Today AI has raised over $200M. Is it China's answer to Meta's Muse? The bigger story isn't just funding. It's the shift from AI that waits for prompts to AI that remembers, anticipates, and acts. But can that become lasting value?

GPU rent is the crude. AI tokens are the gasoline. But who hedges the margin in between? For the same AI model, prices can vary up to 4× across providers. GPU futures alone won't solve that risk. So who writes the grade? Hedging Intelligence | Part 2 📷

One AI provider sells 25 billion tokens a day. Yet there's no market to hedge its token revenue. Token prices were expected to keep falling. Now they're moving both ways. If AI tokens become a commodity, who takes the other side? Hedging Intelligence | Part 1 👇

Zuckerberg was ready to bet $100–200B to Muse before the model existed🤯. And Muse Wonder shipped TODAY. Takeaways from @collision sit-down with Zuck & @alexandr_wang: 1. Muse was trained for personal agents, not coding, the opposite of where the industry was racing. 2. Memory is the bet: #Muse Wonder is built to remember what you told it weeks or months ago. 3.Ignore public benchmarks. @alexandr_wang: chasing them outsources your product thinking to whoever made a viral benchmark. 4. Compute costs ~$50–60B upfront per gigawatt and ~$15B/yr to run. If Muse fails, the compute can be resold or redirected. 5. Talent density over headcount. A day earlier on @davidsenra’s podcast, @alexandr_wang split businesses into two kinds: diamond mines (most bets fail, a few return 1,000,000x) and skyscrapers (you grind out a few % of efficiency at a time). Put the two conversations together and Meta’s strategy is clear: a diamond bet on top of a skyscraper. A small team chases the asymmetric upside, while hard assets cap the downside. Up the AI stack, from compute to agent infra to agent apps, the mix shifts. Know which game you’re in before you pick your investors. #IOSFVC #MuseWonder #Meta

We co-organized the Workshop on Efficient Reasoning at #COLM2026 in San Francisco this Friday 🧠 A full day on making reasoning models faster, cheaper, and deployable at scale, from test-time scaling and RL to agents, world models, VLMs, and KV cache efficiency. Grateful to our speakers: Yuandong Tian (ex-Meta), Niklas Muennighoff (Stanford), Jiajun Wu (Stanford), Kai-Wei Chang (UCLA), Sewon Min (UC Berkeley), Junjie Hu (Wisconsin) and many more! Thanks to everyone who joined in person and online 🙌 🔗 wdlctc.github.io/efficient-r…

What a night. Thanks to @glazecl for hosting a dinner with some of the sharpest AI builders and researchers we know! At the table: · Head of GTM at @fal · Engineering Manager at @nvidia · First author of the Seedance prototype · @McKinsey AI · @Meta AI scientist working on ads · One of @togethercompute 's first 20 hires, and Meta Llama 4 agent training lead · Founder of @novita_labs, a top-3 provider on @OpenRouter · UCSB full professor specializing in LLM training, @amazon AI Lab Scholar · Pioneer in GPU circuit simulation The people training models, running inference and building the silicon, all at one table. More gatherings soon!