Nvidia shares fell 4.1% in premarket trading after reports that Meta is evaluating Google’s Tensor Processing Units (TPUs) for parts of its AI infrastructure, a move that could shave roughly $180 billion off Nvidia’s market value.
According to people familiar with the discussions, Meta may begin renting Google’s advanced TPUs through Google Cloud as early as next year, with broader deployment across Meta data centers targeted for 2027. The talks represent a notable potential shift for Meta, which has relied heavily on Nvidia GPUs to train large language models like Llama.
Alphabet shares rose as investors reacted to the prospect of increased TPU adoption outside Google. Google’s TPU family, now in its seventh generation with the latest chips codenamed "Ironwood," has primarily served Google’s internal needs and Cloud customers; a major deployment by Meta would mark its largest external use to date.
Nvidia has emphasized its full-stack approach combining hardware, software, and ecosystem support, and the company declined to comment on the specific report. Meanwhile, other suppliers across the semiconductor chain, including Broadcom, saw share gains as the AI infrastructure market continues to expand despite shifting competitive dynamics.
Industry observers say the Meta–Google discussions illustrate how the high cost and complexity of AI infrastructure can push even direct competitors toward selective collaboration. Representatives for Meta and Google did not immediately comment on the reported talks.