A house newsletter · daily · researched and written by the Kurzia AI
Monday 16 November 2026 · Issue 18 · 3 min read
Anthropic's IPO is 7x oversubscribed; Google doubles prices; Nvidia gets 10 days
Today's one thing
Techmeme (citing Bloomberg) · today
Anthropic IPO
Takeaway: about seven times oversubscribed
Anthropic's stock is about seven times oversubscribed, sources told Bloomberg, and the company with some early backers aims to raise up to $90.4 billion from the listing plus a private side deal — set to be the biggest IPO on record. For any company buying AI, the significance is that a main supplier is about to answer to public investors with a growth number to defend every quarter. That pressure reaches invoices: pricing, packaging and sales tactics rarely survive a record listing intact. Use the product; don't sign long contracts that assume today's prices hold.
Your supplier's incentives change the day it lists, and defending a record valuation makes revenue growth the only acceptable outcome. The figures come from sources rather than the company, so the final raise may land below the $90.4 billion ceiling.
What is real, what is not
WSJ's Nvidia financing report — The Wall Street Journal reported Nvidia helped finance AI startups that then bought its chips, making demand look tighter than it was. Nothing is proven, and the weekend's Hacker News thread splits two ways: ordinary vendor finance to keep its customer base diverse, or a firm propping up sales as its biggest buyers design their own chips. If the shortage you were quoted on AI compute had a seller behind it, the premium deserved more scepticism than it got. The thread drew 646 comments, some citing Huang's own warning of an oversupply of committed chips and memory. Hacker News (discussing a WSJ report)
“He's a better businessperson than all of you”
Wharton-Gallup Gen Z survey — A Wharton-led survey with Gallup of nearly 2,500 Americans aged 18–28 found 79% think AI makes people lazier and 62% think it makes people less smart — yet that generation leads workplace AI adoption and nearly one in five fears being displaced by it. Bans will not stick with your heaviest users, and their scepticism is about colleagues, not themselves; policy should channel the ambivalence, not deny it. Frequent users worried less, suggesting the anxiety fades with use. Fortune
“You should outsource the crap, not the craft.”
Money and the market
Gemini API prices — Google's two newest Gemini models cost $2 per million input tokens and $10 per million output until 30 November, then double to $4 and $20 on 1 December; the cheaper Flash models double on 1 January. Any AI budget built on this month's rates rests on a discount with an expiry date; price your projects on the December number. The article's worked example: an agent using 50M input and 10M output tokens monthly pays $200 in November and $400 from December. On the free tier, Google says human reviewers may read what is sent. gravity.fast
OpenAI robotics — OpenAI has broken up its stand-alone robotics group and moved the researchers onto its core teams, betting its general models can steer robots without dedicated robotics software. The move also fits a cost squeeze: reported operating losses of $10.9 billion last quarter, with a funding round and a possible early-2027 listing ahead. The best-funded lab in AI is cutting side bets to defend its core — a map of where the big money thinks value concentrates. Altman still targets a first public demonstration of a humanoid robot in 2027. Fortune
Risk the board will raise
Nvidia's ten-day deadline — Treasury Secretary Scott Bessent says Nvidia's board and its outside lawyers have ten days to explain whether the company pushed apparently illegal chip sales to China, and he wants the next chief executive hired from outside. Nvidia has opened an internal investigation with the law firm Wachtell, Lipton. Supplier political risk now moves faster than contracts; check your vendors' China exposure before the board asks. A Singapore firm, Megaspeed, is suspected of smuggling $2 billion of restricted chips into China, and Nvidia lost about $1 trillion of market value in a single week. Bloomberg (via Techmeme)
Samsung, SK Hynix and Micron — A federal judge refused to dismiss a class action accusing the three memory makers of conspiring to cut supply of ordinary computer memory while shifting factories to AI-grade memory. The complaint says they control over 91% of the memory market and that conventional prices rose 171.8% in a year. Nothing is proven, but your 2027 hardware budgets should not assume memory prices fall back while the case runs. The companies must formally answer by 7 December 2026; the first case-management hearing is set for 12 January 2027. LitNews
The big AI stories today
Jensen Huang — has discussed returning as Nvidia chief, with allies lobbying the board and the White House for him, the FT reports — unconfirmed, but the fight is not settled. Financial Times (via Techmeme)
Nvidia in Southeast Asia — The WSJ reports Huang built a multibillion-dollar business letting Chinese firms use Nvidia chips via Southeast Asian data centres, through a sanctions gap. Wall Street Journal (via Techmeme)
Project Zero — Google's bug-hunting team will disclose flaws at 90 days, then 60 by 2027 and 30 by 2028, because AI now finds bugs far faster. Project Zero (Google)
One takeaway
Techmeme (citing Bloomberg) · today
Today's AI price list is a launch offer
The pattern across today's issue: the biggest vendors are either chasing a public-market growth number or quietly ending their launch discounts — and the chip scarcity your budgets were built on may have had a seller's fingerprints on it. One question for your team this week: which of our AI costs are priced at introductory rates, and what is that bill at list price? Decide before the invoices decide for you.