Meta is set to start making its custom-designed AI chips in September to reduce spending on expensive GPUs from rivals like Nvidia. The move reflects Meta's massive investment in compute capacity—between $125 billion（約20兆円） and $145 billion（約23兆円） expected this year—as it ramps up AI model training and deployment across its services. Custom chips are part of a broader industry trend: Amazon and Google build their own chips, OpenAI unveiled an inference processor, and Anthropic is considering a similar path.
Summaries like this, in your inbox every morning.
Sign up free →What happened
Meta is on track to begin production of its custom AI chips in September, according to an internal memo cited by Reuters. At least one chip passed testing in about six weeks. The chips, developed under Meta's MTIA (Meta Training and Inference Accelerator) program and detailed in March, will be manufactured by TSMC with components from Broadcom, Samsung, SanDisk, and Sumitomo Electric.
Why it matters
Meta is spending heavily to power its AI efforts—the company expects capital expenditures between $125 billion（約20兆円） and $145 billion（約23兆円） this year. Custom chips allow the company to reduce its reliance on expensive GPUs from Nvidia and AMD while maintaining production at scale. Meta plans to deploy 7 gigawatts of compute this year and double that next year.
What to watch
The company intends to use MTIA chips for training ranking and recommendation algorithms, broader AI workloads, and inference for its applications. Meta has been producing its own AI chips since 2023 and is taking a modular approach so designs can adapt as AI evolves.
Meta's move to begin manufacturing its own AI chips in September represents a continuation of the company's strategy to own more of its compute stack. Since 2023, Meta has been producing custom chips, and the modular design approach—building each MTIA generation on the last with new chiplets—allows the company to keep pace with rapid AI evolution without costly redesigns. This effort is directly tied to Meta's unprecedented spending on AI infrastructure: with capital expenditures projected at $125 billion（約20兆円） to $145 billion（約23兆円） this year alone, reducing the cost of GPUs by supplementing them with custom silicon is a material lever.
Meta is not alone in this shift. Amazon and Google operate their own chip programs, OpenAI recently unveiled an inference processor built with Broadcom, and Anthropic is exploring custom chips with Samsung. For Meta specifically, the custom chips will handle training for recommendation systems and broader AI workloads, while the company continues to rely on external GPU suppliers for other needs. The memo cited by Reuters also reveals Meta's infrastructure ambition: the company plans to deploy 7 gigawatts of compute this year and double that capacity next year—a scale requiring both purchased and proprietary hardware to remain cost-efficient.
No discussion yet for this article
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
