Meta Platforms plans to start manufacturing a custom artificial intelligence chip code-named "Iris" in September, according to an internal memo seen by Reuters. The move is part of a push to double the company's overall computing capacity to 14 gigawatts next year and reduce its reliance on external chip suppliers like Nvidia and AMD.
The chip is the first in a four-generation project called Meta Training and Inference Accelerators (MTIA) that the company will design in-house. Testing took only six weeks and uncovered no major issues, the memo said, signaling momentum for an effort that has struggled since it began more than five years ago.
Meta is working with Broadcom to help design the chip and with Taiwan Semiconductor Manufacturing Co to produce it. The approach is likely to lower massive computing costs and give the company more independence from Nvidia and AMD, whose graphics processing units are still needed in large quantities for AI applications. Adopting the latest GPUs at Meta's scale "has been a heavy lift, and it has cost us time," the memo showed.
Custom chip design and testing
Meta unveiled Iris under its technical name in March alongside three other AI processors. The company plans to release a new chip roughly every six months through 2027, a faster cadence than the annual or longer cycles typical in the industry. The chip is tailored to Meta's own workloads for Facebook and Instagram, and the bug-testing completion had not been previously reported.
"You can't become an AI titan if you are dependent on another company for chips," said Mike Gualtieri, a vice president and principal analyst at Forrester. "The hyperscalers and even SpaceX all plan chips because it will be the only way to compete on price for model usage."
Massive infrastructure expansion
The memo revealed that Meta plans to deploy seven gigawatts of computing infrastructure this year. It added one gigawatt in the first half and forecasts adding another 5.5 gigawatts by year-end. One gigawatt is enough to power roughly 800,000 homes. The company then intends to double capacity again next year, reaching 14 gigawatts in 2027.
The scale of the buildout is part of a broader industry surge. Meta expects to spend as much as $145 billion on AI infrastructure this year, a significant slice of Big Tech's projected $700 billion-plus outlay. For IT and development teams managing data center operations, AI for IT & Development planning now routinely involves energy and hardware projections at this magnitude.
Supply chain agreements
To support the expansion, Meta has locked in multi-year supply deals with Samsung Electronics for memory chips, Sandisk for flash storage, and Sumitomo Electric for fiber-optic equipment. The memo showed that such long-term agreements have become critical amid a memory chip shortage that has already pushed companies like Apple to raise prices.
Demand for memory and AI chips has grown so rapidly that "chipflation" has turned into a macroeconomic concern, Morgan Stanley analysts said. The memo's details on supply agreements and production timing offer a rare look at how a major tech firm is securing components ahead of expected shortages.
Why this matters for technology leaders
Meta's decision to design its own MTIA chips and double computing capacity provides a concrete case study for chief technology officers evaluating AI infrastructure strategies. AI Learning Path for CTOs often examines how custom silicon can shift cost structures and reduce vendor lock-in. The six-week testing cycle and the plan to launch new chips every six months show that in-house chip development, once seen as too slow and risky, is becoming a realistic option for hyperscalers - and a signal for any enterprise betting heavily on AI.
Your membership also unlocks:
