Alphabet Inc's Google has entered into a multi-year arrangement to lease its tensor processing units - specialized AI accelerators - to Meta Platforms, according to a person involved in the negotiations. The contract, reported to be worth billions of dollars, will have Meta use the hardware in efforts to develop new artificial intelligence models.
The transaction represents a notable commercial deployment of Google’s proprietary AI chips to a large external customer. By making its tensor processing units available to Meta on a rental basis, Google is extending the commercial footprint of its silicon and seeking to monetize technology that underpins its own AI development.
Industry observers see the deal as another point of competition between Google and NVIDIA Corporation, which currently dominates the market for high-performance AI accelerators. The move by Google to place its chips with a major social-media and metaverse-oriented firm signals heightened rivalry among suppliers of advanced AI hardware.
Reports earlier in the week indicated that Google is exploring multiple paths to broaden the market for its AI processors, which have seen uptake among some startups. Those accounts noted that, despite growing adoption in parts of the market, Google remains constrained by manufacturing bottlenecks and faces limited interest from some of its so-called AI hyperscaler peers.
Separately, Meta has also signed its own multi-year supply agreement with AMD this week, underscoring the company’s strategy of sourcing AI hardware from multiple vendors as it seeks to accelerate development of next-generation AI capabilities. The two deals together suggest Meta is pursuing a multi-vendor approach to chip procurement.
Context and implications
The arrangement to rent tensor processing units to Meta expands Google’s commercial offerings for its AI silicon without disclosing new manufacturing investments or changes to its production capacity. At the same time, Meta’s concurrent supply agreement with AMD indicates the company is diversifying its hardware base rather than relying on a single supplier.
While the reported value of the rental contract is described as being in the billions of dollars and structured over multiple years, specifics on exact financial terms, allocation of capacity, and operational timelines were not disclosed in the report.
Bottom line
Google’s decision to lease its tensor processing units to Meta is a significant commercial step for its AI hardware business and highlights ongoing competition among major technology firms to provide the compute foundation for advanced AI models. Manufacturing limits and selective peer interest remain material considerations for Google as it seeks to grow the market for its chips.