Meta set to produce its new AI chips in September

⚡

Key Takeaways

1Meta plans to launch production of its new AI chips in September to reduce its GPU costs.

2The chips, designed with Broadcom and manufactured by TSMC, aim to optimize Meta's AI workloads.

3Meta is heavily investing in AI infrastructure, with projected spending between $125 billion and $145 billion this year.

💡Why it matters — Meta's strategy to develop its own AI chips could transform the component market, reducing its reliance on Nvidia and AMD.

Meta Launches AI Chip Production to Cut Costs

In response to a global component shortage and to reduce its GPU expenses, Meta is preparing to launch the production of its new artificial intelligence chips starting in September. This information, revealed by Reuters from an internal memo, highlights Meta's commitment to optimizing its technological resources.

Collaboration with Broadcom and TSMC for Manufacturing

One of Meta's chips has already successfully passed its testing phase in about six weeks, according to the memo. For the design of these chips, Meta is closely collaborating with Broadcom, while manufacturing will be handled by Taiwan Semiconductor Manufacturing Company (TSMC). Meanwhile, Meta is sourcing RAM from Samsung, storage from Sandisk, and fiber optic equipment from Sumitomo Electric.

The MTIA Program and Its Innovations

In March, Meta unveiled four new chips developed under its Meta Training and Inference Accelerator (MTIA) program. Some of these chips are already being deployed or will be soon. The modular approach adopted by Meta for the design of these chips aims to anticipate the rapid evolution of technological needs.

The company explained that each generation of MTIA chips builds on the previous one, utilizing modular chiplets and incorporating the latest advancements in AI workloads and hardware technologies. This strategy allows for faster and more efficient deployment.

Cost Reduction and Resource Optimization

The new chips are expected to enable Meta to reduce its GPU purchases from major manufacturers such as Nvidia and AMD. However, the company plans to continue investing heavily with these suppliers. The MTIA chips will be used to train models for ranking and recommendation algorithms, as well as for other AI workloads and inferences in its applications. Since 2023, Meta has been producing its own AI chips.

Massive Investments in AI

Meta is allocating significant resources to ensure sufficient computing capacity for its AI projects. In April, the company announced projected capital expenditures between $125 and $145 billion for the year, a large portion of which is dedicated to its AI efforts.

The company has signed agreements for data centers and energy sources globally, investing tens of billions to ensure the computing capacity necessary for training and deploying its new series of AI models, Muse Spark. Meta plans to establish 7 gigawatts of computing capacity this year, with the goal of doubling that figure next year.

Strategic Partnerships to Enhance Capabilities

In addition to its own developments, Meta entered into an agreement with ARM last year to secure computing capabilities for its recommendation systems. It also signed a multi-billion dollar deal with AMD for its Instinct GPUs and another with Amazon to utilize the cloud giant's in-house CPUs for its AI needs.

A General Trend in the Industry

Meta is not the only one looking to reduce its dependence on Nvidia. OpenAI recently announced the development of an inference processor with Broadcom, and Anthropic is considering creating its own chips with Samsung. Amazon and Google are also engaged in developing their own chips for AI, a trend being followed by many startups seeking to meet growing demand.

Meta has chosen not to comment on this information.