Meta Prepares to Launch Custom AI Chips This Fall, Aiming to Cut GPU Costs

Meta is gearing up to start production of its latest artificial intelligence chips in September, according to an internal memo obtained by Reuters. The move is part of the social media giant's broader effort to reduce its reliance on expensive GPUs from Nvidia and AMD amid a persistent shortage. At least one of the new chips passed its testing phase in roughly six weeks, the memo noted. Meta is collaborating with Broadcom on chip design and will rely on Taiwan Semiconductor Manufacturing Company (TSMC) for fabrication.

The chips are the latest additions to Meta's Training and Inference Accelerator (MTIA) program, which the company detailed in March. Meta is taking a modular approach to chip design, allowing it to adapt as AI workloads evolve. "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," the company said at the time. Meta plans to use these chips for training ranking and recommendation algorithms, broader AI workloads, and inference tasks across its apps. The company has been developing its own AI chips since 2023.

Meta has been spending heavily to secure compute capacity, with capital expenditures expected to hit $125 billion to $145 billion this year. The company has inked data center and power deals globally, and plans to deploy 7 gigawatts of compute in 2024, doubling that next year. It also signed a multibillion-dollar deal with AMD for Instinct GPUs and an agreement with Amazon to use its custom CPUs for AI. Meta is not alone in seeking alternatives to Nvidia: OpenAI recently unveiled an inference processor built with Broadcom, and Anthropic is reportedly exploring its own chip development with Samsung. Meta declined to comment on the report.

Source: TechCrunch