Excelsior group logo

The Signal - daily AI evolution

Archives
Log in
Subscribe
November 15, 2026

The Signal — November 15, 2026

The Read

Two days after Nvidia lost the man who built it, the only things that moved were about who fills the hole he left — and in both cases the answer is nobody yet. In Washington, the AP reported that senior officials say President Trump does not plan to confront Xi Jinping over the affair when the two meet in China later this week: the scandal is being spent inside the defense authorisation bill rather than at the summit, which means the China chip regime now gets written in NDAA text instead of negotiated in a readout. In the open-model world, Nathan Lambert named the loss nobody had priced — Huang was the open-weight ecosystem's political patron as much as its funder, and there is no one of equivalent weight to pick up the phone when the open-weight hawks come hunting. Beyond that the day was genuinely empty: the Saturday roundups were almost entirely the preceding week's news re-dated, the oldest item I traced back to 7 November, and I cut all of it. The opening is the one Lambert points at and does not take — a credible American frontier open-weight lab is now the most valuable unclaimed slot in the stack, and the constraint Dylan Patel names is consolidating talent, not raising money, which is a solvable problem for somebody with conviction.


🌊 Tide

One confirmation, no shift. All four tides hold. governance-as-market-structure took the day's only hard evidence, and it arrived as an unnamed official's briefing rather than a rule.

The chip scandal is being spent in the defense bill, not at the summit

The AP reported from Hong Kong early Saturday that senior US officials, briefing on background, say Trump does not plan to confront Xi Jinping over the Nvidia affair — the two are due to meet in China later this week — and that those officials do not expect Chinese retaliation if existing smuggling and chip-use loopholes are closed. Trump has separately said Xi may raise the case himself. Beijing's foreign-ministry spokesperson Mao Ning said on Thursday that the US should stop fabricating a "Chinese spy" narrative as a pretext for imposing technology restrictions. The action is in the legislature instead: the Senate's version of the defense authorisation carries three provisions that would pull export discretion back from the executive branch — the AI OVERWATCH Act, giving Congress a review of licences for advanced AI chip sales to countries of concern; the Chip Security Act, requiring companies to track their AI chips abroad; and the MATCH Act, pressing allies to restrict sales of chipmaking equipment to China. Analysts quoted by the AP also expect the episode to accelerate China's domestic chip programme and to complicate talks on a planned US–China AI national-security hotline.

So what: Stop modelling your China compute exposure off summit outcomes and model it off the NDAA text. A licence regime Congress reviews is slower, noisier and much harder to reverse than one a president can trade away in an afternoon — which cuts in your favour if you are building on US-controlled silicon and against you if your roadmap quietly assumed a carve-out. The good news for planners is that statutory rules are legible: you can read them, price them and build to them, which is more than anyone could do with the deal-by-deal regime they would replace.

Sources: ABC News — The Nvidia scandal is heightening America's AI rivalry with China ahead of Trump's visit · 24/7 Wall St. — Congress Wants China Chip Limits in the Defense Bill After Nvidia's Spy Scandal


🌊 Waves

The open-weight ecosystem lost its patron, and no one of equivalent weight has picked up the phone

Nathan Lambert published "Open-source AI without Jensen" on Saturday, and it is the most useful thing written about the resignation so far because it prices a loss the market did not. Nvidia's open-model spending was a founder-style bet — Lambert's phrase is "flooding the zone with funding" — and the company's priority now is to stabilise the ship and deliver to customers over the next twenty-four months, which is not a posture that renews ambitious side bets. The harder loss is political: Lambert argues Huang "would stop a lot of shit from going wrong" and was willing to spend his own power defending open weights, at a moment when a well-funded safety coalition has, on his reading, open models "as a negative target", and when a major hack driven by a popular Chinese open-weight model is in his words "a matter of when, not if". On capability the gap is real and is not a money problem: Thinking Machines' Inkling 2 Small, whose 594-billion-parameter, 17-billion-active weights were published on Thursday after a hosted-only debut on 5 November, is on Lambert's assessment "the best Western open-weight model in a while" and still "far behind Kimi K3.1 Pro". Dylan Patel's point on the ChinaTalk–SemiAnalysis emergency episode, quoted in the piece, is the operative one — "It's not money, it's talent, man" — because the strongest people are concentrated at Anthropic, OpenAI, DeepMind and Meta and the rest are split across a hundred different startups.

Roadmap implication: If your 2027 plan assumed an American frontier open-weight model you could self-host, re-underwrite it this quarter on two legs instead: Chinese weights under real governance review, and enterprise post-training on a mid-tier open model you actually own. Then look at the position this leaves empty. A credible US frontier open lab is the most valuable unclaimed slot in the stack, the constraint is consolidating talent rather than raising money, and consolidation is exactly the kind of problem capital and conviction solve — which makes this an opening, not an obituary.

Sources: Interconnects — Open-source AI without Jensen · thinkingmachines/Inkling-2-Small · Hugging Face

The reassessment of Nvidia's founder started over the weekend, and the first witness is his own biographer

Stephen Witt, who wrote the 2025 Huang biography "The Thinking Machine", published a New York Times opinion essay that was picked up on Saturday, and it is the opening move of the reassessment phase. Witt writes that "in truth, Mr. Huang was quite arrogant", that "he had little tolerance for disagreement and almost no capacity for open discourse", and that having built Nvidia from nothing he then, "through his arrogance and toxic self-regard", created a leadership vacuum; he describes a culture of silence and deference, compares the management style to the instructor in Whiplash, and says Huang's growing closeness to the administration while pushing to ease China chip export restrictions contributed to his own downfall. Set that against what the company has actually put in writing: Nvidia's 12 November release announced Huang's resignation effective immediately, named chief financial officer Colette Kress interim chief executive while she keeps the CFO role, and installed director Harvey C. Jones as — in the company's own words — its first-ever Chair of the Board. A company that ran for thirty-three years without a board chair has now created one in the same breath as losing its founder. Huang has also agreed, according to multiple press reports, to testify before Congress before Thanksgiving, which falls on Thursday 26 November.

Roadmap implication: Supplier governance just became a diligence line on the one company your compute roadmap cannot route around. The question to put to Nvidia, or to anyone borrowing against it, is not who the next chief executive will be — it is which commitments were institutional and which were one man's phone calls, and who inside the company can still make the second kind. The constructive read is that a board which now has a chair and an independent investigation is a better counterparty over a ten-year build-out than one that had neither; the repricing is uncomfortable and it is also the market doing its job.

Sources: India TV News — Jensen Huang's biographer describes ex-Nvidia CEO as 'quite arrogant' with 'little tolerance for disagreement' · NVIDIA Announces CEO Transition


🌊 Ripples

Nvidia reports on Tuesday and the earnings date did not move, so a five-day-old interim CEO carries the call

Nvidia's 12 November transition release confirmed that third-quarter fiscal 2027 results stay on Tuesday 17 November, as previously scheduled. Colette Kress, named interim chief executive that day, takes the call five days into the job while remaining chief financial officer, with Harvey C. Jones newly installed as board chair and the independent investigation still running. This is the first hard number the market gets since the resignation, and the first time anyone can ask on the record what happens to supply commitments and vendor guarantees that were negotiated personally.

Do this now: If you hold Nvidia-linked exposure, a multi-year capacity agreement, or a compute contract that reprices off anyone borrowing against those guarantees, put Tuesday's call on the calendar as a decision point rather than a data point — and write down now, before the print, which single answer would actually change your plan. The most likely outcome is a boring quarter with excellent numbers, which is itself the information: it would mean the demand curve never depended on the founder.

Sources: NVIDIA Announces CEO Transition · StockTitan — NVIDIA CEO Jen-Hsun Huang resigns; Kress named interim

A founder-restoration campaign ran all weekend, and most of the signatures are not employees

The "Stand with Jensen" campaign kept growing through the weekend after Elon Musk said on X on Friday that a petition to reinstate Huang was under way and he was signing it too. Epic's Tim Sweeney, Replit's Amjad Masad, Y Combinator's Garry Tan and Atreides' Gavin Baker are listed among the supporters; Shopify's Tobi Lütke replied to Musk's post with "You have my sword." As reported on Sunday, the campaign site showed 136,994 names — 6,841 self-identified current Nvidia employees, 1,278 former employees, 73,776 shareholders and 55,099 others. Identities are self-reported and not independently verified, so treat the employee figure as a claim rather than a count. Nvidia's board has given no indication it will reverse course, and a search considering both internal and external candidates is under way.

Do this now: Read this as a sentiment indicator, not a probability — a board that moved under legal pressure and then created its first-ever chair is not a board that un-moves under a petition. The useful signal is the composition: when self-identified shareholders outnumber self-identified current employees by more than ten to one on a reinstatement drive, what is being defended is a risk premium rather than a management style, and a risk premium is something a competent successor can actually earn back.

Sources: International Business Times — Elon Musk, Tech CEOs Rally Behind Jensen Huang as Over 6,800 Nvidia Employees Demand His Return

Azeem Azhar stopped comparing tasks and started pricing the worker

In a Saturday sketchbook post — explicitly raw notes rather than a finished model, with no figures in the free portion — Azeem Azhar proposed inverting the usual AI-versus-human cost comparison. Instead of asking what a task costs a person against what it costs a machine, he asks what it costs the economy to create a worker at all: the parenting, educating and feeding from infancy to adulthood, over roughly two decades, with the outcome unknown for most of that period. Set against that, AI is "manufactured intelligence" — not the same thing as human intelligence, but something that looks like it, and crucially something highly replicable that follows a classic learning curve.

Do this now: Use this as the denominator in your next workforce-design conversation, because the per-task comparison your finance team is running understates the asymmetry: it prices the marginal hour on both sides and ignores that one side is copyable at the margin while the other takes twenty years to originate. The day-zero version of the question is the better one — not which tasks to hand over, but which problems were never worth staffing at human origination cost and are obviously worth attacking at replication cost.

Sources: Exponential View — 🔮 What does it cost to make a worker?


Read this edition and the archive at excelsiorgroup.ai/insights/signal

Don't miss what's next. Subscribe to The Signal - daily AI evolution :
← Newer The Signal — November 16, 2026 Older → The Signal — November 14, 2026
LinkedIn
excelsiorgroup.ai
Powered by Buttondown, the easiest way to start and grow your newsletter.