Nvidia Corp. experienced a decline in its stock value following reports that Meta Platforms Inc. is considering a significant investment in Google’s tensor processing units (TPUs) to enhance its data center capabilities by 2027. The news, sourced from The Information, indicated that Meta might also explore renting chips from Google’s cloud division as early as next year.
As a direct response to these developments, shares of Alphabet Inc., Google’s parent company, rose by as much as 2.7% in late trading, while Nvidia’s stock dropped by the same percentage. The potential deal between Meta and Google presents an opportunity for TPUs to establish themselves as a credible alternative to Nvidia’s chips, which have been favored by major tech companies, including Meta and OpenAI, for their AI computing needs.
Nvidia’s dominance in the market remains strong, but the anticipated agreement could mark a substantial shift in the competitive landscape. Google has already made strides in this area by distributing up to 1 million TPUs to Anthropic PBC, further validating the tensor chips as a viable option for AI development. Analyst Jay Goldberg from Seaport highlighted this deal as “powerful validation” for TPUs, suggesting that increasing numbers of companies may now regard them as serious contenders against Nvidia’s products.
While representatives from Meta declined to comment on the ongoing discussions, Google has yet to provide a response to inquiries regarding the reports.
Experts from Bloomberg Intelligence pointed out that Meta’s interest in utilizing Google’s TPUs reflects a broader trend where third-party providers of large language models are likely to turn to Google for accelerator chips in the near future. These experts estimated that Meta’s capital expenditure would exceed $100 billion for 2026, with a projected $40-$50 billion earmarked specifically for inferencing-chip capacity next year. This growth reflects increasing demand from enterprise customers eager to access TPUs and Google’s Gemini large language models (LLMs) via Google Cloud.
In Asia, stocks linked to Alphabet saw significant gains, indicating positive sentiment stemming from the news. Shares of IsuPetasys Co., which supplies multilayered boards to Alphabet, soared by 18% to an intraday record in South Korea. Taiwanese company MediaTek Inc. witnessed nearly a 5% increase in its stock as well.
A partnership with Meta, one of the largest global investors in data centers and AI, would be a notable achievement for Google. However, much will hinge on TPUs proving their efficiency and computational strength as a long-term alternative to Nvidia’s offerings. Developed over a decade ago, the tensor chips were specifically designed for AI-centric tasks. Their growing appeal comes amid surging concerns regarding overreliance on Nvidia, particularly as competitors like Advanced Micro Devices Inc. lag behind.
While GPUs, which Nvidia dominates, were initially created to enhance graphics rendering—especially in gaming—they have proven effective for training AI models due to their capacity for handling vast datasets and calculations. In contrast, TPUs are specialized application-specific integrated circuits (ASICs) tailored for specific functions. Google’s experience in developing advanced AI models, such as Gemini, through its DeepMind unit has allowed the company to refine its chip design based on real applications, benefiting both chip design and AI development teams.
