Meta Platforms plans to deploy its third-generation MTIA 450 processor, code-named Arke, in data centers during the first half of 2027 as it seeks to reduce AI infrastructure costs, energy use and dependence on Nvidia hardware. Twelve Arke units manufactured by Taiwan Semiconductor Manufacturing Co. arrived on Sept. 1, with early test results within 2% to 3% of Meta's simulations. The company tested Meta, DeepSeek and Alibaba Group models on the chips. Meta is focusing its custom processors on AI inference, the process of running trained models, rather than training workloads or applications requiring near-instant responses. MTIA 500, code-named Astrid, is expected to complete design work within about a month and enter data centers by late 2027, with a broader rollout than Arke. Meta also began manufacturing another in-house processor, Iris, in September, while its earlier MTIA 300 is already used for ranking and recommendation systems. The company expanded its Broadcom partnership through 2029 to cover multiple MTIA generations, including design, packaging and networking, and committed to deploy more than one gigawatt of computing capacity. Meta abandoned Olympus, a planned training-and-inference chip targeted for 2028 or 2029, after concluding its estimated 30% higher cost was unsuitable for deployment at gigawatt scale.