Meta has reportedly turned to MediaTek to develop a new 2nm chip codenamed ‘Arke,’ specifically designed for AI inference tasks, as it faces rising capital expenditures in its AI race. While hiring talent from other companies has been one approach for the tech giant to bolster its AI efforts, it is also facing mounting costs associated with the use of NVIDIA’s GPUs, which are critical for processing millions of queries but come with a hefty price tag. These GPUs also require significant storage and cooling, adding to the financial burden. To offset these costs, Meta is now investing in its own solution. MediaTek, with its experience in designing cutting-edge chipsets and a strong partnership with Meta in developing custom silicon for smart glasses, will play a key role in bringing ‘Arke’ to life. The chip will use the latest 2nm process, with tape-out expected by the fourth quarter of 2025, leveraging TSMC’s advanced lithography technologies. The ‘Arke’ project marks a shift in Meta’s chip development strategy, as it was initially not part of the company’s plans, which had been focused on other chips like Iris and Olympus. However, in light of the rising costs of AI infrastructure, Meta appears to have made the strategic decision to fast-track Arke’s development to help lower dependency on external suppliers like NVIDIA. The mass production of ‘Arke’ is projected for the first half of 2027. This move also underscores the escalating costs of Meta’s AI infrastructure, which are expected to reach up to $119 billion this year alone. The company is clearly determined to develop a self-sufficient solution to handle AI inference more cost-effectively in the long run.