Meta is moving its AI effort into a higher gear, with production of its in-house Iris chip slated for September and a plan to double computing capacity to 14 gigawatts next year. An internal memo frames the move as a bid to lower costs, reduce dependence on Nvidia and AMD, and speed AI development across Facebook and Instagram.
Meta’s in-house silicon push
Iris is part of a four-generation Meta Training and Inference Accelerators initiative built for Meta’s own workloads. The company tailored the data centre chip to its platforms, with Broadcom helping design and Taiwan Semiconductor Manufacturing Co handling manufacturing.
Testing wrapped in six weeks with no major issues, according to the memo. That pace marks an uptick for an in-house effort that had struggled for more than half a decade. Meta declined to comment on the bug-testing completion and production timing.
The chip is designed to augment, not replace, the large fleets of GPUs Meta buys from Nvidia and Advanced Micro Devices. Even so, adopting the latest GPUs at a firm of Meta’s scale ‘has been a heavy lift, and it has cost us time,’ the memo showed.
Here is what Meta’s chip cadence signals next:
– Production of Iris begins in September
– A new chip targeted about every six months through 2027
– Iris was unveiled in March with three other processors
Compute capacity and timelines
The memo says Meta plans to deploy seven gigawatts of computing infrastructure this year. It added 1 gigawatt in the first half and forecasts another 2.5 gigawatts by year-end. One gigawatt of energy is enough to power about 800,000 homes.
The same memo points to a doubling of capacity next year to 14 gigawatts. It also references a total of 14 gigawatts in 2027, indicating aggressive targets tied to the chip rollout and broader data centre expansion.
Costs, supply chain, and chipflation
Meta expects to spend as much as $145 billion on AI infrastructure this year. That is a sizeable slice of Big Tech’s more than $700 billion projected outlay on the technology, underscoring the scale of the current compute arms race.
To lock in critical parts, Meta has secured long-term, multi-year supply agreements, the memo showed. Partners include Samsung Electronics for memory chips, Sandisk for flash storage, and Sumitomo Electric for fiber-optic equipment.
Such deals have become vital as memory chip shortages push up prices and force companies such as Apple to raise prices. Components from memory to AI accelerators have seen rapid increases, with Morgan Stanley analysts warning that ‘chipflation’ is now a macroeconomic concern.
Competitive stakes and investor mood
The strategy directly targets a core bottleneck: reliance on third-party silicon. ‘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.’
Shares dipped after news of the internal plans but rebounded following Meta’s announcement of developer access to an AI coding model that pits it against OpenAI and Anthropic. The stock was trading up 4.6% in late afternoon trading, according to the memo.
What to watch next
Meta unveiled Iris under its technical name in March, alongside three other AI processors, and aims to ship a new chip about every six months through 2027. That breakneck cadence contrasts with the year-or-more intervals common across the sector.
If delivered, the plan could lighten Meta’s GPU dependence while compressing development cycles for its AI models. The real test begins in September, when Iris leaves the lab and enters production at scale.
