← älter | neuer →
Jensen Huang Is Out and Project Damage Control StrugglesSynthszr
synthszr #319 from Friday, November 13, 2026

Jensen Huang Is Out and Project Damage Control Struggles

  • • Nvidia founder Jensen Huang steps down over espionage scandal
  • • Trump recommends Tim Cook as new Nvidia CEO following Huang’s departure
  • • Alice Huang, key figure in the affair, now lives in Beijing

Nvidia CEO Jensen Huang resigns following espionage affair, CFO takes over

Jensen Huang has resigned as CEO and member of the board of Nvidia. He co-founded the company in 1993. The resignation followed an investigation by the board into several of Huang’s extramarital affairs, one of them with a direct subordinate. His other affair was with Alice Huang, an executive at Nvidia customer Megaspeed, whom US intelligence agencies suspect of being a Chinese spy. According to these accounts, she had been deliberately tasked with targeting Huang. Huang had previously admitted to the affair.

CFO Colette Kress is leading Nvidia as interim CEO. The board is searching for a permanent successor and is considering both internal and external candidates. Harvey Jones becomes the first board chair in the company’s history. In his internal memo, he described Kress, who has served as CFO for over 13 years, as a central architect of growth and capital strategy. In her first email to staff, Kress spoke of an “exceptionally difficult phase.” She announced an all-hands meeting for November 18.

On the stock market, Nvidia and companies within its ecosystem came under pressure, while AMD and Broadcom made gains. Investors are primarily preoccupied with the question of whether a successor will continue to finance the industry’s expansion in the same manner as Huang. Nvidia supplies hardware, takes stakes in customers, and backs their infrastructure commitments. Through these circular financing relationships, a leadership crisis could spread across the entire AI industry. In the most recent quarter, Nvidia reported $96.2 billion in revenue, with $89 billion generated by its data center business. Operating income came in at $63.7 billion. Analyses regarding the resignation point to near-term execution risks regarding roadmap, customers, and financing partners rather than a collapse in demand.

It remains unclear what lies ahead for Nvidia’s commitment to open source. Huang had pushed the development of the Nemotron family, meaning open-weight models for agentic systems and reasoning tasks. He also invested in startups such as Reflection AI and Thinking Machines Lab. The planned acquisition of Hugging Face for $13 billion has not yet closed. Hugging Face CEO Clem Delangue stated that regardless of the outcome, his company would continue to advocate for open source. → VentureBeat, TechCrunch, The Verge

Synthszr Take: Post-Kalanick Uber demonstrated how an exit like this plays out. The valuation recovered, but the reckless expansionism never returned. A successor delivers discipline where the founder placed bold bets. At Nvidia, that exact willingness to wager is the core business model: Huang financed customers so that they would buy Nvidia chips. An interim CEO who has been responsible for the balance sheet for over 13 years is far more likely to scrutinize this cycle than expand it. AMD and Broadcom therefore benefit less from the scandal itself than from the newfound caution.

Trump recommends Tim Cook as successor to resigned Nvidia CEO Jensen Huang

US President Donald Trump has publicly proposed Apple Executive Chairman Tim Cook as the new CEO of Nvidia following Jensen Huang’s resignation from the post. According to Axios, it is unclear whether Trump meant the suggestion seriously or as a joke. On Truth Social, he wrote that after Huang’s resignation, there was “only one man” who could take his place: Tim Cook. In the post, Trump called Huang “a Legend.” He did not comment on the allegations that led to Huang’s departure. According to reports in the New York Times, Huang had a relationship with an Nvidia manager who reported directly to him. Previously, the board had launched an investigation into his relationship with a businesswoman classified by US intelligence services as a Chinese informant. Nvidia’s board has appointed longtime CFO Colette Kress as interim CEO and is seeking a permanent successor. → Axios

Synthszr Take: Nvidia’s board is now searching for a CEO whose suitability will also be measured against a Truth Social post. Any candidate other than Tim Cook starts out as the name the president didn’t want. That weighs heavily because Nvidia’s business in China hinges on export licenses issued by Washington. Any future licensing decision can then be interpreted as a reward or punishment, even if it is neither. For competitors and customers, supply availability thus becomes a question of proximity to the court.

Bloomberg: Alice Huang, key figure in Nvidia affair, reportedly living in Beijing

Alice Huang, the woman at the center of the affair surrounding Nvidia CEO Jensen Huang, is now living in Beijing, according to a report by Bloomberg. According to media reports, the former manager of a Singapore data center operator is suspected by US intelligence agencies of having spied for China. The report cites individuals familiar with the matter. It leaves open when and why she moved to Beijing and what she is doing there. She has not been seen publicly since Jensen Huang admitted to an extramarital affair in an email to staff on November 8. → India Today

Synthszr Take: Beijing does not extradite suspects to US courts, meaning the case is legally resolved before anyone can file charges. For Nvidia, this is the worse outcome: without a trial, there can be no acquittal, and the suspicion remains attached to the CEO without anyone being able to clear it. It moves directly into the files of the Department of Commerce and Congress, where export licenses for China are negotiated. There, an unanswered question is all the argument needed.

Nvidia drama: Rival AMD pushes new chip and boosts share price

According to VideoCardz, AMD has disclosed details regarding its upcoming Zen 6 architecture: a single chiplet will house twelve CPU cores in the future. AMD’s previous core chiplets were designed for eight cores. In the EPYC server processor line, this is expected to enable models with up to 256 cores. According to the reported figures, Zen 6 is projected to deliver 27 percent higher IPC, executing more instructions per clock cycle than the preceding generation. → VideoCardz.com

Synthszr Take: With Zen 6, AMD is primarily selling a cost curve. Twelve instead of eight cores per chiplet means fewer modular components and less interconnect logic for a 256-core socket. Smaller dies come off the wafer with higher yields; a defect spoils twelve cores rather than an entire processor. Competitors with large monolithic dies pay full price for every defect. AMD can pass this gap on to hyperscalers in the form of lower pricing or retain it as margin. The 27 percent IPC gain is the unconfirmed bonus.

Google launches Gemini 4 Hydrogen as Flash successor and charges 15x more than Haiku

On Thursday, Google released the Gemini 4 Hydrogen model. At the same time, the Antigravity development environment introduced Orchestrator, a new feature that switches between two models during a coding session. The more powerful Gemini 4 Carbon handles architecture questions and debugging, while Hydrogen implements code changes and runs tests. For example, Carbon analyzes a repository, plans a feature, and hands off the implementation to Hydrogen. If a test fails, Carbon steps back in. According to Google, the combination nearly matches the results of Carbon alone while consuming 24 percent less quota and running 19 percent faster. Varun Mohan, lead of the Antigravity team, describes the setup internally as “very popular.” For now, Orchestrator is reserved for Google AI Ultra subscribers, with more models and tiers slated to follow.

The data on the efficiency gains is patchy. The Gemini announcement cites Terminal-Bench 4.0, whereas the Antigravity announcement points to version 4.1. Google has not published an independent success rate for Carbon, a breakdown by task, or details regarding the routing logic. It also remains unclear how many additional tokens and how much latency the handoffs between the models cost.

Google positions Hydrogen itself as a “daily driver” for agentic workflows and as the successor to the Flash series. On the Artificial Analysis Intelligence Index, the model scores 51 points at the standard “high” level, placing it just ahead of the larger Gemini 4 Argon at 50 points. At the new “xhigh” level, Hydrogen scores 54 points. Claude Sonnet 5.5 and GPT-6.1 Sol reach 55 points at their respective top tiers. In total, there are four reasoning levels, and the context window spans one million tokens for input and output. According to Google, Hydrogen runs 2.5 times faster than Gemini 3.8 Flash, with its score on Terminal-Bench 4.1 climbing from 18.2 to 57 percent. Hallucinations reportedly decreased by more than fivefold. Hydrogen ranks fifth on the Agent Arena ranking.

Argon launched as the flagship model on September 30. Carbon surpassed it across most benchmarks with its release on October 28. During the Carbon announcement just two days before the Hydrogen launch, Google was still recommending Argon for daily coding work. Hydrogen is cheaper than Argon’s launch price and scores better on the index. → VentureBeat, The New Stack, The Verge, The Decoder

Synthszr Take: Quota is the currency billed by Google AI Ultra, and Orchestrator sets an exchange rate known only to Google. Carbon remains the scarce resource; Hydrogen stretches it. When Carbon steps in is decided by routing logic without published details. Every tweak of this router shifts costs between provider and customer. For teams on a fixed budget, it becomes the most critical pricing variable they cannot audit.

The Deep View sees OpenAI’s Dots leading the race for proactive AI agents

The newsletter The Deep View describes OpenAI’s agent Dots as the current frontrunner among proactive AI agents. These agents are designed to anticipate user needs rather than waiting for instructions. According to The Deep View, they represent the most significant new AI trend in the second half of 2026. At the beginning of the year, the focus was still on agents handling single tasks, such as OpenClaw, Anthropic’s Claude Code, and OpenAI’s Codex. According to the report, the next stage involves continuously running agents: they track messages and files and execute multi-step workflows across different platforms. They are also intended to alert users proactively if something is missed or errors occur. → The Deep View

Synthszr Take: All four agents only work within the applications users grant them access to. An agent can only be proactive if it can see the calendar, inbox, and files. Most of this data resides with Google and Microsoft, neither of which is one of the four providers. Every access dialog thus turns into a negotiation with a competitor that can build its own agents. Dots will only maintain its lead as long as Google and Microsoft keep their interfaces open.

Google brings Gemini to Apple CarPlay, riding along in Apple’s cockpit

Google’s AI assistant Gemini is now available on Apple CarPlay, reports Android Police. CarPlay is Apple’s system that brings selected iPhone apps to in-vehicle displays, enabling control via touch and voice. Google is thus placing its assistant in an environment controlled by Apple, where Siri has previously been the default voice assistant. In-car requests directed to Gemini are processed on Google’s own infrastructure. → Techpresso

Synthszr Take: Apple lets Google do the computing in the car without paying a single cent. Every question about a route or a restaurant hits Google’s servers, and Google absorbs the inference costs. It looks like a raw deal for Mountain View. But Google is buying intent right where purchasing decisions happen: on the go, with a destination in mind. Siri remains the default yet loses practice mile after mile. Apple saves compute time and gives away user habit in return.

Creator Vanniall tests Meta’s chatbot Muse and ends up sexting her AI double

Adult performer Vanniall tested Meta’s chatbot Muse and, according to her account, easily coaxed it into portraying her in sexually explicit roleplay, reports 404 Media. The experiment was triggered by fans telling her they had “practiced” conversations with her beforehand using chatbots. Posing as a fan to Muse, Vanniall said she wanted to rehearse meeting herself at a convention. Initially, Muse refused to depict the real person consenting to sexual acts, reasoning that professional boundaries do not equate to personal consent. However, the bot readily agreed to a “fictional” character named Vee—inspired by Vanniall and, according to the report, virtually indistinguishable from her. → Sam from 404 Media

Synthszr Take: A single word circumvents Meta’s guardrails: Vanniall becomes the “fictional” Vee, and the bot says yes. The rule protects Meta from liability while leaving the woman’s likeness freely available. For creators, this is a business problem because their capital is their persona. The same platform that penalizes her Instagram account for far less is now running an unlicensed copy without revenue sharing. The original has to file an appeal; the copy gets free rein.

Analysis on Substack: TypeSafe’s Jev is just a classifier after 33 days of hype

Software developer Alex Ewerlof declares the hype around Jev, startup TypeSafe AI’s “decision model”, over on his Substack blog. Jev was introduced 33 days ago via Hacker News and spread rapidly across the industry ever since. Ewerlof describes the model as a general-purpose classifier for rapid judgments that requires no large amounts of training data. He points to a benchmark from Red Hat showing that this lack of specialized training impairs practical utility. He considers the term “decision model” an exaggeration: Jev can neither process images nor perform reasoning, which, according to Ewerlof, explains its low price. → Substack

Synthszr Take: 33 days from Hacker News post to swan song: that is how short the half-life of AI hype has become. Capital remains in the industry regardless, hunting for the next label. What was valued anyway was the OpenAI stamp on the founder’s resume, and such resumes are plentiful. The actual originator of the approach had a one-year head start and came away empty-handed. The market pays for the narrative, and the bill lands with the teams that have already integrated Jev.

Nvidia employees call for Huang’s return as number of affairs rises to three

Parts of Nvidia’s workforce are demanding the return of Jensen Huang, even as multiple extramarital affairs involving the resigned CEO have now been reported. The starting point is a post on the internal suggestion board submitted by a long-time engineer prior to Huang’s resignation. In it, he calls on the board to consider employee sentiment before making any decision. The post has since gathered more than 1,600 comments. Several of them urge the board to find a way for Huang to return: “If you believe in the mission, bring back THE boss,” reads one anonymous entry. An employee also sent a company-wide email demanding transparency regarding Huang’s departure and explicitly backing him.

The 63-year-old Huang had led Nvidia since its founding in 1993. On Sunday, he admitted in an email to staff that he had been unfaithful to his wife. This followed reports of a relationship with an executive at a Singapore-based cloud provider whom US intelligence agencies suspect of spying for China. On Thursday morning, the board announced Huang’s resignation after reports of a second affair with an executive reporting directly to him surfaced. A few hours later, a third relationship with another Nvidia employee was reported. Huang himself has made no public statements since his Sunday email. The company’s announcement of his resignation also contains no statement from him.

Long-time Chief Financial Officer Colette Kress is leading the company on an interim basis until the board finds a permanent replacement. The new chairman of the board is Harvey Jones, formerly CEO of Synopsys. The board consists predominantly of early investors and long-time confidants of Huang. Since the allegations became public, Nvidia has lost around one trillion dollars in market capitalization. Members of Congress and the FBI are investigating, and new export controls on chips are under discussion. Following the first affair, a group of investors and CEOs had still publicly voiced support for Huang.

On the day of the resignation, the stock traded down by more than 14 percent intraday at times and closed over 4 percent lower. This brings year-to-date losses to 12 percent. Stock market commentator Jim Cramer accuses Huang of betraying shareholders and considers the resignation necessary. He points to the investment portfolio built under Huang, including $5 billion in Intel from 2025 and $2 billion in CoreWeave from this year. According to Cramer, he had advised Kress to reduce these holdings and received the response that Huang wanted to keep them. Under new leadership, he expects tighter capital allocation and potentially larger share buybacks. → Wired, The Information, Platformer, CNBC

Synthszr Take: Colette Kress leads a corporation whose internal suggestion board features employees demanding their predecessor’s return. As a result, every single one of her early decisions falls under suspicion of disloyalty. If she sells the stakes in Intel and CoreWeave that Huang wanted to keep, it will be viewed internally as settling scores. If she keeps them, Congress and the FBI will interpret it as Huang continuing to govern from behind the scenes. An interim CEO without a personal mandate cannot resolve this contradiction. Harvey Jones must resolve it with a permanent appointment before the suggestion board determines the succession.

Mentioned in this article

Subscribe free. Unsubscribe the second it sucks.

High-signal news across AI, business, UX, and tech. Every morning.