Meta in talks to use Google chips in AI expansion

627

Meta is in active discussions to adopt Google’s custom AI chips, known as TPUs (Tensor Processing Units), for running large-scale artificial intelligence models, according to a report by The Wall Street Journal. The move signals a growing effort among Big Tech firms to diversify beyond Nvidia’s high-demand GPUs and explore strategic chip partnerships.

If finalised, this agreement could see Meta deploying Google’s Cloud TPU v5 in its data centers to train and serve large language models (LLMs) at scale. Meta currently relies heavily on Nvidia’s hardware, including H100 chips, for training its Llama models and running inference across its AI products such as Meta AI and AI stickers.

The collaboration is still under negotiation and may not materialise, but it reflects Meta’s larger strategy to hedge against supply chain constraints and reduce dependence on Nvidia, whose chips are both expensive and often backlogged. For Google, partnering with Meta would be a validation of its in-house TPU performance and a direct challenge to Nvidia’s AI infrastructure lead.

The deal could also spark cloud-based AI infrastructure alliances between rival tech giants. While Meta and Google compete in consumer AI platforms, both are looking to optimise model performance, reduce latency, and scale inference workloads efficiently.

Meta’s internal efforts to develop its own custom silicon—Artemis, expected to debut next year—also point to a multi-pronged approach: in-house chips, Nvidia systems, and potential Google TPU integration. As AI workloads grow more complex, securing compute diversity is fast becoming a boardroom priority.