↵ select ↓ ↑ navigate esc close

Daily Briefing: November 18, 2026, morning

Daily Briefing ·

Listen: https://serve.buildtall.systems/b650baf8e3a7792e9d2c38ced80fe2e4c947d133d4f7b90520fa40baa92e42f3.mp3


The briefing for the morning of November eighteenth, twenty twenty-six, covers twenty-one articles.

The largest cluster concerns Nvidia and the departure of Jensen Huang. TechCrunch reports in Nvidia says it doesn’t plan to reappoint Jensen Huang as CEO that the former chief executive offered to return to his top role after resigning last week, and the company does not intend to bring him back. In Nvidia says its open-source AI strategy remains intact after Jensen Huang’s exit, TechCrunch has the interim chief executive reaffirming the company’s open-source artificial intelligence strategy and its plan to acquire Hugging Face. Nvidia is abandoning China after Jensen Huang’s ouster says the interim CEO states the company is done selling chips to China. Nvidia promises that revenue will keep soaring — for now says that despite the turmoil after the departure of the disgraced founder and chief executive, the company projects its record-breaking revenue streak will continue.

Nvidia Rejects Jensen Huang's Offer to Return as CEO, from Slashdot, draws on the Associated Press. The board unanimously rejected Huang’s offer to return despite public support from President Trump and prominent Silicon Valley figures. Nvidia reported third-quarter revenue of one hundred sixteen billion dollars, double the year-earlier figure, and net income of two dollars and seventy cents per share against estimates of two forty-eight. Chairman Harvey Jones said the board unanimously accepted his resignation and does not intend to reinstate him as chief executive. Chief financial officer Colette Kress is running the company while a permanent replacement is sought. Huang acknowledged making serious mistakes in his personal life. Nvidia stock rose five point two percent after hours. The company also announced a new board committee overseeing national security and trade compliance amid investigations into alleged AI chip smuggling to China.

TechCrunch reports in Nvidia’s startup stakes ballooned to $65B, but Huang’s exit puts future deals in doubt that the value of Nvidia’s private investments has nearly tripled over the past nine months, fueled by Jensen Huang’s aggressive dealmaking and rising valuations in its portfolio companies.

The final and longest piece in this cluster is AI Data Centers and the MIT Beer Game from Feld Thoughts. It frames the AI infrastructure boom through the MIT Beer Game, in which retailer, wholesaler, distributor, and factory each react to a single step up in customer demand and produce wild swings upstream. The article argues the AI data center business has more actors: buyers of AI products, labs such as OpenAI and Anthropic, hyperscalers and neoclouds such as CoreWeave and Nebius, and Nvidia as factory. Each sees orders from the actor below, not actual end demand, and delays now come from land, power, permits, construction, elections, and geopolitics. The Commerce Department closed a loophole that let Chinese companies rent Nvidia computing power abroad. Nvidia’s results this week were spectacular: quarterly revenue of one hundred fifteen point eight billion dollars, up one hundred three percent, net income sixty-five point three billion, up one hundred five percent, on a seventy-four point eight percent gross margin. It generated forty-three point eight billion dollars of free cash flow and holds three hundred ninety billion dollars in purchase commitments. But Nvidia has also acted as the bank, taking equity stakes, backstopping neoclouds, and ending the quarter with one hundred eighteen billion dollars in guarantees, including one hundred five billion dollars of credit support for a single Ohio campus. It took a one point eight billion dollar impairment on private stakes. Jensen Huang’s resignation landed on a cooling credit market, with some data center bond deals shelved and CoreWeave’s 2031 bonds yielding fifteen point six six percent. Apollo’s Marc Rowan said that if Nvidia’s guarantor function is removed, access to capital for many neoclouds will disappear. Colette Kress explained the guarantee starts once the builder has raised equity and debt and is ready to go, and if delays occur the guarantee no longer exists. Microsoft already trimmed its capital spending forecast due to delayed third-party capacity. Kress called returning capital a top priority; Nvidia spent twenty-two billion dollars on buybacks last quarter and sixty-one billion this fiscal year, with two hundred twenty-seven billion in remaining authorization. Customer obligations doubled to eighty-one billion dollars. Nvidia’s net cash is about twenty-one billion against commitments of five hundred eighteen billion. The piece compares cycles of railway mania, telecom fiber, shale, and chip hoarding, where demand was real but capital structures were wiped out.

A second cluster covers artificial intelligence models, agents, and companions. Simon Willison's Weblog discusses GPT‑6.2 Astra and GPT‑6.2 Sol. OpenAI introduced Astra as a frontier model for the hardest projects and Sol as an efficient, versatile model for everyday work, available through the API as gpt-6.2-astra and gpt-6.2-sol. Willison shipped LLM 0.38 with support for both. Astra is eighty percent of the price of the previous 6 Astra, at eight dollars per million input tokens and forty dollars per million output, the same price as Claude Fable 5.2. Its cached price is forty cents per million tokens, down from one dollar. Sol remains two dollars for input and ten for output. He ran his pelican-drawing tests and a Yonkers mech animation benchmark. Neither model did as good a job morphing Yonkers into a mech as Claude Fable 5.2 or Gemini 4 Carbon. He then used GPT-6.2 Astra in the Codex desktop app to create a split-screen video comparing the two animations. Reception is mixed: Astra 6.2 is not scoring as high as Fable 5.2 on some benchmarks, and Theo Browne’s review includes an example where the model, two hours into a task, asked an obscure internal implementation detail question that no user could answer.

External sandboxes for coding agents, also from Simon Willison's Weblog, describes work with Prime Radiant on an external sandbox pattern in which code executed by an agent runs inside a dedicated sandbox while the agent lives outside it. They released three sandbox server implementations: external-sandbox-v86 uses v86 to run Alpine Linux inside WebAssembly; external-sandbox-container is a Python web server running Alpine Linux in an Apple Containers instance; and external-sandbox-quicksand uses Microsoft’s Quicksand library to bundle QEMU and Alpine Linux in a Python package. All share the same localhost API and can be configured so Pi, Claude Code, or Codex sends code execution calls to the sandbox. The sandboxes include an optional HTTP and HTTPS allow-list proxy and a credential replacement mechanism for authenticated calls without sharing secrets.

Google To Keep Temporary Gemini Chats For Two Months, from Slashdot, reports Google is extending how long it keeps temporary Gemini chats and conversations with chat history turned off to around two months, up from the previous seventy-two hours. OpenAI may keep temporary chats in ChatGPT for up to thirty days, and Anthropic retains incognito Claude chats for thirty days by default. Google says the change improves safety and security, supports responding to users, processing feedback, and protecting services, and cites potential system failures; temporary conversations are not used to train AI models.

What Goes on in Our Brains When We See an AI Face from Nautilus carries only the statement that deepfakes have jumped the uncanny valley.

Massive New Study Shows How Young People Feel About AI Companions from 404 Media covers a Gallup study with the Boston Children’s Digital Wellness Lab, based on online surveys of more than sixteen thousand people aged fifteen to twenty-two in sixteen countries. Around ninety-three percent of respondents reported at least one strong in-person friendship, and thirty-three percent said they had an AI companion. In-person friendships were rated strongest, and digital friendships complemented rather than replaced them. Daily AI use varied: twenty percent in the United States, Canada, and the UK, almost forty percent in Brazil, and around thirty-five percent in Japan and Italy. Eighty-two percent had used AI to look up information, sixty-eight percent for advice, and thirty-three percent as a companion. Young people with online-only friends were more likely to use AI for companionship, though the study cannot establish whether it supplements or substitutes for human relationships. Australian users were most likely to rate AI friendships as high quality, at nearly half, while only twelve percent in Poland did; in most countries the rating was medium. People who used AI as a friend reported higher loneliness, but causality is not established. Seventy-two percent spent time scrolling short-form content, mostly TikTok and Instagram, and six in ten said they often or always scroll when lonely. Scrolling was perceived as nearly as helpful as spending time with a friend in real life. Strong in-person friendships were associated with lower loneliness, while online-only friendship quality showed no comparable effect. Not having a trusted adult outside the family stood out as a factor. The study concludes that the quality of relationships and wider social context matter greatly, and expectations for AI companions differ from those for human friendships.

Amazon and Google will together own nearly 30% of Anthropic after its IPO from TechCrunch states Amazon will own just under fifteen percent of Anthropic after the IPO, with an option to go much higher, and Google is close behind at nearly fourteen percent.

Daily Reading List – November 17, 2026 (#911) from Richard Seroter's Architecture Musings is a digest of recommended pieces. It covers levels of agentic engineering, Google’s work accelerating development for massive Git repositories, putting MCP next to an API rather than on top, ephemeral disks on Cloud Run, a list of ten AI agent frameworks, Google’s leader position in the 2026 Gartner Magic Quadrant for cloud operational database management systems, an argument that AI slop is acceptable as people learn, Gemini 4 Hydrogen as a faster model for general-purpose robotics, invisible engineering teams, the claim that agents build things but enterprises have to run them, full-stack Android apps with Gemini in Google AI Studio, and an argument on why Git needs help moving from files to systems for source control.

Symposium on Stoicism and AI from Stoicism Today is a call for submissions on the theme of Stoicism and AI. It invites essays, reflections, fiction, poetry, music, or visual art exploring how Stoicism can inform AI technologies, how AI ethics may be interpreted through Stoic ethics, and the human-AI relationship. All submissions must be produced by a human creator. The deadline is January thirty-first, twenty twenty-seven, with acceptance notices by March thirty-first and publication in the online journal.

Space is represented by Rocket Lab’s Electron completes 100th mission, Neutron progresses toward debut mission from NASASpaceFlight.com. Electron completed its one hundredth mission and became the fastest commercially developed rocket to reach that milestone. The mission, called The Mountain God Rises, launched from New Zealand’s Māhia Peninsula on November fourteenth, delivering the YAMATSUMI-II synthetic aperture radar satellite for the Institute for Q-shu Pioneers of Space into low-Earth orbit. Electron’s maiden flight in May 2017 failed when range safety activated the second stage’s flight termination system after telemetry was lost; blame fell on third-party ground tracking equipment, so no hardware changes were made. The rocket first reached orbit in January 2018 on the Still Testing mission. Electron has now logged a ninety-six percent reliability rate across orbital and HASTE flights, with another hundred launches on the manifest. For Neutron, the second stage and thrust module arrived at Wallops Island, Virginia, where the thrust structure was placed on the launch mount for integrated fluid and quick-disconnect tests. The second stage will next undergo cryogenic proof testing and wet dress rehearsals, then receive a single vacuum-optimized Archimedes engine for static-fire testing. The forward dome for stage one has been built and cured. Rocket Lab also announced a NASA collaboration on a robotic arm capable of walking across spacecraft and performing in-orbit repairs, scheduled for a late 2027 demonstration. Its Blue and Gold spacecraft for NASA’s ESCAPADE Mars mission executed powered Earth gravity assists on November seventh and ninth and are on track for Martian orbit in September 2027. Rocket Lab’s backlog stands at a record three point five seven billion dollars.

Stripe backer Atomico calls for Europe to own its success stories from TechCrunch reports that Atomico’s latest State of European Tech report argues Europe is home to many of the world’s fastest growing companies.

US Officially a Measles Endemic Country from Slashdot and The Hill reports the Pan American Health Organization has withdrawn the United States’ measles elimination status, more than two decades after it was achieved. PAHO’s re-verification commission concluded endemic measles transmission has been reestablished in Bolivia, Mexico, and the United States, and director Jarbas Barbosa accepted the recommendation. The US achieved elimination in 2000 and held it for more than twenty-five years, but the past two years saw nearly eight thousand infections and roughly a dozen deaths, with this year the worst transmission in more than thirty years. CDC director Erica Schwartz said losing the designation will not change the federal response, that one hundred million dollars has been invested in state and local health departments, and that the evidence-based approach remains.

Two pieces concern hacking and cyber risk. Me on the cyberattacks by Tyler Cowen on Marginal REVOLUTION argues the problems are serious but the sky is not falling. Cyber risk is considerably higher because AI-driven offense has outraced defense, and many experts say advanced AI agents can hack a wide variety of systems. But not every reported incident is a cyberattack. Some may be normal outages receiving more attention, and some may be failed patches or upgrades. Microsoft’s November twelfth outage, the FAA’s November thirteenth ground stop, and JPMorgan Chase’s November sixteenth outage all seem to have been something other than direct cyberattacks. Cowen cautions against being fooled by randomness and anecdote, while still calling for an emergency national emphasis on cyber defense because the risk of a larger, costlier AI-driven attack remains real.

Hacker Pleads Guilty to Leak of Grand Theft Auto VI Footage from Slashdot reports Artion Kurtaj, the British hacker responsible for leaking early Grand Theft Auto VI footage in twenty twenty-two, pleaded guilty to six offenses stemming from attacks on Rockstar Games, Uber, and Revolut. The twenty-one-year-old used a mobile phone and an Amazon Fire Stick connected to a hotel room television despite bail restrictions limiting his internet access. He admitted four Computer Misuse Act offenses, one fraud offense, and one possession of a mobile phone in prison at Woolwich Crown Court, and will be sentenced next week. Prosecutor Robert Hutchinson called it a calculated and persistent campaign causing millions in losses.

Varieties Of Discourseposting from Astral Codex Ten offers a taxonomy of writer-audience dynamics. The author notes he is teaching at the Inkhaven blogging fellowship and argues without having seen Naomi Kanakia’s class on discourseposting. The piece suggests classifying posting by the psychological need it fulfills, analogous to classifying sex by psychological need. It then defines the Voxsplainer, aimed at validating an already-agreeing audience; the Brutal Takedown, which appeals to the id and common enemy; the Marching Orders, which tells less cool readers what position to hold; the Virtue Signal, which takes a seemingly controversial but actually popular position; the Revenge Killing, which retaliates against a prior insult; the One-Level-Higher-Than-You, which claims to transcend both sides; the Slatepitch, an unpopular but trivial contrarian take for clicks; the True Contrarian, who holds a belief others need to beat up; and the Inkhaven Post, written to fulfill a daily writing requirement. Each category describes the writer’s need and the audience’s need.

The morning’s twenty-one articles are dominated by Nvidia’s leadership rupture and financial commitments, with a broad second theme of AI models, agents, and companions. Rocket Lab’s milestone, the loss of measles elimination status, and cyberattack commentary round out the window.

  1. Quad9's withdrawal from France: A canary in the coal mine
  2. Rocket Lab’s Electron completes 100th mission, Neutron progresses toward debut mission
  3. Amazon and Google will together own nearly 30% of Anthropic after its IPO
  4. What Goes on in Our Brains When We See an AI Face
  5. GPT‑6.2 Astra and GPT‑6.2 Sol
  6. Nvidia says it doesn’t plan to reappoint Jensen Huang as CEO
  7. External sandboxes for coding agents
  8. Google To Keep Temporary Gemini Chats For Two Months
  9. Nvidia says its open-source AI strategy remains intact after Jensen Huang’s exit
  10. Microsoft's ONNX Runtime has disappeared from GitHub
  11. Nvidia is abandoning China after Jensen Huang’s ouster
  12. Daily Reading List – November 17, 2026 (#911)
  13. Stripe backer Atomico calls for Europe to own its success stories
  14. Linklater's film Slacker (1991): Radical and Mediocre
  15. Nvidia promises that revenue will keep soaring — for now
  16. ID.me terms required for IRS sign-in
  17. Nvidia Rejects Jensen Huang's Offer to Return as CEO
  18. Nvidia’s startup stakes ballooned to $65B, but Huang’s exit puts future deals in doubt
  19. US Officially a Measles Endemic Country
  20. Massive New Study Shows How Young People Feel About AI Companions
  21. Interviewed at a YC startup, found out they used my LinkedIn without permission
  22. Me on the cyberattacks
  23. AI is killing All About Berlin
  24. Hacker Pleads Guilty to Leak of Grand Theft Auto VI Footage
  25. Varieties Of Discourseposting
  26. Symposium on Stoicism and AI
  27. AI Data Centers and the MIT Beer Game