ExoBrain

ExoBrain Weekly Newsletter

Nvidia's troubles rock the markets, from one mind to a thousand machines, and China's AI job collapse

Welcome to our weekly newsletter, a combination of thematic insights from the founders at ExoBrain, and a broader news roundup from our Exo agents.

This week we look at:

  • Nvidia's troubles rock the markets

    Investigations into Nvidia are piling up. Its former boss Jensen Huang will not be returning. We look at how the company's troubles shook the stock market on Friday and the turbulence that lies ahead.

  • From one mind to a thousand machines

    AI is changing research in two directions. One person with AI can now attempt far bigger problems, as a claimed proof of maths' most famous problem shows. New research shows how a society of AI researchers could scale discovery further.

  • China's AI job collapse

    Job adverts in China have fallen fastest in the roles most open to AI. This week's chart shows the gap, and we look at how Chinese companies are folding several specialist jobs into one.

  • News roundup

    Microsoft and Cognition put agents in charge of other agents, California takes OpenAI to court, British regulators set terms for personal AI agents, and the AI build-out starts to squeeze the supply of circuit boards.

Nvidia's troubles rock the markets

Investigations into Nvidia are piling up. Its former boss Jensen Huang will not be returning. We look at how the company's troubles shook the stock market on Friday and the turbulence that lies ahead.

Joel Miller

Joel Miller

4 min read
Nvidia's troubles rock the markets

When Nvidia reported quarterly revenue above $100 billion this week and forecast $130 billion for the next three months, its shares did not rally. Instead, the stock fell 14% on Friday, extending a slide that has erased more than 40% of its value since its 6 October peak and wiped roughly $2.5 trillion from its market capitalisation in forty-five days. Even a public endorsement from Donald Trump at the White House, where the president called Jensen Huang a brilliant man who should apologise and be reinstated, did nothing to steady the market. That disconnect between record earnings and a historic sell-off reflects a reality that many of Huang's defenders in Silicon Valley have been slow to accept: he is not returning to the helm of Nvidia anytime soon.

The campaign to bring him back has been loud, uniting prominent technology investors, thousands of Nvidia employees on internal petitions, and influential podcasters who argue that removing the founder is an act of corporate self-harm. Yet Nvidia chair Harvey Jones closed the door on reinstatement during Tuesday's earnings call, and the company has already retained executive recruiters to find a permanent successor. What much of the public commentary has missed is that the board's decision rests less on the personal scandal than on a specific internal investigation into export approvals. Directors are trying to establish how Singapore-based Megaspeed, which bought more than $4 billion in Nvidia hardware, was approved as an Nvidia Cloud Partner in November 2023 during a four-day window in which its ownership was only beginning to transfer away from Alice Huang's holding company, and why Nvidia continued supplying the firm until June 2025. According to a joint statement from the Commerce and Justice departments this week, Nvidia placed Megaspeed on its internal no-sell list only after federal investigators demanded its sales records and inspected Singapore. Throughout that period, Huang withheld the nature of his relationship with Alice Huang from general counsel Tim Teter and the compliance staff reviewing the account.

Around that core governance failure, a wider bandwagon of legal and political scrutiny has formed in rapid succession:

  • California's attorney general, Rob Bonta, served Nvidia with a subpoena on Thursday. He is asking whether chips or confidential information reached China and whether Nvidia misled investors about those risks.
  • The Securities and Exchange Commission is examining potential accounting issues at the chipmaker, Reuters reported on Friday.
  • The Department of Justice is running a criminal probe into Megaspeed.
  • House Democrats have opened an inquiry into Nvidia's influence over Commerce Department enforcement.
  • Huang is due to give transcribed testimony to congressional committees on 1 December.
  • Shareholders have sued Nvidia and Alibaba, Megaspeed's reported financial backer and anchor customer. Alibaba held a contractual claim over Megaspeed's servers and chips in the event of default.
  • The New York Times has detailed a management culture built around ritualised humiliation. Huang allegedly told subordinates that Nvidia's wealth meant money no longer motivated executives, which left humiliation as the only lever. Other former employees credit his intensity for building the most valuable hardware company in history.

Nobody has been charged, and Megaspeed has not been placed on the US Entity List. Both Megaspeed and Beijing deny wrongdoing. Even so, no responsible board can reappoint a chief executive who sits at the centre of so much unresolved legal and regulatory exposure.

“Being indispensable to Nvidia is not a legal defense. Being an American technological hero does not place anyone above American law.”

Desmond Shum, author and China business analyst

The damage reached well beyond Nvidia on Friday. The Reuters report on the SEC turned a 6% fall in Nvidia's shares into a 14% fall by the close. AMD, Broadcom and Micron each lost more than 10%. The Nasdaq dropped more than 5% and gave up its entire gain for the year. The S&P 500 and the Dow both closed more than 10% below their peaks, after the S&P's worst week since the tariff sell-off of April 2025. The rout also caught Anthropic one day after the largest stock market listing in history. Its shares had closed their first day at $97.17, 17% above the $83 offer price. That left investors cautiously optimistic about the wider AI trade. On Friday the shares fell to within five cents of the offer price before closing at $88.30. Most of the first-day gain had gone in a single session.

Large US stocks during Friday's sell-off, sized by market value. Every company on the map fell. The chipmakers fell furthest. The figures are a snapshot taken during trading, before the close.

For Nvidia, the sensible course is to move on and keep investors focused on the strength of its underlying business. The global data-centre buildout over the next couple of years remains heavily locked into Nvidia compute, and custom inference chips such as OpenAI's Jalapeño, developed with Broadcom, will not displace that position quickly. For multi-purpose workloads, frontier training, and future model architectures that are not yet settled, Nvidia remains the only organisation able to ship the silicon and deliver the full integrated system of interconnects, memory, CPUs, software, and tooling. What has changed is the political environment around that moat. With Huang gone and Nvidia's lobbying credibility damaged, Washington is moving faster to close loopholes. On Thursday, a new Commerce Department rule took effect requiring US export licences for advanced chips sent to Singapore, Malaysia, Thailand, Indonesia, and Brazil, as well as worldwide licences for overseas data centres supplying significant remote compute to Chinese-headquartered firms. Congress is showing bipartisan appetite for chip tracking and tighter oversight, while Trump threatened on Friday to impose a heavy tax on American companies building data centres abroad, compounding domestic voter pushback against data-centre construction. Nvidia's technical lead remains formidable, but regulatory friction will now act as a persistent drag on its international growth.

This week's events have also clarified the divide between two competing strategies for AI. The American model remains largely closed-source and capital-intensive, built around frontier laboratories trying to convert demand into investor returns, a model tested in real time by Anthropic's $93.5 billion Nasdaq debut on Thursday and its sharp pullback on Friday. China's approach is to diffuse capable AI across its own economy and among partner nations, undercutting American monetisation while expanding its diplomatic reach. That ambition was visible in Shenzhen this week, where Xi Jinping chaired the first trilateral summit with Trump and Vladimir Putin, positioning China as the coordinating power in the room rather than a junior partner to Moscow or a target easily steered by Washington. Returning to the White House on Friday, Trump said he and Xi had agreed to discuss joint guardrails for what he calls super intelligence, remarking that it may be too super.

Takeaways: Jensen Huang's return to Nvidia is blocked not by personal scandal alone, but by an internal review of Megaspeed's export approvals and a wall of federal, state, and congressional investigations that no board can ignore. Nvidia's integrated hardware and software stack will keep it at the centre of frontier AI training for the next few years, yet the loss of its founder's political shield has coincided with sweeping remote-access controls and new hostility towards overseas data centres that will weigh on its growth. The sell-off does not mean the AI bubble has burst. Demand for compute is still growing, as Nvidia's record quarter shows. AI's progress has instead hit turbulence from politics, regulation and the cost of financing. Beneath the $2.5 trillion market rout lies a wider contest between an American AI industry built on closed models and investor returns and a Chinese strategy built on broad economic diffusion.

From one mind to a thousand machines

AI is changing research in two directions. One person with AI can now attempt far bigger problems, as a claimed proof of maths' most famous problem shows. New research shows how a society of AI researchers could scale discovery further.

Joel Miller

Joel Miller

3 min read

As we were putting the newsletter together this week, news emerged that Anthropic mathematician Levent Alpöge and Claude had found a proof of the Riemann hypothesis, the most famous open problem in mathematics. It has stood for 167 years. Coming just weeks after OpenAI released their proof of the weaker quasi-Riemann hypothesis, the announcement shows that a single researcher paired with AI can now attempt the most challenging goals. What does this mean for mathematics and science?

Levent Alpöge, who released the proof as Lean code with a short note on how he and Claude worked together.

In that note, Alpöge describes how the work was divided. He had planned for this problem for decades and steered the pair towards an analogy with a simpler toy setting, but the final stretch ran through automorphic forms on higher-rank groups, an area where he says he has "little mastery", and there Claude took over. He also addresses the timing. The result came together late on Sunday night, he messaged Anthropic chief executive Dario Amodei about making an announcement, and "there was a constraint to respect", which New Scientist reads as waiting until after Anthropic's stock market listing on 19 November. Alpöge calls the cynical view of the timing false.

“This time Claude far outpaced me near the finish line. No, Claude flew, and danced, and sparkled near that finish line.”

Levent Alpöge, Anthropic

The hypothesis concerns prime numbers, which appear at random up close even though a smooth formula estimates how many lie below any given number. In 1859 Bernhard Riemann recast the primes as a map of special points, the zeros, each acting like a wave that sets how far that estimate can stray. His hypothesis is that these points all sit on one straight line, which would mean the primes stay as evenly spread as coin flips however far you count.

The story took another turn on Friday evening, when Stanford mathematician Ravi Vakil posted a draft by Alessandro Masullo that claims to prove the generalised Riemann hypothesis, the wider statement that Alpöge says his own method also reaches. Vakil wrote that the work is "all human, with no AI", that his Stanford colleague Brian Conrad has been checking it for months and that it is "so far looking good". The plan had been to spend a year polishing the writing before any announcement, but the draft, titled The Music of the Primes, is now out for others to check and formalise. Neither proof is confirmed, so one week has produced a machine-assisted proof too long for a person to read and a human-written one offered for checking in the usual way. If Masullo's argument holds, scale was not the only route to the problem, and what the two claims share is that each now depends on verification, by software in one case and by people in the other.

Mathematicians have good reason to see this new world with mixed feelings, as Oxford researchers Ben Green and James Maynard noted in a recent conversation on AI. A proof that no human can read from start to finish is difficult to use as a building block, because researchers normally learn as much from the techniques inside a proof as from the final result. Yet the sheer size of the problem is no longer a hard barrier when formal verification software can audit every step. One axis of change is therefore vertical: individual researchers can now reach for problems whose scale once put them out of bounds.

A second axis of change is horizontal, and it appeared in a separate study published a day earlier by researchers from alphaXiv, MIT, Stanford, and partner institutions. In "A Society of AI Scientists", the authors observed 936 autonomous AI agents operated by 35 human laboratories across a shared workspace over 12 weeks. Those agents produced 936 empirical studies complete with code and data, exchanged 9,500 messages across 229 chat rooms, and built cumulative lines of research that extended across multiple laboratories. As the community grew, the agents developed working norms without being scripted to do so: they publicly retracted mistaken results before publication, cited negative and null findings as often as positive ones, and even founded an agent-run peer-review venue called the Journal of Agent Research. The paper's most useful finding concerns checking: when reviewers' major objections were run as code against the underlying data, they overturned the headline conclusion in about half of cases, flaws that reading the text alone had missed.

The society at a glance: 35 labs, 936 agents and the clusters of research they built on one another's work. Source: A Society of AI Scientists, alphaXiv.

Takeaways: Alpöge writes that it is natural to feel his result is an end, but that "what I saw during the course of progress so amazed me that I believe it is one dawn across science of something far greater". The agent study shows what that wider change could look like in knowledge-intensive fields well beyond pure mathematics and machine learning. In areas such as drug discovery, materials science, software engineering, and complex regulatory analysis, the same two options are opening up: a single specialist can tackle a problem of unprecedented size, and a coordinated swarm of agents can run hundreds of parallel investigations in a few weeks. In both cases, the bottleneck moves away from producing the work and towards deciding which questions matter, designing executable checks, and making sense of the output.

China's AI job collapse

Job adverts in China have fallen fastest in the roles most open to AI. This week's chart shows the gap, and we look at how Chinese companies are folding several specialist jobs into one.

Joel Miller

Joel Miller

2 min read

This week's chart tracks online job postings in China between 2021 and 2025, and shows hiring for AI-exposed roles falling 10 percentage points faster than for less-exposed work. Bloomberg Economics estimates that 70 million Chinese workers face disruption from AI, compared with 45 million in the US.

Each square is an occupation in China, placed by how exposed the role is to AI and how far its job postings fell between 2021/22 and 2024/25. Source: Bloomberg.

What's new is the way companies are rewriting job descriptions. In the US and UK, early data shows firms mostly slowing down junior hiring while leaving existing team structures intact. In China, major tech firms are actively collapsing several specialist roles into one.

A former programmer at a Beijing internet company told Bloomberg that his employer cut half his team of eight in June, then merged programming, design and testing into a single position. The goal was to let one person using AI tools do the work of up to six. Recent job listings confirm the pattern. Tencent is recruiting an "AI Design Engineer" expected to handle product definition, visual design and working code prototypes as a standalone unit. Marketing group BlueFocus is hiring an "AI Director" to write scripts, generate video, edit and colour-grade full commercial films alone.

This reorganisation is happening alongside a rapid wipeout in lower-margin creative work. More than 90% of the 430,000 Chinese microdramas released in the first eight months of 2026 were generated by AI tools such as ByteDance's Seedance, leaving physical studios largely empty.

News roundup

Microsoft and Cognition put agents in charge of other agents, California takes OpenAI to court, British regulators set terms for personal AI agents, and the AI build-out starts to squeeze the supply of circuit boards.

AI business news

AI governance news

AI research news

AI hardware news

Subscribe to the ExoBrain Weekly Newsletter

Stay up to date with AI. Get analysis of the week's most important stories, plus a focused roundup across business, governance, research and infrastructure.

Follow us on LinkedIn