What does Jensen actually know that we don't? 🔍

Reading that Form 4 won't tell you what you think

“You saw the Form 4, right?”

That’s how a call opened Monday morning — a compliance officer I’ve traded notes with since 2016, three weeks out from a conference we’re both speaking at. He wanted to know if I was writing the Jensen-is-cashing-out piece everyone else was writing. I told him I wasn’t. Then I asked him what he’d noticed in the filings that the reporters missed.

He told me. And that’s what this post is about.

The thing everyone assumes

Here’s the received wisdom on Jensen Huang’s stock sales: insider selling from the CEO of the most important semiconductor company on Earth is a signal. He sees the top. He’s cashing out. The capex cycle is peaking, hyperscaler orders are about to normalize, and the man closest to the demand curve is quietly de-risking his personal balance sheet before the tape figures it out.

That’s the bear read.

The bull read is the mirror image: these are pre-planned 10b5-1 sales, mechanical, scheduled months in advance, meaningless as signal. Move on.

Both sides have a point. Let’s test them.

Testing assumption #1: “Jensen is selling the top”

Let me put a number on it. Since 2020, Huang has sold roughly $1.4 billion of Nvidia stock. That sounds enormous until you index it against what the equity’s done — NVDA is up more than 20x over the same window. Selling into a run like that isn’t timing the top. It’s diversification math any wealth manager would force on a client sitting on that kind of concentration.

Now the specific print everyone’s chewing on:

Huang’s Summer 2025 Sales Under the March 20 10b5-1 Plan · Dates · Shares · Proceeds

July 8–10, 2025 · 225,000 · ~$36.4M

July 18, 2025 · 225,000 · ~$38.7M at $172.05

First time here? Browse past issues

July 21–23, 2025 · 200,121 · $38.22M

July 24–28, 2025 · 202,797 · ~$39M

August 11, 2025 · 225,000 · $40.96M

The July total lands near $323M in the public reporting, with some outlets pushing the summer figure toward $500M once the August tranches are stacked on. Add the recent $180M block and you’re looking at a cumulative program that’s plausibly north of $700M for the year — and the plan authorizes up to 6 million shares through year-end, worth $865M–$925M depending on where the stock trades.

Here’s what the print doesn’t say: if Jensen were actually calling a top, he’d have front-loaded. He didn’t. The cadence is metronomic. Roughly 200,000–225,000 shares per tranche, filed like clockwork.

I’ve watched three CEOs try to time exits inside their own 10b5-1 programs over the years. It never looks like this. It looks jagged, with amendment filings and suspension notices. Jensen’s plan doesn’t have any of that texture.

Verdict on assumption #1: doesn’t survive.

Testing assumption #2: “It’s just a 10b5-1, ignore it”

This is where the bulls get lazy.

Yes, the sales are pre-planned. The March 20, 2025 adoption date is on the filings. That’s the legal shield and the practical explanation for the cadence. But the choice to adopt the plan — and its size — is itself a decision made with information. Huang set the ceiling at 6 million shares. He could have set it at 3 million. He didn’t.

Call it what it’s: a signal about how Huang models his personal exposure, not a signal about the next quarter’s data-center revenue. Those are different things, and conflating them is how retail gets whipsawed.

The read-through here’s behavioral, not fundamental. When a founder-CEO with north of $100 billion in paper wealth decides the right ceiling is $900M-ish of annualized selling, he’s telling you what “prudent concentration” looks like at his level. Nothing more.

Verdict on assumption #2: partially survives. The sales aren’t a demand-curve signal. But dismissing the size of the program entirely is intellectually cheap.

Testing assumption #3: “The last six sales predicted something”

This is the claim in the headline of every clickbait piece running this week. Let’s actually look.

The Last Six Huang Sales vs. What NVDA Did Next 30 Days · Sale Window · NVDA Direction Post-Sale

July 8–10, 2025 · Up

July 18, 2025 · Up

July 21–23, 2025 · Flat-to-up

July 24–28, 2025 · Up

August 11, 2025 · Up

Most recent ~$180M block · TBD

Five of the last five completed sales were followed by NVDA appreciation over the subsequent 30 days. If Huang’s sales “predict” anything, they predict continuation, not reversal. Which is what you’d expect from a pre-planned program running inside a bull tape.

The math doesn’t work for the “insider selling = top” thesis. Not on this data set.

The tell was always going to be in the plan structure, not the print dates. Anyone reading Form 4s as tea leaves is doing sell-side entertainment, not analysis.

Positioning risks, if you’re trading this

Risk Assessment for the NVDA Insider-Sale Narrative Trade · Risk · Likelihood · Impact · What To Do

Huang amends or suspends the 10b5-1 plan · Low · High · Watch for 8-K amendments; that’s the real tell

Sales accelerate beyond the 6M share ceiling via new plan · Medium · Medium · Reread the DEF 14A next spring

Hyperscaler capex guide cuts in Q1 2026 print · Medium · High · This is the actual risk, not the insider sales

Retail-driven volatility on each Form 4 filing · High · Low · Fade the day-of move if you’ve the stomach

The trade that kills you here isn’t Huang selling. It’s a Google or Meta capex guide-down in January that resets the demand story. That’s where the real read-through lives, and it’s nothing to do with what’s in Huang’s brokerage account.

What survives the audit

One thing, mainly.

The surprising fact isn’t that Huang is selling. It’s that a program built on modest-looking 225,000-share blocks — the kind of tranche a mid-cap CFO might unload — compounds into nine-figure quarterly totals and approaches a billion dollars annualized. That’s the scale of the equity. That’s what 20x appreciation on founder stock does to the arithmetic of “normal” diversification.

The insider-sale narrative is priced in, or close to it. The capex-cycle question isn’t.

If you’re spending your Sunday reading Form 4s instead of reading Microsoft’s cost-of-revenue footnote, you’re watching the wrong document. My compliance friend put it more bluntly on the call, but I’m not going to quote him on the record.

More on the capex read after channel checks next week.