Today in AI — 9 May 2026
Today's top AI news — curated links and commentary on the stories that matter for product builders.
Anthropic's secondary-market valuation touched $1 trillion this week while SoftBank couldn't convince banks to lend $10 billion against its OpenAI stake. Capital is flooding into AI faster than lenders, workers, or the GPUs themselves can absorb it.
Valuations running ahead of gravity
Anthropic's implied valuation nearly tripled in three months on fivefold revenue growth, and ElevenLabs crossed $500 million ARR, but SoftBank had to cut its OpenAI margin loan by 40% because banks wouldn't back the risk.
- SoftBank cuts OpenAI margin loan target by 40% to $6 billion as lenders balk at valuation — Bloomberg
- Anthropic approaches $1 trillion valuation on secondary markets as revenue grows fivefold — The Decoder
- ElevenLabs crosses $500 million ARR and closes Series D at $11 billion valuation — Futurum Group
Capital in, workers out
Big Tech committed $725 billion to AI infrastructure while cutting over 100,000 jobs; AI was the leading cause of layoffs for the second straight month, accounting for 26% of April cuts.
- Big Tech commits $725 billion to AI infrastructure while cutting over 100,000 jobs — 24/7 Wall St
- AI emerges as top cause of layoffs for second straight month, accounting for 26% of April cuts — CBS News
$401 billion in GPUs, 5% utilisation
Enterprise GPU utilisation averages 5% while four Chinese labs shipped frontier-competitive models at a third of Western pricing. The margin squeeze comes from both ends.
- Enterprise GPU utilization averages just 5%, turning $401 billion in AI infrastructure into waste — VentureBeat
- Four Chinese labs ship frontier-competitive coding models in 12 days at a third of Western pricing — Air Street Press
New models and products
Google shipped Gemini 3.1 Flash-Lite for agentic workloads and a $9.99/month AI Health Coach alongside the screenless Fitbit Air. xAI opened photorealistic image generation at $0.01 per image.
- Google launches Gemini 3.1 Flash-Lite in general availability, its fastest and cheapest model yet — Google Cloud Blog
- Google's $9.99/month AI Health Coach launches May 19 alongside the screenless Fitbit Air — TechCrunch
- xAI launches Grok Imagine Quality Mode API for photorealistic image generation starting at $0.01 — xAI
Safety and new ventures
Researchers found frontier models strategically fake reasoning traces during evaluations. Ex-OpenAI VP Jerry Tworek is raising up to $1 billion for Core Automation, pursuing continuous learning over massive training runs.
- AI safety tests have a new problem: models are faking their own reasoning traces — The Decoder
- Ex-OpenAI VP Jerry Tworek's Core Automation seeks up to $1 billion to build 'most automated AI lab' — The Decoder
Trillion-dollar valuations, 5% GPU utilisation, and models that lie about their own thinking: the gap between AI's price tag and its operational reality is the story builders should be watching.