Skip to content
Source · Daily Brief

AI Daily Brief — 2 February 2026

Monday opens with two big enterprise plays. OpenAI and Snowflake announce a $200M multi-year partnership embedding OpenAI’s frontier models — including GPT-5.2 — natively across Snowflake Cortex AI and Snowflake Intelligence, reaching 12,600 customers on all three major clouds. xAI ships Grok Imagine 1.0 — 10-second 720p video with native audio and dialogue/ambience alignment, having generated 1.245 billion videos in the prior 30 days. The EU AI Act‘s one-year anniversary of prohibited-practices enforcement triggers the European Commission’s first mandatory review.

Top stories

  • OpenAI × Snowflake — $200M, multi-year. The deal makes OpenAI frontier models and ChatGPT Enterprise natively available inside Snowflake Cortex AI and Snowflake Intelligence on AWS, Azure and GCP. CEO Sridhar Ramaswamy frames it as bringing OpenAI to governed enterprise data; Canva and WHOOP are cited launch customers. Strategic significance: OpenAI distribution moves beyond the Azure-centric route. via OpenAI
  • xAI — Grok Imagine 1.0. “Biggest leap yet” — 10-second videos (up from 8s), 720p resolution, dramatically improved audio with dialogue/ambience alignment, multi-image-to-video (up to 7 reference images), and editing/extension. xAI says Imagine generated 1.245B videos in the prior 30 days. Day-0 rollout extends access beyond X to the xAI API and partner platforms. via xAI
  • MiniMax — M2.1 open-sourced. Sparse MoE with 204,800-token context, FP8 weights, Modified-MIT licence. Weights on Hugging Face; API on platform.minimax.io. Early SWE-bench Verified ~74.0; SWE-bench Multilingual ~72.5 — Microsoft Foundry frames the release as “a massive leap over M2 on software-engineering leaderboards.” via Hugging Face
  • EU AI Act — one-year prohibited-practices anniversary. The Act’s bans on unacceptable-risk AI (government social scoring, manipulative AI exploiting vulnerabilities, real-time remote biometric ID in public spaces, workplace/education emotion recognition, untargeted facial-image scraping) plus AI literacy obligations have been enforceable since 2 February 2025. Penalties run to €35M or 7% of global turnover. Full applicability of the broader Act follows on 2 August 2026. via European Commission

Money & infra

Cerebras momentum dominates the chips beat ahead of Tuesday’s official Series H announcement. Reuters and Tech Funding News report an IPO refile sought at up to $26.6B valuation with a $3.5B raise; the imminent $1B Series H at ~$23B post-money is Tiger Global-led with AMD, Benchmark, Fidelity, Atreides, Alpha Wave, Altimeter, Coatue and 1789 Capital alongside — four months after the Sep 2025 Series G ($1.1B at $8.1B) and two weeks after the $10B OpenAI compute deal (750 MW by 2028). Separately, Tom’s Hardware and MLQ News report the $500B Stargate JV between OpenAI, SoftBank and Oracle has stalled — no staff hired, no datacentres built — pushing OpenAI to pursue a bilateral 4.5 GW Oracle deal directly. Samsung begins mass production of HBM4 memory chips after Nvidia qualification samples in September; 2026 HBM4 output is already sold out. via Cerebras

Open-source pulse

Microsoft Foundry’s first Model Mondays of February features three trending Hugging Face models: Z.ai GLM-4.7-Flash (30B-total / 3B-active MoE, MIT-licensed), Meta SAM3 (unified image + video segmentation foundation model deployable via Foundry), and MiniMax-M2.1. Pre-Lunar New Year (Year of the Horse, 17 February) Chinese-lab release wave continues — Alibaba has committed ~CNY 3B (~$434M) to Qwen user-acquisition and Tencent ~CNY 1B (~$145M) for Yuanbao. via Microsoft Azure AI Foundry

Research

Monday arXiv cs.AI listings include Localizing and Correcting Errors for LLM-based Planners (Kumar & Cohen), Unintended Behavioural Misalignment from Domain-Specific Fine-Tuning (Mishra et al.), Game-Theoretic Reasoning Capabilities of LLMs Using Agentic Tools (Lin et al., accepted to ICLR 2026), and MHDash — a mental-health-aware AI-assistant benchmark. The International AI Safety Report 2026 is in final preparation for Tuesday release; METR opens its Feb-March frontier-risk pilot exercise this week with Anthropic, Google, Meta and OpenAI participating. via arXiv

Quiet corners

Anthropic, Google DeepMind, Meta AI, Microsoft AI and Thinking Machines Lab all silent on first-party shipping today. Anthropic’s Series G ($30B at $380B post-money) is being readied for 12 February.

By the numbers

  • OpenAI × Snowflake: $200M multi-year · 12,600 Snowflake customers across AWS / Azure / GCP
  • Grok Imagine 1.0: 10s · 720p · 7 reference images · 1.245B videos generated in prior 30 days
  • MiniMax M2.1: 204,800-token context · SWE-bench Verified ~74.0
  • EU AI Act fines: up to €35M or 7% of global turnover
  • Cerebras Series H imminent: $1B · ~$23B post-money · Tiger Global-led

Compiled by AI Feed’s editor from publicly reported announcements on 2 February 2026.