Meta hopes to have its custom AI chip installed in its data centers by September, as it seeks to reduce the cost of training and running AI models.
It is the fourth iteration of the company’s custom chip and has reportedly performed well in early tests, with full deployment expected ahead of schedule.
The social media giant designed the chip alongside Broadcom, with Taiwan Semiconductor Manufacturing Co manufacturing it, according to Reuters. It comes as Meta looks to gain more control over costs and operations away from Nvidia and AMD, just a few months after it inked a partnership worth up to $100 billion with the latter.
Meta plans to launch a new chip every six months or so, a faster timeline than most major chipmakers. The project, called Meta Training and Inference Accelerators, focuses on custom silicon that can run AI more efficiently. For Meta, that should reduce AI costs across Instagram and Facebook while improving performance.
Double its overall computing power by 2027
Meta has taken major steps to establish itself as a name in AI infrastructure, but it remains one of the few tech giants without a cloud computing division.
That may soon change, with internal discussions reportedly taking place over whether to sell capacity to Meta partners, or excess capacity to rival AI developers such as Anthropic and OpenAI. This would take it down a similar path to SpaceX, which is generating more than $2 billion per month by selling spare capacity.
According to Reuters, Meta plans to double its computing capacity from seven gigawatts in 2026 to 14 gigawatts in 2027, through a mix of its own data centers, partnerships with neoclouds, and agreements with hyperscalers.
To put that into perspective, OpenAI’s Stargate project, which is expected to cost project leaders OpenAI, Oracle, and SoftBank more than $500 billion by the end of the decade, has 10 gigawatts planned.
More must-read AI coverage
- SS&C Intralinks DealCentre AI vs. Datasite: Which platform is built for the future of dealmaking?
- SS&C Intralinks FundCentre AI vs. Juniper Square: Which platform better supports modern private markets fund managers?
- Why Data, Not Models, Determines AI Success
- The Rise of the AI-Native Factory: How Physical AI Is Transforming Manufacturing
All in on AI and smart glasses
After the failures of the metaverse, Meta has shifted much of its focus to AI. It is now trying to become both a frontier AI model developer and a major provider of consumer AI technology through Facebook, Instagram, and Meta AI. It has recently integrated a new AI image model, Muse Image, into its three main social apps.
Its smart glasses, the only real success of its wearables division, are also being pivoted toward AI. Meta has integrated Meta AI into them, allowing the assistant to view the world through the in-built camera. This may go further in future updates, with Meta reportedly looking to capture video and audio every few seconds to provide a more agentic service based on the user’s surroundings and day.
What this means for users
For everyday users, Meta’s custom chip push will mostly show up in the background. If the company can lower the cost of running AI, it could bring more AI features to Facebook, Instagram, WhatsApp, and Meta AI without charging directly for every new tool.
That could mean faster image generation, more responsive assistants, better content recommendations, and AI features that work across more Meta products. For businesses and creators, cheaper AI infrastructure could also make Meta’s ad tools, customer service features, and content generation products more capable over time.
But there is a trade-off. More powerful AI across Meta’s apps could also raise new questions about privacy, data use, personalization, and how much automated decision-making users want inside their social feeds.
Also read: Meta isn’t alone in building custom AI silicon. Learn how OpenAI is pursuing a similar strategy with its new Jalapeño inference chip.