Meta will begin manufacturing its custom "Iris" AI chip in September 2026 to reduce costs and reliance on suppliers like Nvidia. An internal memo reveals plans to double computing capacity to 14 gigawatts by 2027, part of a $145 billion infrastructure investment. Meta is partnering with Broadcom for design and TSMC for manufacturing, signaling a major push for in-house silicon to power its social media platforms.
Iris AI chip development
- ▪Testing for the "Iris" chip was completed in six weeks with no major issues found
- ▪Meta plans to start manufacturing its custom AI chip, codenamed "Iris," in September 2026
- ▪The "Iris" chip is part of Meta's four-generation project for in-house Meta Training and Inference Accelerators (MTIA)
- ▪Meta plans to launch a new in-house AI processor approximately every six months through 2027
Computing infrastructure expansion
- ▪A surge in demand for AI components has caused rapid price increases, a phenomenon Morgan Stanley analysts termed "chipflation."
- ▪Meta expects to spend as much as $145 billion on AI infrastructure in 2026
- ▪Meta plans to deploy seven gigawatts of computing infrastructure in 2026
- ▪Meta aims to double its computing power to 14 gigawatts in 2027
Supply chain partnerships
- ▪Meta has secured long-term supply agreements with Samsung for memory chips and Sandisk for flash storage
- ▪Meta is collaborating with Broadcom on the design of its "Iris" chip
- ▪Taiwan Semiconductor Manufacturing Co (TSMC) has been contracted to manufacture the "Iris" chip for Meta
- ▪Meta has a multi-year supply agreement with Sumitomo Electric for fiber-optic equipment
Cost reduction strategy
- ▪The in-house chip project aims to lower Meta's computing costs and reduce its dependence on suppliers like Nvidia and AMD
- ▪Meta's in-house chip development effort, which began over five years ago, is now showing positive momentum after previously floundering
- ▪The custom "Iris" chip is intended to augment, not replace, the large quantities of GPUs Meta buys from Nvidia and AMD
- ▪An internal Meta memo stated that adopting the latest GPUs from external firms "has been a heavy lift, and it has cost us time."
Story comments
Loading comments…