Meta is set to begin producing the latest versions of its in-house AI chips in September, according to a Reuters report citing an internal memo. The move is aimed at reducing the company's reliance on costly GPUs from suppliers such as Nvidia and AMD, particularly amid an unprecedented shortage of components. The push reflects a wider industry trend, as major AI players seek to curb the vast sums flowing to Nvidia by developing their own silicon.

The chips are being developed under Meta's Meta Training and Inference Accelerator (MTIA) programme, with Broadcom involved in the design and TSMC handling manufacturing; Meta is also sourcing RAM from Samsung, storage from Sandisk and fibre-optic equipment from Sumitomo Electric. At least one chip reportedly passed testing in around six weeks, and the modular "chiplet" chips will be used for ranking and recommendation models, broader AI workloads and inference. Meta, which has made its own AI chips since 2023, expects capital expenditure of between $125bn and $145bn this year, plans to deploy 7 gigawatts of compute in 2026 and double that in 2027. Rivals are moving similarly, with OpenAI unveiling a Broadcom-built inference processor, Anthropic reportedly weighing Samsung-made chips, and Amazon and Google already producing their own.