AI Daily Brief — 03 February 2025
Monday opened with the year’s most consequential bilateral AI announcement. In Tokyo, Sam Altman and Masayoshi Son unveiled SB OpenAI Japan — a 50-50 joint venture — and signaled that the Stargate $500-billion data-center buildout would extend into Japan. SoftBank previewed Cristal, an enterprise AI offering for planning, marketing, and coding. US federal agencies including NASA, the Pentagon, the House, and the Navy began blocking DeepSeek over data-security risk. Trump paused the Canada and Mexico tariffs for 30 days after negotiations but kept the China 10% tariff on track for Feb 4. The Paris AI Action Summit sat one week away.
Top stories
- OpenAI and SoftBank unveil SB OpenAI Japan; Stargate expands to Japan. Altman and Son announced a 50-50 joint venture and signaled the $500B Stargate buildout would extend into Japan. SoftBank previewed Cristal — its enterprise AI offering for planning, marketing, and coding tasks. Altman highlighted Sunday’s Deep Research launch, available in Japan in Japanese. via Euronews · via SoftBank
- US federal agencies block DeepSeek. NASA, the Pentagon, Congress, and the Navy all began blocking employees from using the Chinese chatbot. The House issued a notice stating DeepSeek was “under review and currently unauthorized for official House use” and warned threat actors were already exploiting it to deliver malware. Concerns centered on data sent to PRC servers and obligations Chinese firms have to share data with intelligence services. via BankInfoSecurity
- Trump pauses Canada and Mexico tariffs 30 days; China 10% stays on track. After negotiations Feb 3 morning, Trump paused the 25% IEEPA tariffs on Canada and Mexico for 30 days but kept the 10% tariff on Chinese imports on track for Feb 4 — a market-moving move for AI hardware supply chains given how much chip packaging and assembly routes through North America. via Congress.gov CRS
- o3-mini week one: external experts prefer to o1-mini 56% of the time, 39% fewer major errors. First-week evaluations posted strong reception. o3-mini-medium matched o1 on math/code/science while being roughly 93% cheaper per token — framed across coverage as OpenAI’s direct response to DeepSeek pricing. via OpenAI
- Qwen ecosystem surpasses Llama derivatives — 90K+ open-source models. As of early February, Alibaba’s Qwen base counted over 90,000 derivative open-source models on community trackers, surpassing Meta’s Llama-derivative count. Qwen2.5-Max had launched Jan 29 and Qwen2.5-VL (3B/7B/32B/72B) in mid-January, reinforcing China’s open-weight surge heading into Paris. via Understanding AI
Who shipped
OpenAI and SoftBank‘s JV was the day’s anchor; SoftBank’s Cristal was the only adjacent product preview. Anthropic, Google DeepMind, Meta, xAI, and Mistral made no dated Monday launches.
Open-source pulse
The Qwen-derivative milestone marked the first time a Chinese open-weight family decisively outpaced Llama on community tracker counts. DeepSeek sat at the centre of cascading regulatory action — Italian block, Irish DPC inquiry, Texas state ban, US federal-agency lockouts — with no new release on Feb 3.
Money, infra & hardware
Mistral was finalizing its Le Chat consumer app (public launch Feb 6) and an enterprise partnership with Cisco, both timed to the Paris Summit. Macron would explicitly tell viewers on France 2 Feb 9 to “download Le Chat… rather than ChatGPT.” via Aragon Research
Quiet corners
The EU AI Act first business day under Article 5 / Article 4 began. Counsel notes circulated through the week summarizing the prohibition catalog and the AI-literacy obligation reach. via DLA Piper
By the numbers
- 50/50 — SB OpenAI Japan equity split
- 30 days — Canada / Mexico tariff pause window
- 10% — China tariff effective Feb 4
- 90K+ — Qwen-derivative open-source models
- 56% / 39% — o3-mini preference rate / major-error reduction vs o1-mini
- Most-mentioned lab: OpenAI
Compiled by AI Feed’s editor from verified web sources for 3 February 2025.