Microsoft Announces Maia 300 Inference Chip, Claims ~40% More Tokens per Megawatt
O'Reilly Radar's weekly AI roundup reports that Microsoft announced Maia 300, its latest custom inference chip, which the company expects to generate roughly 40% more tokens per megawatt than the preceding Maia 200. Microsoft plans to deploy the chip across its AI products, including Copilot. The roundup also notes that ASML reported customer demand for advanced lithography equipment already exceeds its delivery capacity, with orders extending into 2029 and 2030, and that OpenAI's own usage data shows weekly active users of Codex and ChatGPT Work among European small and medium-sized businesses up eightfold since May.
A more energy-efficient custom inference chip from Microsoft could ease the rising compute and power cost of running AI models as agentic workloads grow, though the 40% tokens-per-megawatt figure is Microsoft's own claim rather than an independent measurement.
This source is a digest of several news items (“This Week in AI: AI’s Growing Pains”). Find this story in the original.
#Microsoft#Maia 300#Maia 200#Copilot#ASML#OpenAI#Codex#ChatGPT Work