NEW YORK/SAN FRANCISCO: Meta Platforms plans to begin manufacturing its latest in-house artificial intelligence chip in September as it accelerates efforts to expand computing capacity and reduce reliance on outside chip suppliers, according to an internal memo reviewed by Reuters.

The chip, code-named "Iris," is part of Meta's four-generation Meta Training and Inference Accelerators (MTIA) program, under which the company is developing its own AI processors to power services across Facebook and Instagram.

According to the memo, testing of the chip was completed in just six weeks without any major issues, marking significant progress for a project that has struggled since it began more than five years ago.

Meta is developing the chip with Broadcom and will use Taiwan Semiconductor Manufacturing Co. to produce it. The custom-designed processor is expected to help lower the company's computing costs while reducing dependence on suppliers such as Nvidia and Advanced Micro Devices.

Meta declined to comment.

The company said the new chip is intended to complement, rather than replace, the large numbers of graphics processing units (GPUs) it continues to purchase from Nvidia and AMD for AI workloads.

However, the memo noted that adopting the latest GPUs across Meta's operations "has been a heavy lift, and it has cost us time."

Shares of Meta briefly fell following the Reuters report before recovering to trade 1.4% higher in afternoon trading.

"Investors increasingly see custom silicon as a way to optimize costs over time, but in the near term it's accompanied by enormous capex, growing financing needs, and more questions about when those investments translate into meaningful returns," said Emarketer analyst Jacob Bourne.

Meta introduced Iris under its technical name in March alongside three other AI processors. The company plans to release a new AI chip roughly every six months through 2027, a faster schedule than the typical annual release cycle followed by many competitors.

The memo also outlined Meta's plans to deploy seven gigawatts of computing infrastructure this year before doubling capacity to 14 gigawatts in 2027.

The company expects to spend as much as $145 billion on AI infrastructure this year. To support that expansion, it has secured long-term supply agreements with Samsung Electronics for memory chips, Sandisk for flash storage and Sumitomo Electric for fiber-optic equipment.

Demand for AI infrastructure has fueled shortages of memory and other semiconductor components, driving up prices as technology companies race to build more data centers.