It’s September 3, 2026. Somewhere, a filing clerk hits submit on a Form 4, and a trust associated with an Nvidia director sells 1,022,239 shares of the most talked-about stock on Earth. The next day, more sales. No press release. No keynote. No leather jacket. Just paperwork, posted to a government website that almost nobody reads, describing a transaction larger than the annual revenue of most companies you’ve heard of.
By the time the quarter closed, Mark Stevens had sold nearly $947 million in Nvidia stock, putting him at the top of the insider-selling list for Q3 2026. That’s not his first trip to the register, either. He sold $38.5 million in March 2026 and $186 million in June 2026, which adds up to over $534 million across those filings in less than a year, before the September round pushed the quarterly figure near a billion.
What the list actually looks like
Stevens wasn’t alone. Jayshree Ullal sold around $446 million in Arista Networks stock. Jeff Bezos moved over 1.2 million shares of Amazon for roughly $346.5 million. Between the three of them, the combined fortune sits at about $302 billion. The selling wasn’t limited to AI names, either — Illumina director Keith Meister sold approximately $436 million in Illumina stock, and SharkNinja chair Xuning Wang also made the quarter’s rankings.
So the honest read is that big holders of big positions sold big chunks. That happens every quarter in every sector. What makes this one interesting isn’t the mechanics. It’s the mood.
My angle, as someone who reviews AI tools for a living
I spend my weeks testing agents that promise to replace your sales team, your QA department, and occasionally your judgment. Most of them don’t. The gap between what AI products claim and what they do is the entire reason this site exists. So when I look at a number like $947 million in a single quarter, I’m not reading it as a market prediction. I’m reading it as a reminder about who’s capturing value in this cycle and who’s paying for it.
Here’s what I’d ask you to hold onto: the money in AI right now is flowing most reliably to the people who sell the picks and shovels, and to the people who got in early enough to sell equity instead of product. Nvidia makes the chips. Arista makes the networking gear that stitches the data centers together. Amazon rents the capacity. None of those businesses depend on your AI agent actually working. They get paid when the training run happens, whether the resulting product is useful or embarrassing.
Meanwhile, the tools layer — the chat wrappers, the autonomous agents, the “AI employees” with pricing pages that start at $499 a seat — is where the risk lives. That’s the part of the stack that has to prove it does something a human wouldn’t do faster and cheaper. And that’s the part I keep finding held together with prompt duct tape.
What this is not
I want to be careful here, because the internet will absolutely turn this into a crash narrative, and I don’t have the facts to support that. A few things I genuinely cannot tell you from the available information:
- Whether any of these sales were pre-scheduled trading plans set up months in advance, which is routine for executives and directors
- What percentage of each person’s holdings the sales represent
- What any of them believe about the next twelve months
- Whether this quarter’s totals are unusual compared to prior quarters at the same companies
Insider selling is one of the noisiest signals in finance. Early investors diversify. Directors fund foundations, estates, taxes, houses, and other large boring things. Treating a Form 4 as a prophecy is how retail investors talk themselves into bad decisions.
The useful takeaway for people who buy AI software
Stop treating company valuation as product validation. Those are two different scoreboards, and the AI era has blurred them into mush. A vendor’s funding round, its chip supplier’s stock chart, and the net worth of its board members tell you nothing about whether its agent can reliably close a support ticket without hallucinating a refund policy.
So use the same discipline on tools that you’d want from someone managing your money. Ask what the thing does on your actual data. Ask for failure rates, not demo videos. Run a two-week pilot with a task you can measure. Cancel fast when it underperforms.
The people at the top of the insider-selling list already know how to convert AI enthusiasm into cash. You probably can’t do that. What you can do is refuse to pay for enthusiasm in the first place.
🕒 Published: