Meta Strikes Multi-Billion Dollar Deal to Rent Google TPUs

calendar Feb 28, 2026 · 4 min read · meta google tpu ai hardware infrastructure aws nvidia amd chips  ·
Share on: linkedin copy

Meta Platforms has signed a multi-billion dollar agreement to rent Google’s Tensor Processing Units (TPUs) through Google Cloud to develop its new AI models, significantly expanding Google’s TPU commercialization strategy and allowing Meta to diversify its computing infrastructure beyond Nvidia and AMD.

The Deal

According to an original report by The Information and widely covered by Reuters, Meta has signed a contract to use TPUs via Google Cloud. The agreement is part of a broader effort by Google to challenge Nvidia’s dominance in the AI hardware market.

Google Cloud executives estimate that expanding TPU sales could capture up to 10% of Nvidia’s annual revenue — a market currently dominated by Nvidia’s GPUs.

Partnership Expansion

The agreement goes beyond simply renting computing capacity:

  1. Cloud Access: Meta uses TPUs through Google Cloud for model development
  2. Direct Purchase: Meta is in negotiations to buy TPUs for installation in its own data centers as early as 2027
  3. Joint Venture: Google is forming a joint venture to lease TPUs to other AI customers
  4. PyTorch Compatibility: In December, Google and Meta collaborated on the “TorchTPU” project to ensure full compatibility between TPUs and the PyTorch framework

Context: Meta’s Multi-Vendor Strategy

This agreement is part of a multipronged silicon strategy at Meta, described by Morningstar as an approach that leverages different vendors for different objectives:

  • Nvidia: For frontier model training
  • AMD: For inference needs (up to $60 billion 5-year deal announced on 02/24)
  • Google TPUs: For possible Llama workloads
  • Meta MTIA: For core recommendation algorithms (proprietary chips)

Meta also recently expanded its partnership with Nvidia to deploy millions of Blackwell and next-generation Vera Rubin processors.

Massive Infrastructure Investment

Meta is committing massive resources in 2026:

  • Capex Budget: $115 to $135 billion in AI infrastructure
  • Comparison: Nearly double the $72 billion spent in 2025
  • Data Centers: Plans for 30 data centers, with 26 located in the United States

Mark Zuckerberg, Meta’s CEO, stated that the AMD deal is an important step to diversify compute and deliver “personal superintelligence.”

TPU History

TPUs (Tensor Processing Units) are custom AI chips developed by Google more than a decade ago. Originally restricted to Google’s internal use and select Google Cloud customers, TPUs are now being aggressively commercialized.

In October 2025, Anthropic signed a tens of billions of dollars agreement for access to up to 1 million TPUs, preceding Google’s expanded commercial strategy.

Analysis

This agreement represents a turning point in the AI hardware market:

1. Strategic Diversification

Meta is building operational resilience by not relying exclusively on Nvidia. The scale of the company is such that it requires multiple alternatives, as analyst Nguyen noted in The Guardian.

2. Hardware Competition

Google is positioning its TPUs as a viable alternative to Nvidia’s GPUs, capturing a significant share of the AI training market.

3. Unusual Partnerships

Meta and Google, competing on multiple fronts (advertising, metaverse, search), are cooperating on AI hardware — a demonstration that in the AI era, competitive boundaries are being reconfigured.

4. Unprecedented Scale

The $135 billion in projected capex for Meta in 2026 is nearly double the previous year and represents an unprecedented level of investment in AI infrastructure.

Outlook

This agreement signals:

  1. TPU Market Maturation: Google’s TPUs are emerging as a legitimate competitor to Nvidia’s GPUs
  2. Multi-Vendor Standard: Large AI companies are adopting multi-vendor strategies to mitigate supply chain risks
  3. Infrastructure War: The race for computing capacity is heating up, with tens of billions in contracts being signed

Meta’s vendor diversification reflects the enormous scale of its AI operations, requiring the company to build redundancy and flexibility into its infrastructure.

Sources