AI infrastructure spending dominated the day: AMD announced up to $5B in compute investment with Anthropic for 2GW of MI450 GPUs, unveiled its Helios rack-scale systems, and partnered with Cerebras on ultra-low-latency inference, while Intel rode the same AI capex wave to its fastest revenue growth in 15 years. On the model side, Black Forest Labs shipped FLUX 3 and Microsoft rolled out its own MAI-Image and MAI-Voice models, underscoring how many players are now building first-party models rather than just reselling frontier APIs. A coalition of 25 companies came out defending open-weight AI development, notably without Anthropic or OpenAI, sharpening the open-vs-closed fault line just as a VC essay argued open weights are decoupling model quality from competitive moat. On the tooling side, OpenAI pushed full-duplex voice control into Codex and ChatGPT desktop and Runway shipped a model router for generative media, while engineering-leadership reads pushed back on naive AI productivity gains and unpacked why a 'GPU-hour' is anything but a commodity.

• AMD bets big on AI infra: $5B Anthropic compute deal, Helios systems, Cerebras inference tie-up.
• 25 AI firms defend open weights; Anthropic and OpenAI notably abstain.
• OpenAI brings full-duplex voice control to Codex and ChatGPT desktop.
• Black Forest Labs ships FLUX 3, its latest open image-gen model.
• Intel posts fastest revenue growth in ~15 years on AI-driven demand.