Meta expects to begin manufacturing its latest in-house AI chips in September, according to an internal memo obtained by Reuters. The move is part of a broader push to reduce its dependence on expensive GPUs from Nvidia and AMD amid a global shortage of computing hardware. One of the new chips already cleared testing in roughly six weeks, the memo noted.

The chips are the latest generation of Meta’s MTIA program, which the company first detailed in March. Designed in partnership with Broadcom, they will be fabricated by TSMC. The hardware stack also includes RAM from Samsung, storage from Sandisk, and fiber-optic gear from Sumitomo Electric. Meta is taking a modular approach to the design, using chiplet architecture so it can swap components as AI workloads and model requirements shift quickly. “Each MTIA generation builds on the last, using modular chiplets, incorporating the latest AI workload insights and hardware technologies, and deploying on a shorter cadence,” Meta said at the time. The chips will handle training for ranking and recommendation algorithms, broader AI tasks, and inference for Meta’s apps. The company has been building its own silicon since 2023.

Meta isn’t abandoning external suppliers. It still expects to spend heavily with Nvidia and AMD, and has inked multibillion-dollar deals with both. The company also signed with ARM last year for recommendation system compute and with Amazon to use its custom CPUs for AI. This year, Meta forecasts capital expenditures between $125 billion and $145 billion, much of it going toward AI infrastructure. It plans to deploy 7 gigawatts of compute in 2025 and double that in 2026. Meta is not alone in the push to chip independence: OpenAI recently unveiled an inference processor built with Broadcom, and Anthropic is reportedly exploring custom chips with Samsung. Amazon and Google already design their own AI accelerators, and a wave of startups is chasing the same demand. Meta declined to comment on the memo.

Source: TechCrunch