Skip to content
Source · Daily Brief

AI Daily Brief — 28 January 2025

Tuesday turned into the day Washington picked a side. White House AI czar David Sacks went on Fox News with “substantial evidence” that DeepSeek had distilled OpenAI models; Trump called R1 a “wake-up call” at the House GOP retreat in Doral; the US Navy banned DeepSeek service-wide and the Pentagon’s DISA blocked it on government networks. NVIDIA rebounded 8.9%, recovering roughly $260 billion of Monday’s $589 billion loss. Sam Altman broke OpenAI’s silence, calling R1 “impressive” and promising to “pull up some releases.” Italy’s Garante opened a formal data-privacy probe — the first major Western regulatory action against the Chinese lab.

Top stories

  • White House AI czar David Sacks alleges DeepSeek distilled OpenAI models. On Fox News: “substantial evidence” R1 was trained on outputs from OpenAI models, estimating the US lead at three to six months. Sacks amplified a SemiAnalysis report from Dylan Patel arguing DeepSeek’s real compute-cluster spend was over $1 billion — far above the widely-cited ~$6M training-cost figure. via TechCrunch · via Fox Business
  • Sam Altman breaks silence: R1 “impressive,” OpenAI will “pull up some releases.” Altman on X: “deepseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price. we will obviously deliver much better models and also it’s legit invigorating to have a new competitor! we will pull up some releases.” Follow-ups doubled down that more compute is still essential. via Axios
  • US Navy bans DeepSeek; Pentagon’s DISA blocks it across government networks. The Navy issued a service-wide directive telling members to avoid DeepSeek “in any capacity”; DISA began blocking access on Pentagon IT after reports DoD employees had used it for ~2 days. The White House confirmed a National Security Council review was underway. via CNBC
  • Italy’s Garante opens DeepSeek data-privacy inquiry. The first major Western regulatory action: the Italian Data Protection Authority demanded within 20 days details on personal-data collection, training-data sources, storage locations, and legal basis. The probe would escalate days later into a full processing ban on Italian users. via Hunton
  • NVIDIA rebounds 8.9% the day after the historic crash. JPMorgan and Citi analysts publicly defended NVDA, arguing DeepSeek’s sub-$6M training figure was understated because it leaned on distillation from Meta’s open-source Llama. No major Wall Street analyst downgraded NVIDIA after the sell-off. The bounce recovered roughly $260B of Monday’s $589B loss; the stock still finished the week down ~8%. via CNBC

Who shipped

No new lab releases. Anthropic stayed publicly silent — Dario Amodei’s “On DeepSeek and Export Controls” essay would land Wednesday. Google DeepMind, Meta, xAI, and Mistral made no dated announcements. Alibaba/Qwen‘s Sunday Qwen2.5-VL release continued landing in Hugging Face download charts; Qwen2.5-Max was a day away.

Open-source pulse

DeepSeek hit #1 on the US Google Play Store, completing a sweep of both major mobile platforms — over 1.2M Play Store downloads and over 1.9M App Store downloads since mid-January. The lab continued limiting new sign-ups after the prior day’s cyberattack, narrowing registrations to mainland-China phone numbers, email, or Google login. Existing users could still log in. via TechCrunch

Money, infra & hardware

Apple sat as the most-valuable US company entering Tuesday’s session with Microsoft third; NVIDIA had passed Apple on Jan 21 to take the crown before Monday’s plunge erased it. Cathie Wood told Bloomberg DeepSeek’s emergence is positive and would accelerate AI cost reductions — inference costs already falling 85-95% per year — with share shifting toward platforms and inference-focused companies and away from pure SaaS. via Bloomberg

Quiet corners

Trump told the House GOP retreat in Doral, Florida: “It should be a wake-up call for our industries that we need to be laser-focused on competing to win” — but framed cheaper Chinese AI as “a positive” — “instead of spending billions and billions, you’ll spend less.” OpenAI and US government officials separately held discussions about DeepSeek’s development practices, with OpenAI saying it had evidence Chinese rivals used its model outputs to train competing systems. via The Hill · via PYMNTS

By the numbers

  • +8.9% — NVIDIA Tuesday rebound (recovering ~$260B of Monday’s $589B loss)
  • 3-6 months — Sacks’ estimate of US lead over China
  • >$1B — SemiAnalysis estimate of DeepSeek’s true compute spend
  • 1.9M / 1.2M — DeepSeek US App Store / Play Store downloads since launch
  • 20 days — Italy Garante response deadline
  • Most-mentioned actor: White House

Compiled by AI Feed’s editor from verified web sources for 28 January 2025.