Over the weekend, a post surfaced in my Hacker News feed with a headline I initially took for clickbait: “Owed a billion dollars in Nvidia stock.” Its author is Eric Gullichsen, invited onto Nvidia’s technical advisory board in 1993, when the company was selling not data-center graphics but hope. The billion-dollar headline turned out to be arithmetically exact—and that is perhaps the only thing in this story that adds up without a stretch.
It wasn’t the money that hooked me. I’m interested in what an error looks like when it passes for normal for 30 years simply because no one checks it. And there is one more thing: the central document in this story is not a contract for 25,000 options, but a short letter sent out of courtesy.
I’ll start with the mechanics, because without them the whole story sounds like a parable, when it is arithmetic. An option is the right to buy a share at a fixed price at a prearranged time. The quantity in the contract matters, but the second number determines a person’s life: the vesting schedule, the timetable by which their right to those shares comes alive. In 1993, Gullichsen received 25,000 options at $0.05 a share, grant number 007. The grant sheet, signed September 9, 1993, contains the sentence from which everything grew: the first 25 percent after 3 months from the grant date, then 25 percent in quarterly installments—so that by the end of the first year all the shares would have vested.
But there is a second document from the same period. In the letter officially inviting Gullichsen to become the first member of Nvidia’s technical advisory board, Jensen Huang, then president and CEO, wrote: “I am pleased to offer you an option for 25,000 shares which vests over 4 years.” 4 years is standard industry practice; 4 quarters is not. 2 documents from the same company contradict each other in almost every respect except the number of shares.
The context helps explain why no one took the 1996 letter to lawyers. Gullichsen was then running Sense8—one of the first virtual-reality companies—making rendering software, with clients ranging from NASA to the artist Jenny Holzer. He lived aboard the SS Vallejo, a floating house in Sausalito built in 1879 and once owned by Jean Varda and Alan Watts. That was where Kurt Prym, an Nvidia cofounder, brought Huang and Chris Malachowsky to see a demo: fast biquadratic texture mapping, a way to wrap a texture around a curved surface without losing speed. The technology went straight to the heart of the company’s first chip, the NV1, which rendered quadrilaterals rather than triangles.
Then came the mishap recorded in patent and corporate history: Microsoft released DirectX, which had no support for quadrilaterals whatsoever, and the NV1 was left without a software base. The company lost a significant share of its workforce, while Gullichsen had by then moved to the Kingdom of Tonga and gone into domains—alongside the crown prince, he was building up the national .TO domain, the first in the world to be marketed as an alternative to .com. That matters for the tone of the letter that reached him in 1996: it was written not by today’s trillion-dollar giant, but by a firm at rock bottom, brought back to life only by the RIVA 128 chip.
The letter of April 16, 1996, was signed by CFO Marcel Gani. Its contents were dry: the advisory relationship was ending; 15,625 options had vested at a price of $0.05; he had 90 days to exercise them; buying them all would cost $781.25. Gullichsen paid, completed the paperwork, and forgot about it for 30 years.
The number 15,625 is exactly 62.5 percent of 25,000—10 quarters of a 4-year schedule. No other schedule produces that share: under quarterly vesting, everything comes alive within a year. The difference is 9,375 shares which, under the signed grant sheet, should have vested before the letter was ever typed.
Then the stock-split multiplier kicks in. Nvidia split its stock 6 times: 2-for-1 in June 2000, 2-for-1 in September 2001, 2-for-1 in April 2006, 3-for-2 in September 2007, 4-for-1 in July 2021, and 10-for-1 in June 2024. Multiply them together: 480. 9,375 multiplied by 480 is 4.5 million shares. At the closing price on September 25, 2026 ($225.07), that comes to $1.01 billion. To get there, all it would have taken was a check for $468.75.
Formally, this should be the dullest section, but this is where the real drama lies. The papers turned up in 2024, when a trader friend was holding forth over lunch about Nvidia and Gullichsen remembered the folder. Lawyers Allan Steyer (Steyer Lowenthal) and Chris Burke (Korein Tillery) took the case on contingency.
A year of correspondence followed: Nvidia did not dispute the authenticity of the grant sheet. It argued something else—that the claim was time-barred. At a meeting with lawyers from Cooley, the company’s outside counsel, the answer translated into human language was: “Well, go ahead and sue.”
California’s statute of limitations for a claim under a written contract is 4 years (Code of Civil Procedure, section 337). For torts and mistake it is 3 years, but with a special caveat: the clock starts not when the violation occurred, but when the injured party learned the facts constituting the fraud or mistake (section 338). Gullichsen built his case on the latter route: he calls the 1996 letter a bad-faith representation, a species of fraud under the California Civil Code.
But there is no way to prove that Nvidia knew about the discrepancy rather than simply applying a standard template. Without that, the exception never kicks in, leaving those same 4 years counted from 1996. The attorneys judged the odds of surviving a motion to dismiss as nonzero but low, and the client agreed.
Commenters in the thread split about evenly, and both sides had a point. Some said: sell the right to sue to a third-party fund—this is lawful practice for contract claims, though prohibited by state law for personal-injury claims. Others replied that not even an optimist would find a buyer for a time-barred case.
A third group—and they seem closest to the truth to me—pointed to a hole in the construction of the claim. Even if a court found that the 9,000 options existed, it would not necessarily hand over the shares. It would assess what the person lost in 1996, not what it would be worth today.
One commenter carried that thought all the way to the size of the check: Nvidia would, in that case, issue a fresh option with a strike price around $225 and the same lifespan—compensating the option premium, not the path to the stock price. Others immediately objected that keeping short-dated options rolling for 30 years would have cost more than any compensation. There is no consensus in the thread, and I’m not going to force one.
This is where the rough spots begin, and I will honestly leave them rough. First, the grant sheet prints the exercise price as $0.50, while the CFO’s letter says $0.05. The latter is confirmed by the arithmetic on the check: 15,625 multiplied by 5 cents is exactly the $781.25 in the letter. So the grant sheet contains a typo. For a document underpinning a billion-dollar claim, that is an uncomfortable detail—it shows how carefully such papers were filled out in 1993.
Second, the grant sheet contains a provision that works against the plaintiff: any discrepancy between the sheet and the attached legal terms of the option is governed by the attached terms. Gullichsen did not publish those very terms—only the sheet, the letter, and the invitation are available. Yet they may be precisely where the 4-year schedule appeared, along with the option’s expiration period after the relationship ended.
Third, Huang’s letter. It explicitly promises 25,000 shares with 4-year vesting—which is exactly what the person received. That means the entire claim rests on the notion that a later clerical grant sheet overrode an earlier promise by the chief executive, and that the sheet itself was not overridden by documents sitting in the same folder. I do not know what to do with that.
Strictly speaking, the grant sheet may take precedence. But in human terms, everyone involved—the executive, outside counsel, the holder of the papers himself—shared the 4-year version for 30 years and acted accordingly.
And one final detail, the reason I took all this apart in the first place. The man who lost a billion because he did not reread his own documents does not look crushed at all. He simply notes that in 1996 the stock was worth 5 cents, and no one was obliged to believe in such a future. There is also a second check that was never written—those same $468.75. Perhaps that is the most honest price of the whole matter: the sum no one could be bothered to clarify.