Meta’s new AI chips will begin production in September

In an effort to cut its GPU costs amid an unprecedented component shortage, Meta is on track to start manufacturing the latest versions of its artificial intelligence chip in September, Reuters reported, citing an internal memo.

At least one chip passed the testing phase in about six weeks, the memo said. Meta is working with Broadcom to design the chips, but will use Taiwan’s TSMC to manufacture them. It is also buying RAM from Samsung, storage from SanDisk, and fiber-optic equipment from Sumitomo Electric, according to the report.

Meta detailed the four new slides, developed under the Meta Training and Inference Accelerator (MTIA), in March, some of which are currently in publication or will be published this year or next. The company is taking a modular approach to designing these chips, and expects its needs to change as artificial intelligence develops rapidly by the time the chips are produced.

“Each generation of MTIA builds on the last generation, using standardized chipsets, incorporating the latest AI workload insights and hardware technologies, and deploying on a shorter cadence,” the company wrote at the time.

The chips are expected to help the company save money on purchasing GPUs from chipmakers like Nvidia and AMD, though it still expects to spend a lot with those providers as well, Reuters reports. Meta intends to use MTIA chips for training models for classification and recommendation algorithms, broader AI workloads, and inference targeting its applications. The social media company has been producing its own AI chips since 2023.

Meta is spending heavily on securing enough computing capacity to support its various AI efforts. The company said in April that it expects capital expenditures of between $125 billion and $145 billion this year, much of which will go toward artificial intelligence efforts.

The company has closed deals for data and power centers around the world, spending tens of billions to secure the computing power to train and deploy its new Muse Spark series of AI models. It plans to deploy 7 gigawatts of computing this year, and double that next year, according to Reuters, which cited the memo.

It also signed a deal with ARM last year to secure compute for its recommendation systems, as well as a multibillion-dollar deal with AMD for its Instinct GPUs, and a multibillion-dollar deal with Amazon to use the cloud giant’s native CPUs for AI-related needs.

Meta isn’t the only company trying to stem the flow of capital to Nvidia. Last month, OpenAI unveiled an inference processor it is building with Broadcom, and Anthropic is said to be considering developing its own chips with Samsung. Amazon and Google are both developing their own chips for AI training and inference, and there are a host of startups working in this space to meet the growing demand.

Meta declined to comment.