Meta’s new AI chips will begin production in September

Meta is gearing up to begin production of its latest AI-focused chips this September, as part of a strategy to reduce reliance on expensive GPU components during a significant supply shortage. According to an internal memo reported by Reuters, one of the new chips successfully completed its testing phase in just six weeks. The tech giant is collaborating with Broadcom for the chip's design, while manufacturing will be handled by Taiwan Semiconductor Manufacturing Company (TSMC). In addition, Meta is sourcing RAM from Samsung, storage from SanDisk, and fiber optic equipment from Sumitomo Electric. In March, Meta introduced four new chips under its Meta Training and Inference Accelerator (MTIA) program. Some of these chips are already being deployed or are expected to launch within the year or next. The company is adopting a modular design approach, which allows for adaptations in response to the rapidly evolving landscape of AI technology. "Each MTIA generation builds on the last, utilizing modular chiplets that integrate the latest AI workload insights and hardware innovations, deployed on a shorter timeline," the company stated. These new MTIA chips aim to help Meta cut costs associated with purchasing GPUs from major manufacturers like Nvidia and AMD, though reports indicate that the company will continue to invest significantly in these suppliers. The chips will be instrumental in training models for Meta's ranking and recommendation systems, as well as handling broader AI workloads and inference tasks across its applications. Since the beginning of 2023, Meta has been actively developing its own AI chips, investing heavily to secure the necessary computing power for its various initiatives. Earlier this year, the company projected capital expenditures ranging between $125 billion and $145 billion, much of which is directed towards AI advancements. Meta is currently negotiating data center and power agreements globally, with plans to deploy an impressive 7 gigawatts of computing power this year and potentially double that next year. In addition to its partnerships, Meta has also signed a deal with ARM for computing needs related to its recommendation systems and has secured multi-billion dollar agreements with AMD for Instinct GPUs and Amazon for its proprietary CPUs. The competitive landscape is heating up, with other companies like OpenAI and Anthropic also exploring chip development to mitigate costs associated with Nvidia’s offerings. Meta has not provided comments regarding this initiative.