Key facts
- Meta plans to begin manufacturing its custom AI chip, codenamed 'Iris,' in September.
- The 'Iris' chip is designed in-house and aims to improve AI powering Facebook and Instagram.
- Meta aims to double its computing capacity to 14 gigawatts by 2027.
- The company is working with Broadcom for design and TSMC for manufacturing.
- Meta expects to spend up to $145 billion on AI infrastructure this year.
Meta Platforms intends to commence manufacturing of its custom-designed artificial intelligence chip, internally codenamed 'Iris,' in September. This initiative is a key component of the company's broader strategy to significantly expand its overall computing power, with a target of reaching 14 gigawatts by 2027. The 'Iris' chip, part of the Meta Training and Inference Accelerators (MTIA) project, is tailored for Meta's specific needs and aims to enhance the AI capabilities powering its Facebook and Instagram platforms.
The relatively swift six-week testing period for the 'Iris' chip, which revealed no major issues, signals positive progress for Meta's in-house silicon development efforts. The company is collaborating with Broadcom for chip design and Taiwan Semiconductor Manufacturing Co (TSMC) for its production. This approach is expected to help Meta reduce its substantial computing expenses and decrease its reliance on external chip providers like Nvidia and Advanced Micro Devices.
Meta's strategy involves augmenting the vast number of graphics processing units (GPUs) it currently purchases for AI applications. The memo indicated that adopting the latest GPUs has been a significant undertaking for the company. In addition to its custom chip, Meta plans to deploy 7 gigawatts of computing infrastructure this year, with an additional 2.5 gigawatts projected by year-end. The company anticipates spending as much as $145 billion on AI infrastructure in the current year, a substantial part of the projected over $700 billion outlay by major tech firms.
To support its expansion goals, Meta has secured long-term supply agreements with companies including Samsung Electronics for memory chips, Sandisk for flash storage, and Sumitomo Electric for fiber-optic equipment. These agreements are crucial amid a memory chip shortage that has impacted other tech giants. The surge in demand for components like memory and AI chips has led to rising prices, with 'chipflation' emerging as a macroeconomic concern, according to Morgan Stanley analysts.