• Meta Platforms is negotiating a multi-billion-dollar deal to purchase Google's custom AI chips (TPUs), with rentals possible as early as 2025 and purchases starting in 2027.
  • This would mark Google's first major push to sell its Tensor Processing Units beyond its own data centers, potentially capturing up to 10% of Nvidia's annual AI chip revenue.
  • The deal could reshape the competitive landscape of AI computing infrastructure, offering Meta diversification while challenging Nvidia's market dominance.

Strategic Shift in AI Hardware

Meta Platforms is in advanced talks with Google to secure billions of dollars worth of custom artificial intelligence chips, according to people familiar with the negotiations. The potential agreement would see Meta begin renting Google's Tensor Processing Units through Google Cloud as early as next year, with direct purchases commencing in 2027.

The discussions represent a significant strategic pivot for both technology giants. For Google, it would mark the first time the company has sold or lent its proprietary TPUs to a major third party on this scale. Google Cloud executives believe this strategy could eventually capture up to 10% of Nvidia's yearly revenue from AI chips, these people said.

Competitive Implications

If finalized, the partnership would create a formidable challenge to Nvidia's dominance in the AI accelerator market, where the chipmaker currently controls over 90% of the market. The timing is particularly notable given the ongoing supply constraints for high-end AI hardware that have plagued the industry since the AI boom began in late 2022.

Meta's interest in Google's TPUs comes as the social media giant continues heavy investment in AI infrastructure to support its development of large language models and other AI initiatives. The company reported approximately $34 billion in revenue for the third quarter of 2024, with substantial portions allocated to AI research and development.

Google Cloud CEO Thomas Kurian has been spearheading efforts to boost external adoption of the company's custom AI hardware. The cloud division recently crossed the $10 billion quarterly revenue threshold for the first time in mid-2024, and expanding TPU availability to major customers like Meta could further accelerate that growth.

Market Dynamics

The potential deal highlights the increasing efforts by major tech companies to diversify their AI chip suppliers beyond Nvidia, whose graphics processing units have become the industry standard for training and running sophisticated AI models. Both Amazon and Microsoft have been developing their own custom AI chips, though neither has yet secured a commitment of this magnitude from a competitor.

Spokespeople for Meta and Google declined to comment on what they described as market speculation. Nvidia did not immediately respond to requests for comment.

Industry analysts note that while Nvidia's CUDA software ecosystem remains a significant advantage, the emergence of viable alternatives could gradually erode the company's pricing power. For Meta, access to Google's TPUs would provide crucial diversification in its AI infrastructure while potentially offering cost savings over time.

The negotiations are ongoing and terms could still change, according to people familiar with the matter.