Google Challenges Nvidia Dominance With Meta TPU Deal

Alphabet Inc. is aggressively expanding the commercial availability of its custom Tensor Processing Units (TPUs), moving from internal use to external sales to compete directly with Nvidia Corporation. In a potential multibillion-dollar shift, Meta Platforms Inc. is negotiating to integrate Google TPUs into its data centers starting in 2027, with a preliminary phase of leasing capacity via Google Cloud as early as 2026. This move aims to reduce Meta's heavy reliance on Nvidia GPUs and lower infrastructure costs amid massive capital expenditures.

Google has bolstered this hardware push by releasing the Gemini 3 AI model, which was trained entirely on in-house TPUs. The company has also secured a deal to supply up to one million chips to Anthropic PBC and provided TPU clusters for Apple's foundation models. To facilitate adoption, Google enhanced its JAX software to improve compatibility with open-source frameworks like PyTorch, reducing the industry's dependency on Nvidia's CUDA platform.

Market reaction was immediate, with Nvidia shares falling between 3% and 7% in late November 2025, while Alphabet shares rose toward a $4 trillion market capitalization. Nvidia CEO Jensen Huang defended the company's position, asserting that its platform remains "a generation ahead of the industry" and is the only one capable of running every AI model across all computing environments. Despite this, analysts suggest the shift toward application-specific integrated circuits (ASICs) for inference tasks could erode Nvidia's market share from 85% to 75%.