Meta's Self-Developed AI Chip Iris to Mass Produce in September, Partnering with Broadcom and TSMC
Meta is making a bold move in the AI hardware space. The company is teaming up with Broadcom and TSMC to develop its own data center processor, codenamed Iris, which is slated to begin mass production in September 2026.
According to Reuters and internal memos, Iris is part of Meta's custom-built infrastructure project (MTIA). The chip has already passed initial vulnerability testing within six weeks. Its main job? To supercharge the core content ranking, recommendation systems, and generative algorithms that power Facebook and Instagram. The goal is to cut data center costs and reduce Meta's reliance on third-party GPUs from companies like NVIDIA and AMD.

But Meta isn't stopping there. Facing global hardware shortages and what some call "chip inflation," the company has locked in long-term supply contracts with industry heavyweights like Samsung Electronics, SanDisk, and Sumitomo Electric. This ensures a steady flow of components for its ambitious plans.
Speaking of plans, Meta is thinking big on computing power. The company aims to deploy about 7 gigawatts of computing capacity by the end of 2026, and it expects to double that to 14 gigawatts by 2027. To make that happen, Meta's infrastructure capital expenditure for 2026 is set at a whopping $14.5 billion.
This self-developed chip marks a key turning point for Meta in the trillion-dollar computing power arms race among tech giants. While Iris currently serves as a complement rather than a full replacement for existing NVIDIA and AMD GPUs, Meta's aggressive development cycle—refreshing the chip every six months—will give it greater hardware deployment autonomy. On an industry level, this not only helps ease supply chain bottlenecks but also deepens Meta's push to build an independent AI ecosystem that isn't tied to external software and hardware constraints.
Key Points
- Meta partners with Broadcom and TSMC to develop its own AI chip, Iris, for data centers.
- Iris is designed to optimize ranking and recommendation systems on Facebook and Instagram.
- Mass production is scheduled for September 2026.
- Meta plans to deploy 7 gigawatts of computing power by end of 2026, doubling to 14 GW by 2027.
- Infrastructure spending for 2026 is $14.5 billion, with long-term supply contracts secured.