Google and Meta Ink AI Chip Deal to Counter Nvidia

The deal involves Meta renting Google's custom Tensor Processing Units (TPUs) in a multi-year, multi-billion-dollar agreement to train and run its next-generation AI models. Meta is also reportedly in discussions to purchase TPUs directly from Google for its own data centers starting as early as 2027. This partnership follows Meta's recent, massive chip acquisition strategy, including a deal to deploy up to 6 gigawatts of computing power from AMD, and another to purchase millions of chips from Nvidia. This flurry of agreements highlights a broader industry trend of major AI players diversifying their hardware sources to mitigate reliance on a single provider. Nvidia remains the undisputed leader in the AI chip market, controlling an estimated 70% to 85% of the market share with its highly sought-after GPUs and CUDA software platform. Google has been aggressively positioning its TPUs as a viable, cost-effective alternative to Nvidia's dominant hardware. Meta's move to secure external chip supplies comes amid reported setbacks in its own custom silicon project, the "Meta Training and Inference Accelerator" (MTIA). The company has reportedly halted the development of its most advanced training chip, codenamed "Olympus," making external partnerships critical to keep pace in the AI development race. The competitive pressure for such alliances has been intensified by Amazon's recent strategic partnership with OpenAI. That deal includes a massive $50 billion investment from Amazon and deepens the integration of OpenAI's models with Amazon Web Services (AWS) and its custom Trainium AI chips. Google has ambitious goals for its custom silicon, with internal targets of capturing a significant portion of Nvidia's revenue through TPU sales. The company has also established a joint venture to facilitate the leasing of TPUs to a wider range of customers, signaling a serious push to challenge Nvidia's market dominance.