Skip to content
Source · Daily Brief

AI Daily Brief — 23 July 2025

A seismic policy Wednesday. Trump’s AI Action Plan crowded everything off the front page — three pillars, 100+ recommendations, three EOs. Alibaba’s Qwen team open-sourced Qwen3-Coder-480B-A35B-Instruct in parallel. Frontier US labs went quiet for the day.

Top stories

  • Trump unveils “America’s AI Action Plan”. 23-page, three-pillar plan (Accelerate Innovation, Build US AI Infrastructure, Lead in International Diplomacy/Security) with 100+ recommendations spanning trade, national security, energy, labor, antitrust and education. Drafted by AI/Crypto Czar David Sacks, Sriram Krishnan and OSTP head Michael Kratsios. via White House
  • EO: Accelerating Federal Permitting of Data Center Infrastructure. Directs federal agencies to streamline NEPA/Clean Water/Clean Air/FAST-41 reviews, opens federal land for siting, and tasks Commerce with loans, loan guarantees, grants, tax incentives and offtake agreements. Revokes Biden EO 14141. via White House
  • EO 14320: American AI Exports Program. Establishes a program run by Commerce, State and OSTP to bundle US semiconductors, AI models, software and standards into “full-stack” export packages backed by federal technical, financial and diplomatic support. Implementation deadline October 21. via White House
  • EO 14319: Preventing Woke AI in the Federal Government. Uses federal procurement power to require LLMs purchased by US agencies meet “Unbiased AI Principles” — Truth-seeking and Ideological Neutrality — explicitly targeting DEI, critical race theory, intersectionality and “systemic racism” concepts. Vendors must disclose how models are built, trained and modified. via White House
  • Trump keynote at “Winning the AI Race” summit. At the DC summit (Hill and Valley Forum + All-In Podcast), Trump pledged “whatever it takes” to win the global AI race, said fewer regulations are needed, framed the plan as removing “red tape” for Silicon Valley. Confirmed rollback of Biden-era AI chip export curbs and NVIDIA’s clearance to resume H20 sales to China. via NPR

Who shipped

The White House shipped a plan and three EOs. Alibaba shipped Qwen3-Coder. US frontier labs shipped silence.

Open-source pulse

Alibaba launched Qwen3-Coder-480B-A35B-Instruct — 480B MoE (35B active), 256K context (1M extrapolated), trained on 7.5T tokens at 70% code ratio. Apache-style weights on Hugging Face and GitHub. SOTA among open models on agentic coding, browser-use and tool-use, comparable to Claude Sonnet 4. Also open-sourced companion Qwen Code CLI mirroring Claude Code / Codex CLI. via Qwen Team

Money, infra & hardware

Indian conversational-AI firm Gupshup closed $60M+ co-led by Globespan Capital Partners and EvolutionX Debt Capital. April raised a $38M Series B for AI tax-planning. NVIDIA H20 China-sales policy was reaffirmed alongside the Action Plan — a 15% US revenue-share arrangement on NVIDIA and AMD China sales was reported in the surrounding week. via CNN

Quiet corners

Industry reaction split: Dell CEO Michael Dell publicly praised the plan; BSA CEO Victoria Espinel welcomed the workforce/governance pillars. Online-safety advocates and civil-society groups warned the deregulatory thrust and the “woke AI” procurement carve-out would let labs evade accountability for AI harms. via Tech Policy Press

By the numbers

  • 23 / 100+ / 3 — Action Plan pages / recommendations / accompanying EOs
  • Oct 21, 2025 — AI Exports Program implementation deadline
  • 480B / 35B / 256K (1M) — Qwen3-Coder params / active / context
  • 7.5T / 70% — Qwen3-Coder training tokens / code ratio
  • 15% — reported US revenue-share on NVIDIA/AMD China sales

Compiled by AI Feed’s editor from verified web sources for 23 July 2025.