Key Points
Meta moves Iris AI chip into production
Meta Platforms plans to start manufacturing its custom artificial intelligence chip, code-named Iris, in September. Iris is a data-center chip developed under the Meta Training and Inference Accelerators (MTIA) program, an in-house effort to design dedicated AI processors.
The internal planning around Iris marks a new phase for Meta’s chip strategy after several years of development work. The chip is tailored to the company’s own workloads and is aimed at deployment in its data centers to support large-scale AI applications.
Design, manufacturing and testing progress
Meta is working with Broadcom (AVGO) to help design the Iris chip and with Taiwan Semiconductor Manufacturing Co. (TSM) (TSMC) to manufacture it. This partnership structure allows Meta to retain architectural control while relying on established semiconductor expertise and fabrication capacity.
An internal memo indicates that testing of Iris took only about six weeks and found no major issues. The relatively short bug-testing period suggests that the current design stage has progressed without major technical setbacks, supporting the planned production timeline.
Role of Iris in Meta’s AI hardware stack
Iris is intended to augment, rather than replace, the large quantities of graphics processing units that Meta buys from Nvidia (NVDA) and Advanced Micro Devices (AMD). The chip targets Meta’s specific training and inference needs, working alongside GPUs that continue to provide general-purpose AI compute.
The memo notes that adopting the latest GPUs at a company of Meta’s scale has been a heavy lift and has cost time. Introducing an in-house accelerator is intended to address some of these constraints by giving Meta more direct control over part of its compute stack.
Planned chip cadence and infrastructure expansion
Meta unveiled Iris under its technical name in March, together with three other AI processors. The company plans to launch a new chip roughly every six months through 2027, establishing a multi-generation roadmap for its MTIA program.
The same internal document outlines an aggressive expansion of computing infrastructure. Meta plans to deploy seven gigawatts of computing capacity in 2026 and increase that to 14 gigawatts in 2027, effectively doubling capacity in one year.
The memo also shows that Meta expects to spend as much as $145 billion on AI infrastructure this year. This spending encompasses the hardware and data-center investments required to support its expanding portfolio of AI models and services.
Key Takeaways
- 01Meta is formalizing an in-house accelerator roadmap with Iris as the first MTIA chip slated for production.
- 02Partnerships with Broadcom (AVGO) and TSMC allow Meta to combine internal chip design goals with external manufacturing expertise.
- 03Iris is positioned as a complement to, not a replacement for, Nvidia (NVDA) and AMD GPUs in Meta’s data centers.
- 04The planned move from seven to 14 gigawatts of compute by 2027 underscores the scale of Meta’s AI infrastructure ambitions.
- 05Anticipated AI infrastructure spending of up to $145 billion this year highlights how central large-scale compute has become to Meta’s strategy.
References
- https://www.cnbc.com/2026/07/09/meta-to-put-ai-chip-into-production-in-september-report.html
- https://cnbc.com/2026/07/09/meta-to-put-ai-chip-into-production-in-september-report.html
- https://economictimes.indiatimes.com/tech/artificial-intelligence/meta-to-put-ai-chip-into-production-in-september-as-it-looks-to-double-computing-capacity-memo-shows/articleshow/132285360.cms
- https://money.usnews.com/investing/news/articles/2026-07-09/exclusive-meta-to-put-ai-chip-into-production-in-september-as-it-looks-to-double-computing-capacity-memo-shows