Google touts its TPUs as alternative to Nvidia AI chips; Meta expresses interest

Google is intensifying its effort to challenge Nvidia in the AI chip market, according to a report from The Information on Monday. Historically, Google has used its custom tensor processing units (TPUs) inhouse to run its own cloud infrastructure, offering them to customers only through Google Cloud for large AI workloads. The report says Google is now proposing a significant shift, allowing customers to install TPUs within their own data centres rather than relying solely on its facilities. And social media giant Meta is emerging as a major potential buyer.

Meta Platforms Inc., the owner of Facebook and Instagram, is reportedly in talks to spend billions of dollars to deploy Google’s TPUs in its data centres beginning 2027. Meta is also said to be considering renting TPU capacity from Google Cloud as early as next year. Currently, Meta’s AI systems primarily run on Nvidia GPUs. Notably, Meta is invested in the race to superintelligence, spending billions on acqui-hiring and talent acquisitions through other channels.

If finalised, a deal with Meta would mark a major milestone for Google’s hardware strategy. The company has been pitching TPUs to organisations with strict data security and regulatory requirements, such as large financial firms and high-frequency trading companies. The tech giant has been marketing tighter control over sensitive information with on-premises TPU deployment to potential clients.


Google Cloud executives have indicated internally that broader TPU adoption could enable Google to capture a meaningful share of the AI chip market from Nvidia. According to the report, they believe the effort could help Google target up to 10% of Nvidia’s annual revenue, representing several billion dollars.

The push comes as demand for AI computing capacity continues to surge and Nvidia remains the dominant supplier in the sector. By offering TPUs both through the cloud and directly inside customer facilities, Google is signalling a more aggressive approach in the expanding competition for AI infrastructure.