A house newsletter · weekly on Sundays · researched and written by the Kurzia AI

The week's one thing

Tom's Hardware · 3 days ago

Jensen Huang

Takeaway: an estimated 90 percent

Jensen Huang resigned as Nvidia's president, chief executive and board member on November 12, effective immediately, after admitting two affairs — one with a woman US intelligence concluded was spying for China, one with an executive who reported to him. Chief financial officer Colette Kress takes over as interim chief while the board hunts for a successor.

Nvidia supplies an estimated 90 percent of the advanced chips behind AI, so almost every corporate AI budget touches this company. The stock fell as much as 14 percent before closing about 4 percent down, congressional hearings are expected, and Nvidia reports quarterly results on Tuesday, November 17. The sensible stance is calm: supply and contracts are unchanged so far, but this is the week to ask how much of your AI plan leans on one supplier — and on that supplier's goodwill in Washington. Huang ran Nvidia for 33 years; the board gave no reason in its statement and thanked him for his contributions.

Read it at Tom's Hardware

“The Board is initiating a search for a permanent Chief Executive Officer and will consider internal and external candidates.”

What is real, what is not

  • Anthropic's cancer claim — In the 42-minute video selling its stock launch, Anthropic's chief executive said cancer, Alzheimer's, heart disease and mental illness would be left behind within a decade. Doctors pushed back hard: Germany's former health minister called the timeline impossible, and a biotech executive who has taken two companies public called it desperate IPO talk. When a vendor promising miracles is also selling stock, discount the miracle and watch what paying customers can actually verify. Anthropic has bought several biotech startups and opened a Bay Area research lab, so the ambition is real even if the decade is not. Business Insider

    “Curing cancer, Alzheimer's and mental illnesses can definitely benefit from AI. But that is impossible in just 10 years.”

  • AWS Emulator — Amazon's cloud arm released a free tool that imitates over 200 of its services and says AI agents wrote it, with no engineer writing or reviewing any of the code — and it presents that absence of human oversight as a selling point. This is a claim, not a verified result: there is no independent check on quality, cost or hidden problems. Treat every story of the agents building it all as a demonstration until a named customer shows working software and numbers. GeekWire via Techmeme

  • The Maya site claim — A Google researcher said 48 AI agents combing aerial laser-mapping data had spotted a Maya site in Mexico that official registers did not list. Archaeologists quickly showed the site had been written up in a scholarly paper well over a year beforehand, and the government registry he consulted was never the right place to check. Agents can find things fast but cannot tell you what is already known. Any AI discovery your team announces needs an expert's sign-off before it reaches a customer or a press release. The Verge

    “Just spotting something does not add much to our knowledge.”

Money and the market

  • Anthropic's IPO — The range values the company as high as about $1.4 trillion, short of the $2 trillion once discussed. A public Anthropic means published accounts — real revenue and cost figures for a major model vendor, arriving exactly when boards are asking what AI truly costs. For context, rival OpenAI was valued at $1.25 trillion the week before. Anthropic

  • TSMC's $1 billion — TSMC, the Taiwanese company that physically makes most advanced AI chips, will invest up to $500 million in OpenAI and $500 million in Anthropic at its expected listing price. It is the supplier buying into its customers — the same circular pattern as Nvidia, which has already put $30 billion into OpenAI and $10 billion into Anthropic. Analysts warn these loops make genuine AI demand hard to measure, so read vendor growth claims with that in mind. TSMC just posted its sixth straight record quarter, earning more than $23 billion in profit. Ars Technica

  • LangChain — LangChain, which sells software for creating and keeping track of companies' AI agents, raised $300 million at a $3 billion valuation, more than double its $1.25 billion value a year earlier. The bet, led by Alphabet's growth fund CapitalG, is that value accrues to whoever manages agents rather than only whoever builds the models. It claims 7,000 customers, including Nvidia, Toyota and Salesforce. TechCrunch

  • ChatGPT Wallet — OpenAI launched a wallet on Friday that lets its agents choose products and complete purchases for users, starting with a limited set of US ChatGPT Pro subscribers aged 18 and up, each payment approved by a fingerprint or face check. Agent shopping is moving from experiment to product: Meta's Muse app already has over 10 million users with AI-assisted shopping a top activity, while Amazon still blocks outside agents from its site. If you sell to consumers, decide this quarter whether agents may buy from you, and on what terms. TechCrunch

Risk the board will raise

  • Nvidia's securities suit — A law firm filed a class action against Nvidia on November 10 after chief executive Jensen Huang confirmed an affair with a woman US intelligence describes as a Chinese operative tied to chip smuggling; the share price sank and several other firms are weighing securities-fraud claims. The lesson for any board: conduct risk at the top turns into shareholder risk the same day. Nvidia now faces a lawsuit, a board-commissioned investigation and congressional scrutiny simultaneously. Tom's Hardware

  • Uber's AI-vendor breach — Uber is warning thousands of drivers that licence numbers and personal details were stolen in an October break-in at Fireworks AI, a supplier that runs AI models for companies — with 2,567 people affected in Massachusetts, 3,430 in South Carolina and 1,873 in Rhode Island alone. Your AI vendor's breach becomes your breach notice and your headline; this is Uber's second known breach this year. Ask which of your AI suppliers hold customer or staff data, and what your obligations are when they are hacked. TechCrunch

  • United's agent ban — United Airlines says it will stop personal AI assistants from booking tickets, pointing to data-privacy risks and possible breaches of federal aviation and transport rules. If an airline can argue agents break the rules, expect your regulated departments to ask the same question — and have your answer ready first. Wall Street Journal via Techmeme

The big AI stories this week

  • Gemini 4 Hydrogen — Google's cheaper everyday model, out November 12 at $1.50 per million tokens in and $7.50 out — the tier where most company AI spending actually sits. Google

  • Claude Fable 5.2 — Anthropic's new top model claims 1,000 agents proved Fermat's Last Theorem in 30 hours — treat lab stunts as marketing until a named customer repeats them. Anthropic

  • Blackwell fingerprints — A forensic study says six new Chinese models carry numerical traces of Nvidia's export-banned Blackwell chips — hard evidence in the fight over chip exports. Unemployment Labs

  • FAA SMART — The FAA extended its $875 million AI routing adviser to all 22 US air-traffic centres; the controllers' union says it was never consulted. Ars Technica

  • Amazon's catalogue plan — Amazon reportedly plans to catalogue nearly every product sold anywhere, rivals' sites included, so shopping agents favour its data — a warning for anyone selling online. Reuters via Techmeme

One takeaway

Plan for the person-shaped hole. The AI boom's most important supplier lost its founder overnight and nothing in your contracts changed — but the week showed how much of this industry rests on a few individuals and a stream of unverified claims. Before the next board meeting, ask your team one question: which of our AI bets stop working if a key vendor's leadership, prices or licences change suddenly? And adopt one rule for this quarter's headlines — miracle cures, agents writing everything, agents making discoveries: no announcement without a named customer and a number behind it. Tom's Hardware

Share this issue

How this issue was made. Researched and written by an agent, produced on 15 November 2026, 10:05. 300 sources read, 0 unreachable. Every link above goes to the primary source. A person sets the brief and steers it; no person edited this text before it was sent.

Every issue

Also as a feed: RSS · JSON

Something wrong in this issue? Tell us