Google and Blackstone to Create TPU Cloud Service
Alphabet’s Google has entered a joint venture with asset management company Blackstone to deliver services via a new U.S.-based company equipped with Google's Tensor Processing Units (TPUs).
The news highlights Google’s strategy to begin offering its chips separately from Google Cloud, though the new company apparently will offer both alternatives to customers. It also adds to the momentum behind challenges to NVIDIA’s dominance in the AI accelerator chip market—a force also evident in Cerebras’s recent IPO.
Google Cloud CEO Thomas Kurian stated in the press release:
“This joint venture with Blackstone helps meet growing demand for TPUs, which are optimized specifically for efficiency and performance in the AI era. Together, we’re accelerating AI transformation and providing more options for organizations to access accelerated compute capability.”
A New TPU Neocloud
The new company will be majority owned by Blackstone, which is initially investing $5 billion to bring 500 MW of capacity for it online in 2027, with plans to grow beyond that “significantly over time,” according to the press release.
The year 2027 has also been cited by Alphabet management as the start of hardware sales to third parties outside of Google Cloud. Revenues from those sales should materialize significantly in 2027, Alphabet said.
On Alphabet’s Q1 earnings call April 29th, Alphabet CEO Sundar Pichai stated:
“As TPU demand grows from AI labs, capital markets firms, and high performance computing applications, we will begin to deliver TPUs to a select group of customers in their own data centers in a hardware configuration to expand our addressable market opportunity.”
Where the new facility will be located hasn’t been revealed. Blackstone declined to comment on that as of this writing. But a CEO has been chosen: Benjamin Treynor Sloss, presently Chief Programs Officer and former VP of Engineering at Google, who has been with the company since 2003.
Growing Interest in Alternative Chips
The Blackstone JV is the latest evidence of the growing popularity of Google’s TPUs. In April 2026, Anthropic announced an agreement with Google and Broadcom for 3.5 GW of capacity starting in 2027 and mostly sited in the U.S.
As noted, there is momentum building for other vendors’ chips as well, particularly inference chips, though Google appears to be the first cloud hyperscaler to offer its chips outside of its own cloud services, though Amazon seems to be heading in the same direction.
Amazon recently cited an annual run rate of over $20 billion for all of its chips, including Graviton, Trainium, and Nitro. OpenAI has signed on for 2 GW of Trainium capacity—on AWS cloud infrastructure. And Anthropic has agreed to use 5 GW of AWS capacity based on Graviton and Trainium chips, as well as any future Amazon chips, to train its Claude model and offer it to customers through AWS. All told, Anthropic has committed to investing a whopping $100 billion more in AWS technologies over a ten-year period.
The push for alternative chips to NVIDIA’s comes as enterprises, increasingly intent on inference workloads, seek to avoid lock-in to NVIDIA chips and the many related costs they involve. Shortage of memory chips is also causing companies to diversify their purchases of AI chips.
Futuriom Take: Blackstone and Google’s joint venture for a neocloud based on Google’s TPUs represents Google’s ambition to sell its chips outside of its cloud services. It also indicates the momentum growing behind alternatives to NVIDIA GPUs.
