Meta reportedly to lease Google AI chips in multibillion-dollar deal

Meta Platforms has struck a multi-year, multi-billion-dollar deal to lease AI chips from Google, The Information reported, citing a source involved in the discussions.

The Information said the chips would be used to develop new AI models amid increased industry investment in AI infrastructure.

In December, Google reportedly pushed its Tensor Processing Units (TPUs) as an alternative to Nvidia’s GPUs, with TPU sales becoming a key driver of its cloud revenue.

The Information reported that Meta is in talks with Google to buy TPUs for its data centers as early as next year, though the status of those talks could not be determined.

The Information also reported that Google signed an agreement with an unidentified large investment firm to fund a joint venture leasing TPUs to other customers.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

A software push could help unlock Meta’s hardware deal with Google

Meta’s choice to rent Google’s AI chips likely depended on recent software work.
Google has been working to make Tensor Processing Units (TPUs) run better with PyTorch. This includes shifting from the older PyTorch/XLA “lazy tensor” approach toward a more “native” PyTorch TPU backend. SemiAnalysis reports the push came mainly from Meta’s renewed interest in TPUs plus Meta’s preference to avoid moving to JAX, a machine-learning framework 1.
PyTorch on Google’s TPUs used to feel clunky. Stronger support makes TPUs a more practical option than Nvidia hardware for Meta teams that build AI models in PyTorch 1.
Pricing also matters. SemiAnalysis estimates that, for external customers leasing TPU v7, total cost of ownership (TCO) per hour could be up to ~30% lower than Nvidia’s GB200 under its assumptions, plus ~41% lower than GB300 1.

AI’s biggest buyers are dismantling the one-supplier model

Meta’s reported agreement with Google matches a broader effort to limit reliance on Nvidia. One commentary source estimates Nvidia controls about 85% of the AI chip market 2.
Large AI builders are spreading orders across more chip vendors.
Meta is also chasing large purchases from AMD. Anthropic runs its Claude models across Google TPUs, Amazon Trainium, and Nvidia GPUs through a multi-cloud architecture 3.
Mixing suppliers can strengthen buyer leverage. One report says OpenAI negotiated roughly 30% off its Nvidia compute fleet costs by using the competitive threat of TPUs, even before deploying them 1.