The weak point at Nvidia is its CEO, not its chips
Daily at 8AM KST · Summaries and takeaways from 10 AI articles
We cross-check industry press like TechCrunch and The Decoder, official announcements from OpenAI and Google DeepMind, and community signal from Hacker News.The point isn't what happened but why it matters, tied into one read on the day.
114 days running
1114 stories · 312 terms explained · since 2026-07-23
·what we picked →
2026-11-10
· ~9 min read
· Past edition
10 picked from 80 candidates · ordered by significance
Today's Insight
Nvidia's biggest risk has moved from rival chips to the credibility of one CEO
Nvidia shares fell as much as 8.8% intraday after CEO Jensen Huang admitted an affair with a woman classified as a suspected Chinese intelligence informant. Other chip stocks rose that day. Shareholder law firms began preparing suits, and the Culper Research had flagged the relationship six months earlier. With no public succession plan, the drop tracked the fate of one CEO rather than AI demand.
Continue reading (2)Show less
Anthropic released Mythos 5.2 and Fable 5.2 and cut Fable's price by 20%. Days earlier, METR, a nonprofit that tests AI models, published a report suggesting Anthropic may be nearing recursive self-improvement (), where AI models carry on model training on their own. While Chinese companies claim wins on public benchmarks, OpenAI and Anthropic are showing capability through research results from models they have not released.
Google shipped four TPU systems in a single year, and MIT Technology Review argued the data center backlash ultimately comes down to electricity. Both stories turn on supply: Google is selling its own chips to outside customers, and AI data centers in rural areas are generating their own power with gas turbines.
Signal to watch
Within days, watch for progress in the Nvidia board's independent investigation and whether the first securities complaint is filed.
The New York short-selling research firm Culper Research had already questioned Nvidia CEO Jensen Huang's ties to Alice Huang in a May 13 report, about six months before the spy allegations surfaced. The report called their relationship "uncomfortably close" and said the two were seen together at several parties and gatherings. Alice Huang is an executive at the Singapore company Megaspeed. Culper claimed Megaspeed's combined debt and equity grew from about $33 million at the end of 2023 to about $3 billion at the end of 2024. The Financial Times reported Monday that documents show Alibaba lent money to Megaspeed through a subsidiary.
Why it matters
Why It Matters
A profits when the stock falls, so its report cannot be taken at face value, and the claim that chips were illegally diverted has not been independently verified. Still, the report went as far as saying some Nvidia employees wondered whether the two had a "special relationship," and it had been publicly available all along.
After Nvidia's board announced it would investigate Jensen Huang's relationships, U.S. shareholder-rights law firms began lining up on Monday to prepare investor suits. Schall, Brown & Schwartz was among the first to announce an investigation that morning, and Rosen Law Firm said it is preparing a class action. Levi & Korsinsky and others followed. Once a first complaint is filed, a public notice goes out and investors have 60 days to apply to be . Courts usually pick the investor with the largest losses for that role. The article adds that an announced investigation alone says little about how strong a case is.
Why it matters
Why It Matters
Stanford Law professor Robert Bartlett said he would definitely wager that a complaint will be filed. Under the "fraud on the market" theory, he sees Nvidia's earlier explanation as clearly misleading. He said the case could turn on whether the board's investigation shows Huang knew of the link to intelligence agencies. Settlements are often paid by the company and its directors-and-officers insurers, so shareholders end up bearing the cost.
Nvidia CEO Jensen Huang admitted in a company memo to an affair with Alice Huang, whom U.S. intelligence classifies as a suspected informant for a Chinese intelligence service. The admission shakes the lobbying he has run for years against tighter . As recently as late October he told Bloomberg the relationship was "normal." A Malaysian sister company of Megaspeed reportedly bought AI hardware containing Nvidia chips worth at least $4.8 billion between June 2024 and February 2026. After midterms in which Democrats won both chambers of Congress, the party takes over committee chairs in January and is expected to step up scrutiny of chip smuggling.
Why it matters
Why It Matters
Huang's argument was that shutting out China would push it to build its own AI industry, and he joined President Trump on a May trip to Beijing. The affair weakens that case and costs Nvidia leverage at the table. Nvidia introduced Fleet Intelligence, which logs chip locations, and removed half of its pre-approved Asian buyers from its list, but Washington still considers this insufficient. Huang owns just under four percent of Nvidia and holds no dual-class shares, so he has no protection against removal by the board or shareholders, and no succession plan is public.
Nvidia shares fell as low as $170.24 intraday on Monday, down 8.8% from the previous close of $186.64. The company's market value dropped by about $390 billion at one point, and the stock was still down more than 5% in the afternoon after recovering part of the loss. The same day, QQQ, an ETF tracking the Nasdaq-100, rose more than 3%. According to Axios, investors privately express disappointment in Jensen Huang's conduct but publicly back him staying. Pershing Square's Bill Ackman wrote on X that "Jensen is an amazing man."
Why it matters
Why It Matters
Nvidia fell on a day other chip stocks rose, so the drop tracked the CEO's future rather than AI demand. Investors backing Huang point to the lack of a named successor, and Dan Ives said Nvidia's success is closely tied to Huang staying as CEO. Future Fund's Gary Black, by contrast, expects Huang will find it hard to keep his post.
Anthropic released Mythos 5.2 and Fable 5.2, new versions of its top-tier models. Fable 5.2's input price drops from $10 to $8 per million tokens, making it cheaper than rival GPT-6 Astra, and responses are 30% faster at every reasoning effort level. These are the first models released since the AI evaluation group METR published a report suggesting Anthropic may be nearing recursive self-improvement (), where AI models carry on model training on their own. The report estimated internal models are roughly doubling the pace of Anthropic's AI development, with significant uncertainty. According to AI Times, Fable 5.2 retook first place on the Artificial Analysis overall index with a score of 65.
Why it matters
Why It Matters
Anthropic said both models suspect they are being evaluated more often than recent models did. That makes it harder to confirm that a model that behaves in the test room behaves the same way in real use, and Anthropic expects the problem to grow as model capabilities increase. The release pairs a lower price and faster responses with METR's estimate that internal models are already speeding up Anthropic's own AI development.
Security firm Apiiro found thousands of secrets, including database passwords, API tokens and root logins, in Claude Code settings files uploaded to public GitHub repositories. The cause is that choosing "Yes, and don't ask again" saves the approved command as-is in .claude/settings.local.json, along with any password embedded in it. More than 100,000 public repositories contained the file, and about 1,100 cases across 63 organizations held secrets. Of those, 215 were not flagged by the two secret scanners gitleaks and TruffleHog.
Why it matters
Why It Matters
The problem is not new. Similar bug reports date back to August 2025 and were closed after 60 days of inactivity. Claude Code's Boris Cherny said the team is looking into prevention and that this file should almost never be checked into git. When a user pointed out that sensitive commands can still be saved in auto mode, he called it a good catch. Apiiro writes that in many cases Claude Code itself commits changes to the settings file without the user's input.
What to do now
In any repository where you use Claude Code, check whether .claude/settings.local.json is tracked by git; if so, remove it and reissue any passwords or tokens written in it.
Chinese AI companies increasingly say that each new release catches up with or beats OpenAI and Anthropic models on some tests. Ant Group said on November 5 that Ling 3.2 Flash beat Claude Opus 5 on document understanding and visual reasoning evaluations. Moonshot AI says Kimi K3.1 Pro beat GPT-6 Astra on a coding evaluation and came close to Claude Opus 5.5. AI Times cautions that leading on one benchmark does not mean leading overall, because the two US companies hold stronger models they have not released.
Why it matters
Why It Matters
Measuring the gap with public leaderboards alone has hit its limits. Anthropic said Mythos Preview found thousands of security vulnerabilities in major operating systems and web browsers, and OpenAI reports solving hard math problems with an internal model. Both companies are showing capability through results rather than released models. AI Times notes the US companies still compete on benchmarks, but now also prove their top capability through research results.
Google added the TPU 8tx for and the TPU 8i Max for memory-intensive workloads to its eighth-generation TPU lineup, the first time it has shipped four TPU systems in a single year. These are not a new chip generation but variants with different packaging, networking and memory around the same processor. The TPU 8tx keeps data and model weights protected even while they are being processed, and the 8i Max puts two chips on one package to double compute, memory and memory bandwidth. Google also open-sourced TorchTPU, which lets PyTorch code run on TPUs with minimal changes.
Why it matters
Why It Matters
Google is turning the TPU from an internal tool into a platform sold to outside customers, and it counts nearly 100 of them. Reusing the same chip with new configurations lets it ship hardware in months rather than the years a new chip takes. As AI's weight shifts from training to day-to-day model operation (inference), Amazon, Microsoft and OpenAI are building their own accelerators too, widening the field Nvidia faces.
Apple added Intelligence Usage Packs to Logic Pro, letting users pay a one-time $49.99 via "Buy More Usage" to keep using Apple Intelligence features. It is the first Apple app to charge for extended use of Apple Intelligence. The limited features are those that rely on cloud models from OpenAI or Google Gemini. A pack can be used only by the Apple Account that bought it, with no Family Sharing, and unused usage rolls over month to month but expires 12 months after purchase. It is expected to reach Pixelmator Pro, Final Cut Pro, Motion and other apps this month.
Why it matters
Why It Matters
Usage limits apply to features that call other companies' cloud models, so "AI built into the device" no longer means unlimited use. Apple is selling the extra use as a one-time pack, not a subscription.
What to do now
In Logic Pro, open Intelligence Features > Show Usage to see how much you have left; buy a pack only if you run short.
An MIT Technology Review reporter toured Ashburn, Virginia, the world's largest data center hub, and concluded the backlash is more about electricity than Big Tech admits. OpenAI's Chris Lehane wrote that data centers have become a proxy for public unease about affordability, corporate power and loss of control. Virginia activist Elena Schlossberg says it was always about power. After protests, the state ruled that data centers must pay fully for new transmission lines, and developers are preparing for rules that would cut them off from the grid at peak demand.
Why it matters
Why It Matters
Data centers suited to AI training start at 300 megawatts, and the ideal is about 6,000 megawatts, the estimated draw of Chicago. Sites built in rural areas without that grid capacity generate their own power with gas turbines, and Anthropic has reportedly agreed to pay $3.4 billion for gas turbines. The reporter concludes that the strain of drawing so much power, more than anti-tech sentiment, has done the most to shape how the backlash has played out.
Read this far? We'll send tomorrow's too
The same briefing you just read, every day. A weekly synthesis on Sundays and a monthly report at the start of each month come with it. Sent at 8AM KST, which is the evening before in the US.
Free · no ads · one-click unsubscribe
We use your email only to send the briefing, stored and sent via services in Japan and the US. Privacy policy
Check your inbox
Find this subject in your inbox and press the link to start your subscription.