OpenAI, Oracle, SoftBank Announce 7GW AI Buildout

25 Sep 2025


OpenAI’s Sam Altman outlined an “Abundant Intelligence” vision and a plan to produce roughly 1 gigawatt of AI compute capacity every week. Hours later OpenAI, Oracle, and SoftBank announced five new “Stargate” data center sites across the U.S. — a buildout Oracle says will deliver roughly 5.5 GW in Texas, New Mexico, and the Midwest (and create 25,000+ jobs), while SoftBank is fast-tracking two sites in Ohio and Texas for another ~1.5 GW expected within 18 months. Combined with OpenAI’s Abilene campus and CoreWeave projects, the partners say they’ve reached about $400B in investment commitments and roughly 7 GW of planned capacity, putting them ahead of a 10 GW / $500B target by the end of 2025.

The newsletter also highlights an Attention Labs demo that routes on-device voice to the assistant actually being addressed (no wake words or cameras), claiming sub‑100ms latency and ~97% accuracy — a privacy-friendly “social layer” for ambient voice AI. Alibaba rolled out six new Qwen-based releases (e.g., Qwen3Guard, Qwen3-Max, LiveTranslate, Qwen3-Coder, Qwen-VL, travel planner), and Microsoft has been in talks about a publisher content marketplace to compensate creators when their work is used by AI.