Week two of our five-company series. The affair, the Rubio memo, and what it means for your next GPU order.

I had a whole issue planned about NVIDIA selling the system around the chip. Then Sunday happened.

Jensen Huang admitted to an affair, and NVIDIA’s board opened an independent investigation into reports that he was the target of a foreign intelligence effort. The personal side belongs to his family. What belongs in this newsletter is the business question: what happens to your AI plans when your most important supplier has a key-person problem?

The rest of the issue fills in the picture: a stock that was already in a five-week slide, how much China actually matters to NVIDIA’s numbers, regulators asking about the Hugging Face deal, and competitors working to make leaving easier.

On November 8, Huang told NVIDIA employees in an email that he had been unfaithful to his wife and apologized to his family, staff, board and shareholders (Yahoo Finance). NVIDIA’s board said an independent committee working with outside counsel will investigate, that it has no reason to believe national security or company secrets were compromised, and that it has confidence in Huang’s leadership (Axios).

Then the story got bigger. The New York Times reported Sunday that Secretary of State Marco Rubio oversaw a classified memo this spring warning senior administration officials about Huang’s relationship with Alice Huang, a Megaspeed executive US officials believe passed information to Beijing — the memo went out before President Trump’s May trip to China (Yahoo News). China has called the spying allegations groundless speculation (Taipei Times), and Trump said Sunday he hasn’t spoken with Huang but still thinks well of him (CNN).

So is Huang a supply risk? Not for your next order. NVIDIA’s chips come off TSMC lines and through a supply chain that doesn’t run on one executive’s calendar. The risk is slower and more structural. First, key-person risk: Huang biographer Stephen Witt has pointed out there’s no second in command and no obvious successor at the top (Yahoo Finance). Second, policy risk: Huang has been the public face of NVIDIA’s push to sell into China, and analyst Dan Ives expects Sunday’s news to bring more scrutiny to those sales (CNN). If Washington responds by tightening export rules or allocation conditions, that ripples into what everyone else pays and waits for.

I’ve watched a lot of founder-led companies over 30 years, and when the founder is the brand, a personal crisis becomes a governance story fast. NVIDIA says the news won’t slow the company and Huang will keep leading it without interruption (Axios). My read: he isn’t a supply risk this quarter. He’s a concentration risk, and so is any vendor you can’t replace. So if NVIDIA is a big line item in your roadmap, this is a good week to write down your second source: another cloud region, another accelerator, or one workload you could move if you had to. You don’t need to panic. You do need a plan B that exists somewhere besides your head.

Read the full story →

The date three House committee chairs proposed for Huang’s closed-door briefing, the day after NVIDIA reports third-quarter earnings (Free Press Journal). He was asked to accept or propose another date by November 11 (IBTimes). Two days, one company, and a lot of questions from investors and lawmakers.

Source: Free Press Journal

NVIDIA fell 8.76% in the week ended November 6, its worst week since April 2025 and its fifth straight weekly loss, leaving shares about 22% below their October 6 record (Yahoo Finance). The company is still worth about $4.51 trillion, and its last earnings report named supply, not demand, as the main limit on growth. Analyst Dan Ives called Sunday’s news a black eye for Huang that will bring more scrutiny to chip sales to China (CNN). A stock price doesn’t change your delivery date. Political pressure on where chips go can, so ask your vendors how their allocation commitments hold up if export rules tighten.

NVIDIA’s first H200 shipments to China under the US licensing scheme accounted for less than 1% of its $89 billion second-quarter data-center revenue (South China Morning Post). The arrangement that allowed them, announced by President Trump in December 2025, sends 25% of that revenue to the US government (Tom’s Hardware). So the near-term revenue exposure is small. The bigger exposure is political: the deal Huang lobbied for is now under a brighter light, and any change to it lands on NVIDIA’s whole allocation picture, not just its China line.

Reuters reported on October 28 that Justice Department staff are examining whether NVIDIA’s $12.9 billion purchase of Hugging Face would give the chipmaker control of a key route for AI models (The Next Web). NVIDIA has pledged that Hugging Face will stay open to every model builder and won’t require NVIDIA compute (NVIDIA). If your team pulls models from Hugging Face, nothing changes today. Keeping a copy of the models you depend on somewhere you control is cheap insurance either way.

Google announced open source TorchTPU on October 20, bringing native support for the PyTorch machine-learning framework to Google’s TPU accelerators. This challenges an important part of NVIDIA’s advantage: the effort required to try another computing platform. NVIDIA’s competitors have to compete with the accumulated software around its hardware too. Making familiar workloads easier to move is one way to reduce that advantage. If this week has you thinking about a second source, this is the kind of work that makes one realistic.

NVIDIA’s primer explains why model design affects hardware performance. Start with its distinction between total throughput and the response speed one user experiences. Those goals can favor different engineering choices — a single speed number can’t describe both.

Read it →