Meta Platforms has plans to start production of an artificial intelligence (AI) chip starting in September. The initiative aims to increase the company’s total computing capacity to 14 gigawatts (GW) next year, as revealed in a recent internal memo.

The technology giant’s data center chip, called “Iris”, is part of a four-generation project called Meta Training and Inference Accelerators (MTIA), developed internally. The main goal is to utilize this custom silicon to enhance the artificial intelligence that powers its social media platforms such as Facebook and Instagram.

The chip tests were completed in just six weeks, without any major problems being identified, according to the same memo. This relatively rapid progress indicates positive momentum for an internal effort that has struggled since its launch more than five years ago.

Meta customized the chip to its own requirements and collaborates with Broadcom on the design, while Taiwan Semiconductor Manufacturing Co. will be responsible for manufacturing. This strategy tends to reduce the company’s high computing costs and provide greater autonomy in relation to chip suppliers such as Nvidia and Advanced Micro Devices.

The completion of bug testing and the production schedule had not previously been disclosed. Meta, however, chose not to comment on the matter.

This new chip is designed to complement the large quantities of graphics processing units (GPUs) acquired by Meta from Nvidia and AMD, which are used in its various AI applications.

However, the memo pointed out that incorporating the latest GPUs in a company the size of Meta “has been an arduous process, costing the company time.”

The company’s shares fell after the announcement, but recovered after the company announced developer access to an AI model for coding, which puts it in direct competition with OpenAI and Anthropic. At the end of afternoon trading, shares were up 4.6%.

“It is not possible to become an artificial intelligence giant while being dependent on another company to produce chips,” said Mike Gualtieri, vice president and principal analyst at consultancy Forrester. He added that “big technology companies and even SpaceX plan to develop chips, as this will be the only way to compete on price for using models.”

Meta presented Iris, under its technical name, in March this year, along with three other artificial intelligence processors. The company aims to launch a new chip every six months until 2027, a considerably faster pace, as companies generally launch AI chips at intervals of a year or more.

Expansion of computing infrastructure reaches 7 gigawatts in 2026

Meta aims to deploy seven gigawatts of computing infrastructure this year, according to the memo. To reach this mark, the company added 1 GW in the first half of the year and plans to add another 2.5 GW by the end of 2024.

To put it into context, one gigawatt of energy is enough to supply approximately 800,000 homes.

The company plans to double that capacity again next year, targeting a total of 14 gigawatts in 2027, the memo said.

The company is expected to invest up to 145 billion dollars in AI infrastructure this year, representing a considerable portion of the more than 700 billion dollars that large technology companies expect to spend on this technology.

To enable the expansion of its computing infrastructure, Meta entered into long-term, multi-year supply agreements, the document revealed. Partners include Samsung Electronics, for memory chips; Sandisk, for flash storage; and Sumitomo Electric, for fiber optic equipment.

These long-term contracts have become crucial to data center expansion goals, amid a shortage of memory chips that has led companies like Apple to raise prices.

Sandisk chose not to comment on the matter. Samsung Electronics and Sumitomo Electric did not respond to requests for a statement.

Components such as memory and artificial intelligence chips have seen a significant increase in demand as technology companies race to expand their data centers and keep up with the growing need for AI computing power.

The prices of memory chips and other types rose quickly and substantially, to the point that “chipflation” (chip inflation) has become a macroeconomic concern, according to an analysis by experts at Morgan Stanley.