[News] Google TPU Push Reportedly Meets Cautious Early Adoption from AI Infrastructure Providers
While Google has signaled a strategic shift to offer its Tensor Processing Units (TPUs) to select external customers—moving beyond its legacy internal-only model—market reception among leading AI infrastructure providers remains cautious. Citing The Information, the Chosun Daily reports that specialized AI cloud powerhouses, including Nebius, Lambda, and CoreWeave, continue to anchor their hardware strategies firmly with NVIDIA.
As noted by Yahoo Finance, Alphabet Inc. said during its earnings call last week that it plans to sell its TPUs directly to select customers, enabling them to deploy the chips in their own data centers. The move comes shortly after Google unveiled two new TPU variants—TPU 8t for training workloads and TPU 8i for inference.
However, The Chosun Daily, citing The Information, notes that challenging NVIDIA remains an uphill battle. Earlier, in September 2025, The Information reported that NVIDIA signed a US$1.5 billion agreement with Lambda to lease around 18,000 GPUs over four years. More recently, in November 2025, Lambda also announced a multibillion-dollar deal with Microsoft to build AI infrastructure powered by tens of thousands of NVIDIA chips, according to CNBC.
The Information also cites Nebius Chief Revenue Officer Mark Boroditzky, notes that about 99% of customer demand is still for NVIDIA GPUs, adding that TPU interest is rare and mostly comes from former Google users.
Meanwhile, CoreWeave VP Nick Robbins, cited by the report, said that even if GPU demand were to ease slightly, the company would remain firmly focused on NVIDIA-based systems.
NVIDIA’s Entrenched Ecosystem behind Muted TPU Adoption
A key factor behind the muted TPU response is NVIDIA’s deep ties with major U.S. AI data center operators, not only as a core chip supplier but also as an investor, according to The Chosun Daily, citing The Information.
The reports point out that NVIDIA is deeply intertwined with these companies, citing its US$2 billion investment in CoreWeave in January, participation in Lambda’s US$480 million funding round last year, and a strategic partnership with Nebius announced in March that included a US$2 billion investment.
The reports further note that despite efforts by Google and other tech giants to commercialize in-house AI silicon, data center demand remains heavily concentrated on GPUs. While Google has reportedly explored deploying TPUs alongside GPUs at data centers, progress so far has been limited.
Google may still face an uphill battle in breaking into the AI infrastructure supply chain as a full-fledged provider, but it has made some progress in hyperscaler TPU leasing. As previously reported by The Information, Google has signed a deal to rent out its TPU to Meta Platforms. Under the multi-year agreement, Meta will use the chips to develop new artificial intelligence models, with the deal reportedly worth billions of dollars, the report said.
Read more
- [News] Google Makes First Appearance in Samsung’s Top Five Clients in 2025 amid Soaring TPU Demand
- [News] Broadcom Reportedly Eyes $100B AI Chip Revenue in 2027, Backed by 6 Key Clients Like Google
(Photo credit: Google)
