Meta shares rose after the company announced it will deploy its own AI chips in data centers

Meta Platforms shares rose 2% on Tuesday morning after Bloomberg reported that the company plans to deploy its own in-house AI chips in data centers next year. The gain later narrowed to 0.75%.

The social media giant is currently testing the third generation of its MTIA chip line, the MTIA 450, also known as Arke. The next version, called MTIA 500 or Astrid, is expected to complete its design phase in about a month and should be deployed in data centers by the end of 2027. Meta first announced its plans to build its own AI chips in 2023.

The company is developing the chips in collaboration with Broadcom, while using Taiwan Semiconductor Manufacturing Co. (TSMC) for production. Meta's goal is to reduce its reliance on Nvidia processors and control AI infrastructure costs. Over the next 12 months, the company plans to deploy a volume of these chips corresponding to more than 1 gigawatt of capacity.

According to Yee Jiun Song, Meta's vice president of engineering, each new generation of chips delivers higher performance per unit of energy and per dollar spent. He noted that when running AI models, the company's in-house chips operate more efficiently compared to Nvidia's current solutions.

On September 1, Meta received 12 test samples of the new chips from TSMC. Their real-world performance fell short of simulation-based results by only 2%-3%. On the very first day, the team ran Meta's own models, as well as models from DeepSeek and Alibaba Group, on the chips.

The chips use high-bandwidth memory and are primarily geared toward general-purpose inference — that is, generating responses from already-trained AI models — rather than ultra-fast inference.

Meta's decision to develop its own AI chip ecosystem is also tied to the need to reduce costs. The company previously canceled a chip project known as Olympus, planned for 2028-2029, and shifted its focus to inference. According to Song, when the company is building infrastructure at gigawatt scale, a potential 30% increase in costs becomes unacceptable.

Tuesday's rise in Meta's shares further reinforces the positive sentiment surrounding the company's AI strategy. Investor attention has also recently been drawn to Meta's launch of a new AI agent, Muse.