its own hyperscale cloud platform. The new venture creates a separate distribution channel for TPU-based infrastructure outside Google Cloud’s traditional consumption model — a notable shift as enterprises increasingly seek alternatives to Nvidia-dominated AI infrastructure and reassess long-term AI infrastructure sourcing strategies. “The frontier of AI is shifting from models that answer to agents that act,” the companies said in the announcement. That transition is driving growing demand for infrastructure capable of supporting autonomous AI systems, enterprise copilots, and agentic AI workloads that require large-scale inference capacity and lower operating costs. Gogia said the development reflects a broader transition in enterprise technology procurement, where organizations increasingly evaluate AI infrastructure separately from cloud platforms themselves. “CIOs will increasingly have to buy AI as a portfolio of capacity, not as a feature of cloud,” Gogia said. He added that enterprises are increasingly making separate decisions around compute sourcing, accelerator access, orchestration tooling, governance, and infrastructure placement rather than treating AI purely as another cloud service layer. The venture positions Google more directly against a growing class of AI-focused “neocloud” providers such as CoreWeave , Lambda , and Crusoe , which have largely built their businesses around Nvidia GPU infrastructure. While the move could increase enterprise access to TPU infrastructure, Gogia cautioned against viewing it as an immediate replacement for Nvidia-based AI environments. “The real shift is from single-stack dependence to infrastructure portfolio management,” he said. Inference economics emerges as an enterprise priority The announcement also reflects how enterprise AI spending is increasingly shifting from model experimentation toward long-term inference economics as organizations move AI workloads into production environments. Gogia said many enterprises remain overly focused on foundation model benchmarks while underestimating the operational cost implications of sustaining large-scale AI deployments. “Training makes headlines. Inference makes invoices,” Gogia said. As enterprises deploy AI copilots, autonomous agents, and workflow automation systems, inference workloads are becoming continuous operational processes rather than isolated AI experiments, increasing pressure on compute availability, infrastructure efficiency, and long-term operating costs. The companies said the platform is designed to support both AI training and inference workloads, areas where infrastructure demand has intensified amid ongoing GPU shortages and escalating AI deployment costs. Private equity deepens role in AI infrastructure The partnership also highlights the expanding role of private equity firms in financing the AI infrastructure boom as hyperscalers and AI companies race to secure compute capacity, data center power, and AI chip supply chains. Blackstone has become one of the largest investors in AI-related infrastructure, including data centers, cloud platforms, and energy assets tied to AI expansion. The company said the TPU venture would combine Google’s AI technology with Blackstone’s infrastructure development and financing capabilities. Jon Gray, president and chief operating officer at Blackstone, said demand for AI infrastructure continues to accelerate globally. “We believe AI will drive one of the largest infrastructure buildouts in history,” Gray said in the statement. Gogia noted that the rise of private equity-backed AI infrastructure platforms reflects a broader shift in how the industry increasingly views AI compute. “The bottleneck is not merely the chip. It is the powered, cooled, connected, financeable site,” he said. For enterprise IT leaders, the development may signal a more diversified AI infrastructure market where organizations increasingly evaluate not only model performance but also compute availability, infrastructure resilience, long-term capacity commitments, and AI supply-chain flexibility as AI deployments scale.