Anthropic wants a $1.4 trillion IPO
The big story today is money, not models. Anthropic's IPO filing shows how much compute frontier labs have already committed to, and reports that Nvidia financed its own customers show how fragile some of that demand may be. If you buy compute, the question is not only what it costs today but whether your provider can keep those terms when its own funding shifts. Meanwhile, Thinking Machines just released an open model you can use commercially, which gives you a real option to weigh against closed models.
- Anthropic's $1.4 trillion IPO filing
- Thinking Machines' free Inkling 2 Small model
- Nvidia's financing of its own customers
- Quick hits: hiring, robots, analytics, essays
- Lounge AI Index and today's new members
Top stories
All news →What happened: Anthropic's amended prospectus proposes selling 825 million shares at $70 to $78 each, though the offering remains subject to completion.
- Some shares come from selling stockholders, and TechCrunch says the price range implies a valuation of about $1.404 trillion.
- The company reports about $900.9 billion in committed spending, nearly all for compute and infrastructure.
- About $70.7 billion falls due within the next 12 months, and much of the rest runs through 2045.
Why it matters: Frontier labs are locking in multi-year compute contracts sized in the hundreds of billions. If you buy capacity from the same providers, watch how those obligations shape pricing and who gets allocation first.
What happened: Thinking Machines Lab made the open weights for Inkling 2 Small available on November 12, one week after its closed-weight debut on Tinker.
- The Apache 2.0 license allows commercial use, modification and redistribution, and the model is a sparse mixture-of-experts design with 594 billion total parameters, 17 billion active per token.
- On the Artificial Analysis Intelligence Index it scores 31, up from 13 for Inkling Small, which the lab says puts it roughly level with Gemini 3.8 Flash.
- It placed fourth overall and first among open-weight models on DesignArena with an Elo of 1,376, but the main checkpoint needs at least 500GB of VRAM, such as two B300 or four H200 GPUs.
Why it matters: A permissive license means you can build it into a product you sell and change it to fit your needs. The hardware bar is high, so treat it as a serious option for teams with the infrastructure, not a cheap default.
What happened: A Wall Street Journal profile of interim CEO Colette Kress says Nvidia financed customers to keep demand for its chips strong.
- Nvidia reportedly pushed for more such arrangements in mid-to-late 2026 to absorb an oversupply of TSMC wafers and support GPU prices.
- The largest case involved an OpenAI data center in Ohio expected to use about $300 billion of Nvidia hardware, where a $250 billion guarantee was considered and then roughly halved in an agreement signed August 17. Reuters cites $120 billion.
- Nvidia shares fell 5% after the July 26 report on the guarantee plan, and Nvidia has since paused some financing and added board-level reviews. No Nvidia statement or filing confirms a deliberate scarcity strategy.
Why it matters: When a chip supplier also funds its buyers, demand can look stronger than it is, and borrowers carry that risk. If your cloud provider is also a funder, ask what happens to your pricing if that support is pulled back.
Quick hits
- Coverage of the same WSJ reporting says some funded small clouds could not afford the GPUs they were expected to buy. techpowerup.com →
- HiringNow's agent rechecks 247,000 career sites every five minutes and applies only after you approve. Product Hunt →
- Gemini Robotics 2.5 reports 96% success in a Mercedes-Benz parts-kitting pilot with Apptronik's Apollo 2 humanoid. Product Hunt →
- Dex says adding declarations to analytics models lifts an agent's ADE-bench score from 73.3% to 80% with Claude Sonnet 5.5. Product Hunt →
- An essay argues AI awe reflects tech leaders' overconfidence, using Bill Gates as its main case study. netmeister.org →


