Meta expands plans to rely on its own AI chips

11:25

By: Azzat Manal

Meta expands plans to rely on its own AI chips

Meta is preparing to increase its use of internally developed artificial intelligence chips as it seeks to control the rising costs of operating advanced AI systems and improve the efficiency of its data center infrastructure.

The company is testing the third generation of its custom AI accelerators, known as MTIA 450 and reportedly codenamed “Artemis.” At the same time, Meta is nearing completion of the design of a subsequent generation, referred to as the MTIA 500 or “Astrea,” according to information cited by Bloomberg.

The newer chips are expected to begin deployment in Meta’s data centers during the first half of 2027. Their use is likely to expand gradually as the company’s demand for computing power grows alongside the development and operation of increasingly sophisticated AI models.

Meta’s move reflects a broader trend among major technology companies toward designing specialized processors for their own workloads. Instead of relying exclusively on commercially available chips, companies are developing custom hardware that can be optimized for specific AI applications and data center requirements.

Yi-Jiun Song, a Meta vice president of engineering, said the company is seeking continuous improvements in chip efficiency. The objective is for each new generation to deliver greater computing performance for every unit of energy consumed and every dollar invested.

According to Song, Meta expects its deployment of these internally developed processors to exceed one gigawatt over a 12-month period. The pace could increase further if demand for AI computing capacity remains strong, highlighting the scale of infrastructure required to support the company’s expanding AI operations.

Energy consumption has become a major consideration for technology companies building large AI systems. Training and running advanced models require powerful computing infrastructure, while data centers also need substantial amounts of electricity for both processors and cooling systems.

Custom AI chips can offer companies greater control over how computing resources are allocated. By tailoring hardware to specific workloads, technology firms can seek improvements in energy efficiency, processing performance and operating costs compared with more general-purpose solutions.

For Meta, the strategy is also linked to the rapid expansion of its AI ambitions across its social media platforms, recommendation systems, generative AI products and other services. These applications require increasingly large amounts of computing power as the company expands its use of artificial intelligence.

The development of its own processors does not necessarily mean Meta will abandon external chip suppliers. Specialized in-house hardware can instead complement commercially available processors, allowing the company to distribute different workloads across several types of computing systems.

Meta’s investment in custom silicon also illustrates the growing importance of hardware in the global AI race. While software models and algorithms remain central to the development of artificial intelligence, access to efficient and scalable computing infrastructure has become an increasingly important factor for companies seeking to expand their AI capabilities.

The planned deployment of newer MTIA processors in 2027 will therefore represent another step in Meta’s effort to build a more diversified computing infrastructure. If demand for AI services continues to grow, internally designed chips could become an increasingly significant part of the company’s long-term strategy for managing energy consumption and infrastructure costs.