🗞️ AI Daily Briefing — 2026-04-29

🔥 Top Story

OpenAI misses revenue and user targets ahead of its IPO — and the market felt it. A Wall Street Journal report revealed that ChatGPT fell short of its internal goal of one billion weekly active users, and monthly revenue missed targets multiple times this year as Gemini and Claude ate into its share. CFO Sarah Friar has raised doubts about OpenAI’s IPO readiness. SoftBank shares dropped 9.9% in Tokyo — their worst day in six months — and chip stocks sold off broadly. This is the first real crack in the narrative that OpenAI’s dominance is inevitable. (CNBC) (TechCrunch)

🚀 Model & Research News

  • GPT-5.5 officially launched last week, but the afterglow is fading: OpenAI shipped GPT-5.5 on April 23 with autonomous reasoning and deep computer-use capabilities. Sam Altman called it “the last major milestone before AGI.” But the revenue miss story has overshadowed the launch — the model is impressive, but it hasn’t reversed ChatGPT’s user growth slowdown. (CNBC) (TechCrunch)
  • Sergey Brin assembles DeepMind “strike team” to close the coding gap with Claude: In an internal memo, Brin wrote: “To win the final sprint, we must urgently bridge the gap in agentic execution.” The team is led by Sebastian Borgeaud (former pre-training lead) and focused on long-horizon coding tasks. Google currently uses AI for ~50% of its code; Anthropic claims nearly 100%. (The Decoder) (Sherwood News)
  • Claude Opus 4.7 holds #1 on LMArena at 1504 Elo: Grok 4.20 Beta1 has climbed to #4 globally (1491 Elo), surpassing GPT-5.4. Claude 4.7, Gemini 3.1 Pro Preview, and GPT-5.4 are tied at 57 on the Artificial Analysis Intelligence Index. The leaderboard now tracks 351 models with 5.97M votes. (Arena.ai)
  • TurboQuant (Google) gains traction at ICLR 2026: The vector quantization algorithm combining PolarQuant rotation and Johnson-Lindenstrauss compression is seeing real adoption for deploying large context-window models efficiently. Practical stuff for anyone running inference at scale. (ScienceDaily)

🛠️ Tools & Developer Updates

  • OpenAI rebuilt custom GPTs into shared agents: GPTs are being repositioned as agents that live inside Slack and Salesforce, not just in ChatGPT. This is OpenAI’s clearest move yet into enterprise workflow automation territory — directly competing with LangChain and similar orchestration tools. (Asanify)
  • Critical RCE vulnerability in Hugging Face’s LeRobot: Security researchers disclosed a flaw in the open-source robotics platform that allows unauthenticated remote code execution. Exploited within 13 hours of disclosure. If you’re running LeRobot, patch immediately. (The Hacker News)
  • Hannover Messe 2026 showcases industrial humanoid robots: Multiple humanoid robots demoed for assembly, logistics, and factory-floor operations. The physical AI wave Jim Fan has been evangelizing is hitting real industrial deployments. (NVIDIA Blog)

💰 Funding & Business

  • David Silver’s Ineffable Intelligence raises record $1.1B seed at $5.1B valuation: The AlphaGo/AlphaZero creator left DeepMind to build a “superlearner” that discovers knowledge without human data, using pure reinforcement learning. Sequoia and Lightspeed co-led; Nvidia, Google, and the UK Sovereign AI Fund participated. Largest European seed round ever. (TechCrunch) (CNBC)
  • Google commits up to $40B in Anthropic: $10B cash now at a $350B valuation, plus $30B contingent on milestones. Google Cloud delivering 5 GW of compute over five years. Combined with Amazon’s $5B (plus up to $20B future), Anthropic is now the best-capitalized AI lab for infrastructure. Revenue has hit $30B annualized, up from $1B at end of 2024. (TechCrunch) (CNBC)
  • China blocks Meta’s $2B acquisition of Manus: Beijing’s NDRC ordered the deal cancelled, citing export control and national security concerns. Meta had already integrated Manus into its systems, making the unwind messy. Two Manus co-founders are banned from leaving China. A sharp escalation in the US-China AI cold war. (CNBC) (Fortune)
  • Avoca hits $1B valuation with AI voice agents for plumbers: The YC-backed startup raised $125M+ across Seed through Series B (Kleiner Perkins, General Catalyst, Meritech). On track to book $1B in jobs this year. Vertical AI agents for blue-collar services are now a unicorn category. (Fortune)
  • Big Tech brain drain accelerating: VCs have poured $18.8B into AI startups founded since 2025 (Dealroom data). Notable: Tim Rocktaschel (ex-DeepMind) seeking $1B for Recursive Superintelligence; Yann LeCun’s Ami Labs raised $1B in March; Humans& (ex-Anthropic/xAI founders) raised $480M. (CNBC)

🐦 Notable from the Timeline

  • @sama called GPT-5.5 “the last major milestone before AGI” — bold framing, but the revenue miss report the same week undercut the moment. He and CFO Friar issued a joint statement: “We are totally aligned on buying as much compute as we can.”
  • Google vs. Anthropic on Pentagon ethics is the defining split in AI safety right now. Google signed a classified deal letting the Pentagon use Gemini for “any lawful government purpose” with adjustable safety filters. Anthropic was blacklisted for refusing to remove guardrails on autonomous weapons and surveillance — then a federal judge blocked the blacklisting. Meanwhile, the Pentagon was already using Claude to select strike targets in Iran. (Bloomberg) (CBS News)
  • 600+ Google employees signed a letter urging Sundar Pichai to reject the classified Pentagon AI deal. Echoes of Project Maven, but this time Google signed anyway.
  • Q1 2026 venture funding hit $300B globally — an all-time record. AI accounted for $242B (80%). The money firehose shows no signs of slowing. (Crunchbase)

📊 Benchmark Watch

  • LMArena (Arena.ai): Claude Opus 4.7 still #1 at 1504 Elo. Grok 4.20 Beta1 is the surprise climber at #4 (1491 Elo). GPT-5.5 benchmarks haven’t fully propagated to the leaderboard yet — watch for movement this week.
  • SWE-bench Verified: Claude Opus 4.7 leads at 82%. DeepSeek V4 was reported at 81% pre-launch — if that holds, it’ll be the most cost-efficient model at that tier ($0.30/MTok).
  • Intelligence Index tie: Claude 4.7, Gemini 3.1 Pro Preview, and GPT-5.4 are deadlocked at 57 on the Artificial Analysis index. The frontier is genuinely crowded.

🎙️ Podcast Highlights

  • TBPN is now owned by OpenAI. The daily tech show hosted by John Coogan and Jordi Hays was acquired in a nine-figure deal on April 2. It now reports to OpenAI’s Chris Lehane. They claim editorial independence is “explicitly protected” — draw your own conclusions. The Hacker News thread and Slate both called it a PR play. (OpenAI Blog) (NPR)

🔗 Worth Reading

  • “Why Google Formed a Strike Team Because of Claude — The Flywheel Explanation” — Deep dive into why Anthropic’s internal AI-coding flywheel (AI writes code → better AI → more code) is what really spooked Brin. (Revolution in AI)
  • “Google Is Getting a Screaming Bargain on Its $40B Anthropic Investment” — Motley Fool breaks down why, at $30B annualized revenue, Google’s valuation terms are actually cheap. (Motley Fool)
  • Stanford AI Index 2026 Report — 12 key takeaways on the state of AI research, funding, and policy. The definitive annual snapshot. (Stanford HAI)