Key Takeaways
- Meta Platforms is set to manufacture its first AI chip, code-named Iris, starting in September.
- The company aims to double its computing capacity to 14 gigawatts by 2027 as part of its AI expansion plans.
- Iris will work alongside existing Nvidia and AMD GPUs in Meta’s data centers.
Meta Platforms has announced plans to begin manufacturing a new artificial intelligence chip, code-named Iris, starting in September. This move is part of the company's broader strategy to enhance its computing infrastructure and support its AI systems across platforms such as Facebook and Instagram.
According to an internal memo reviewed by Reuters, Meta is designing the processors internally to better align with its AI workloads. The chip will be manufactured by Taiwan Semiconductor Manufacturing Company (TSMC) in collaboration with Broadcom on the design front. This partnership marks a significant step for Meta, which has been working on custom-chip development since 2018.
The rapid testing of Iris was completed within six weeks and found no major issues, indicating positive progress for Meta’s custom-chip program. However, this project has faced challenges in the past, with Meta declining to comment on previously unreported testing results and September production schedule.
Meta plans to operate seven gigawatts of computing infrastructure by 2026, expanding its capacity to 14 gigawatts by 2027 as part of its Meta Compute initiative. One gigawatt is approximately enough electricity to power 800,000 homes, highlighting the scale of this expansion.
The company expects to spend up to $145 billion on AI infrastructure in 2026, a significant portion of the more than $700 billion that major tech companies are projected to invest in AI during the year. This substantial investment underscores Meta's commitment to advancing its AI capabilities and reducing reliance on external chip suppliers.
Forrester analyst Mike Gualtieri noted that major AI companies increasingly need their own processors to compete effectively on the cost of running AI models, a trend that Meta is embracing through its custom-chip development. Iris will be part of a four-generation project, with one new chip introduced every six months until 2027.
While Iris will work alongside existing Nvidia and AMD GPUs in Meta’s data centers, the company does not plan to replace these suppliers entirely. The integration of the latest GPUs has required significant effort and cost, according to the internal memo. This move could help Meta lower its computing expenses and reduce dependence on external chip providers.
