Every time a tech executive sells stock, a certain kind of commentary follows: the smart money is running for the exits, the AI boom is cracking, and ordinary people are the last ones holding the bag. I want to push back on that. Insider sales are one of the noisiest signals in finance, and the latest round of headlines about Nvidia is a good case study in how easy it is to read a spreadsheet and come away with the wrong story.
Here’s what actually happened. Bloomberg and Quartz reported that in the third quarter of 2026, Mark Stevens topped the list of insider stock sellers, offloading more than $47 million worth of Nvidia stock. Stevens is a venture capitalist and a director at Nvidia. The quarter was a rough one for technology stocks generally, with plenty of public warnings about the risks building up around AI. Other AI-linked names showed up on the seller list too, including Arista Networks CEO Jayshree Ullal and Amazon founder Jeff Bezos. The Washington Service, which tracks insider buying and selling, noted that those three people together are worth about $302 billion.
Why that $47 million number is smaller than it sounds
Put $47 million next to $302 billion in combined net worth and the proportions get clarifying fast. For people at that level of wealth, selling shares is less a verdict on the company and more a routine piece of financial housekeeping. Stevens’ sales were reported through trusts holding indirect interests, which is the kind of structure wealthy families use for estate planning, charitable giving, and tax management. Those sales happen on schedules set long before any particular week’s news cycle.
There’s also the plain arithmetic of concentration. If almost all of your wealth sits in one stock that has climbed enormously, every financial advisor on earth will tell you to sell some of it. That advice has nothing to do with whether you believe in the company. It has to do with not betting your entire family’s future on a single ticker.
And the seller list itself undercuts the tidy AI-panic narrative. The third-largest seller that quarter was an Illumina director, in genomics, not AI. A SharkNinja chair also ranked. If AI insiders were uniquely fleeing, you would expect the leaderboard to look more lopsided than it does.
What insider sales can and can’t tell you
I am not arguing that insider activity is meaningless. It’s just that the information content runs in one direction more than the other.
- Selling has many explanations. Taxes, diversification, philanthropy, a house, a new fund, a pre-set trading plan. Pessimism about the company is only one item on a long list.
- Buying has fewer. When an executive spends their own money purchasing more of their company’s stock, the list of motives gets short.
- Patterns beat snapshots. One quarter of sales tells you little. Years of consistent, scheduled sales tell you about someone’s financial plumbing, not their convictions.
That distinction is the part that usually gets flattened in a headline, because “director sells $47 million” is a better hook than “director continues long-running diversification plan.”
The part that matters for anyone following AI
This site exists to explain AI agents to people who don’t write code, so let me connect the dots. The skill this story asks for is the same skill you need to evaluate any claim about AI: separating a number from the meaning someone has attached to it.
AI agents are sold to us on numbers all the time. Benchmark scores. Task completion rates. Percentage improvements over last quarter’s model. Those figures are real in the same way that $47 million is real, and they’re just as easy to point at the wrong conclusion. A model that scores well on a coding benchmark might still fumble your specific workflow. A high task-completion rate measured in a lab says little about a messy inbox or a customer support queue with real people on the other end.
So the question to carry into both situations is the same: what would have to be true for this number to mean what the headline says it means? For insider sales, it would mean the sellers had no other reason to sell, which is almost never the case. For an AI agent demo, it would mean your work looks like the demo, which is also rarely the case.
A more useful place to look
If you want to understand whether the AI buildout is solid or shaky, the honest answer is that stock sales by a handful of very wealthy people won’t settle it. The genuine signals are slower and less dramatic: whether companies renew the AI tools they piloted, whether agents get trusted with work that has real consequences, whether the costs of running these systems come down.
Insider filings are a fine thing to read. They’re just a poor substitute for watching what the technology actually does once the cameras are off.
🕒 Published: