Skip to content
Source · Daily Brief

AI Daily Brief — 06 February 2025

Thursday delivered the year’s first sovereign-AI consumer launch and Amazon’s Q4 capex shock. Mistral released Le Chat for iOS and Android, headlining a Cerebras-powered “flash answers” mode at up to 1,000 tokens per second running Mistral Large 2 — with a $14.99/month Pro tier undercutting ChatGPT Plus and Claude Pro. President Macron endorsed it on French TV, telling viewers to “download Le Chat instead of ChatGPT.” Amazon reported AWS Q4 at $28.8 billion (+19% YoY) and signaled roughly $100 billion of 2025 capex — the largest hyperscaler AI budget at the time — while deepening its Anthropic partnership: Anthropic will use Trainium2 to train next-generation models. Hugging Face shipped Open Deep Research within 24 hours of OpenAI’s launch. NVIDIA released DeepSeek-R1 as a preview NIM microservice. ChatGPT search opened to all users with no signup required.

Top stories

  • Mistral launches Le Chat mobile (iOS/Android) with Cerebras 1,000 tok/s “flash answers.” Free app plus $14.99/month Pro tier and $29.99/user/month Teams tier (enterprise tier in private preview). The headline feature runs Mistral Large 2 (123B) on Cerebras inference at up to 1,000 words per second. Pricing undercuts ChatGPT Plus and Claude Pro ($20 each). Macron endorsed it the same day on French TV: “Download Le Chat, which is made by Mistral, rather than ChatGPT by OpenAI.” via TechCrunch
  • Amazon Q4: AWS $28.8B (+19% YoY), ~$100B 2025 capex signaled, Anthropic on Trainium2. AWS operating income $10.6B (vs $7.2B Q4 2023). $26.3B Q4 capex, $83B full-year 2024; CEO Andy Jassy called that “reasonably representative” of 2025 — implying $100B+. Jassy flagged AI chip, power, and motherboard shortages as constraints. Anthropic will use Amazon’s Trainium2 silicon to train next-generation models. via Futurum
  • Hugging Face ships Open Deep Research within 24 hours. An open-source agent framework that browses the web and writes research reports as a same-week response to OpenAI’s Feb 2 Deep Research launch. Scored 54% on GAIA versus OpenAI Deep Research’s 67.36%, fully open and runnable on any OpenAI-API-compatible model. Set the template for a wave of open agent frameworks. via Hugging Face
  • DeepSeek-R1 lands on NVIDIA NIM; sweeps Chinese cloud deployments. NVIDIA released R1 671B as a preview NIM microservice — turnkey for enterprise NVIDIA-hardware deployment. China Telecom’s Tianyi Cloud “Xirang” platform became the first operator-level Chinese cloud to complete full-stack domestic-silicon inference deployment of R1. NetEase Youdao announced full R1 integration; Zhewen Interconnect completed localized deployment before Spring Festival. via CFR
  • ChatGPT search opens to everyone — no signup required. OpenAI dropped the account wall on what had been a Plus-only and then Free-logged-in feature. ChatGPT search returns timely answers with source links, accessible at chatgpt.com plus desktop and mobile apps. Direct frictionless competition with Google web search the same week as Alphabet’s Q4 print. via OpenAI

Who shipped

Mistral, OpenAI, and Hugging Face all shipped product on the same day. NVIDIA shipped DeepSeek-R1 as a NIM. Anthropic, Google DeepMind (still post-Wednesday Gemini push), Meta, and xAI made no dated launches.

Open-source pulse

Open Deep Research’s 24-hour turnaround was the day’s open-source signal. The Chinese R1 deployment sweep — operator-level cloud, edutech, and enterprise — cemented R1 as the de facto Chinese reasoning model two weeks after release.

Money, infra & hardware

Amazon’s $100B capex pile alongside Alphabet’s roughly $75B and Meta’s $60-65B anchored a hyperscaler arms race north of $300B for 2025 — exactly the spending Jevons-paradox bulls had argued DeepSeek would not derail. Jassy’s note on AI chip, power, and motherboard shortages signalled the binding constraints were physical, not financial.

Quiet corners

Paris AI Action Summit week opened formally with nearly 100 satellite events worldwide. Institut Polytechnique de Paris hosted an international AI-and-society conference Feb 6-7 chaired by Michael Jordan (UC Berkeley/Inria) and Eric Moulines (École Polytechnique); the IASEAI Conference convened at the OECD.

By the numbers

  • 1,000 tok/s — Le Chat flash answers (Cerebras + Mistral Large 2)
  • $14.99 / $29.99 — Le Chat Pro / Teams pricing
  • $28.8B / +19% — AWS Q4 revenue and growth
  • ~$100B — Amazon signaled 2025 capex
  • 54% / 67.36% — Open Deep Research / OpenAI Deep Research GAIA scores
  • Most-mentioned company: Amazon / Mistral

Compiled by AI Feed’s editor from verified web sources for 6 February 2025.