Meta plans Iris chip, previews Muse Spark 1.1

Meta’s shares rose after the company announced plans to begin production of an in-house AI chip, Iris, in September and opened a public developer preview of Muse Spark 1.1.

Iris is part of Meta’s four-generation Meta Training and Inference Accelerators program and is designed to handle training and inference workloads for Facebook and Instagram. Meta completed testing of the chip in six weeks without finding major faults. Broadcom assisted with the chip design and Taiwan Semiconductor Manufacturing Co. will handle production.

Meta intends to use Iris alongside graphics processing units it buys from Nvidia and AMD to reduce computing costs and lower reliance on external suppliers.

The company plans to deploy seven gigawatts of computing capacity this year and reach 14 gigawatts by 2027. To support the expansion it secured long-term supply agreements with Samsung Electronics for memory chips, Sandisk for flash storage and Sumitomo Electric for fiber-optic equipment.

Muse Spark 1.1 is available as a public developer preview and is focused on coding tasks and agentic AI capabilities. Developers can join a waitlist for API access through Meta’s developer portal. New API users will receive $20 in free credits before usage-based pricing begins. Meta is initially restricting API access to its own ecosystem rather than third-party AI marketplaces.

Alexandr Wang, Meta’s head of AI, described Muse Spark 1.1 as “is going to be served on top of the computer infrastructure that we’ve built” and called it the company’s “strongest model for agentic and coding work yet.” He characterized the pricing as “very aggressive and attractive” and noted the model was trained to strengthen coding abilities because “you kind of have to build coding capabilities as part of that in service of overall agentic capabilities.”

Meta reported that the fast six-week test run reflected progress on an in-house chip project that has been under way for more than five years. The company expects to spend as much as $145 billion on AI infrastructure this year.

Meta said it will build its first Canadian data center in Sturgeon County, Alberta, at an investment of more than CAD 13 billion. The project is expected to support about 3,000 construction jobs at peak and more than 300 operational roles, and it will run on 100% clean, renewable energy.

Meta stated the combined hardware and software initiatives are intended to lower long-term computing costs, reduce dependence on external suppliers and expand its portfolio of proprietary AI products as it competes with OpenAI, Anthropic and Google.