Nine thousand three hundred seventy-five options. Post-split, that becomes 4.5 million shares. At current prices, roughly $1.05 billion.
Two documents describe that grant. One says the options vest over four years. The other, signed, says they vest in full after one year. Both are dated, both are on file, and neither has been disputed as a forgery.
That is the entire case. Not a hack. Not a fraud. A mismatch between a letter and an agreement, thirty years old, now worth a billion dollars.
Nvidia's defense is procedural. Time-barred. And here is the detail that should stop you cold: Nvidia does not challenge the authenticity of the documents. Read that again. A company with one of the most expensive legal benches in America is not arguing the paper is fake. It is arguing the paper is late.
When a defendant skips the substance to reach the calendar, it is telling you something. The substance is expensive. The calendar is free.
The dispute traces to 1993. A technical advisor — Gullichsen — sits on what is described as a Technical Advisory Board, not as an employee. That distinction carries more weight than the headline number, because it fixes the legal relationship as a contract rather than employment. Equity incentive agreement. Securities law and contract law, intersecting.
The timeline is thin, and that thinness is itself evidence.
1993: options granted. 9,375 units.
April 1996: a CFO letter acknowledges 15,625 shares as vested at that time. Gullichsen accepts them. Buys stock. Does not object. Does not sue.
Then silence. Thirty years of it.
2024: the claim is "rediscovered." 2025 into 2026: lawyer letters cross. No complaint is filed. As of this writing there is no docket, no court, no judge.
Sit with the arithmetic for a moment. 9,375 options became 4.5 million shares — a 480x adjustment across three decades of splits. But the 1996 CFO letter cites 15,625. Fifteen thousand six hundred twenty-five divided into 4.5 million is 288x. Those two ratios do not reconcile onto a single cap table without an adjustment factor nobody has published.
That is not an accusation. It is a ledger observation. You cannot audit a contract you cannot read, and the plan documents, the board resolutions, and the governing-law clause have not been produced. The reporting is single-sourced. No judicial filing. No statement from Nvidia on the substance. No full text of the agreement.
This is the discipline I applied in 2017, when I spent three weeks reverse-engineering the 0x protocol's token swap logic ahead of its public launch. You do not analyze a contract by reading its summary. You read the commit history. Based on that habit, and on weeks spent dissecting the BlackRock and Fidelity spot Ethereum ETF prospectuses in 2024, I can tell you how this always resolves: the operative clause is never the sentence in the press release. It is buried in an exhibit, written in the passive voice, and cross-referenced three times. Anyone analyzing this case without the plan document is analyzing a rumor with numbers attached.
Whoever found the discrepancy did so by reading, not by remembering. Sleep is for those who can. The archive does not care what time it is.
Start with the clock, because the clock is the case.
California's four-year limitations period for written contracts sits at CCP §337. Delaware's is three years, at 10 Del. C. §8106. Nvidia is a Delaware corporation, and equity incentive agreements from Delaware entities almost universally carry Delaware governing law plus an exclusive forum clause. That clause is the single most consequential variable in this dispute, and the reporting does not mention it.
But the choice-of-law question does not change the outcome. It changes the reasoning. Three years or four, a 1993 grant and a 1996 acknowledgment sit far outside any limitations window. If the case were that simple it would already be over. It is not over, because accrual is a question of fact, not a stopwatch.
The discovery rule delays accrual until the plaintiff knew or reasonably should have known of the injury. Fraudulent concealment can toll it further. Gullichsen's entire theory of survival rests on a single word in his own timeline: rediscovered. Not discovered. Rediscovered — implying the information existed, was available, and was lost.
The accrual date, not the contract text, is the whole ballgame. Every dollar of that $1.05 billion rides on when a reasonable person in his position should have understood what he was owed. And the 1996 CFO letter is the event a court will most likely point to. He had the letter. He had a number. He acted on the number. He did not act on the discrepancy.
That brings us to the most underweighted document in the file. The 1996 CFO letter is not a smoking gun for Gullichsen. It is a poison pill.
Contract law has a doctrine called course of performance — how the parties actually behaved under the agreement. When one party performs and the other accepts without objection, that conduct informs interpretation. Nvidia's 1996 letter, followed by Gullichsen's acceptance and stock purchase without protest, is textbook course of performance. It feeds waiver and estoppel as well. Under California's parol evidence framework, evidence of course of performance is admissible even when the writing is integrated.
Silence is nothing in code. In contract law, silence is a signature. Thirty years of non-objection, layered on top of a documented reconciliation in 1996, is not a footnote. It is the spine of a defense Nvidia may not even need, because it already owns the calendar.
Now the document hierarchy, which is where this gets interesting for anyone who builds.
Corporations run on three layers of paperwork. The equity plan is the constitution — rarely read, rarely amended, always annexed. The board resolution is the statute — the actual authority to grant. The grant agreement is the operative code. And the offer letter? The offer letter is a press release. It is written before diligence, before counsel review, before the plan documents are finalized, by people whose job is to close a hire.
If the offer letter touted four-year vesting and the signed agreement said one-year cliff-less vesting, parol evidence normally bars the earlier document from contradicting the integrated writing. Notice the posture: here, that rule would favor Gullichsen, not Nvidia. The signed deal is the better document for him.
So why isn't Nvidia fighting it? Because time-bar is a complete defense, and a complete defense means you never open the archive. No discovery into 1993 board minutes. No depositions of directors who may be deceased. No forensic examination of thirty years of cap table migrations. The cost delta between a limitations dismissal and a substantive trial is measured in years and seven figures.
Which brings us to Cooley's response letter — the "so sue us" letter that has circulated widely. Read it as strategy, not arrogance. It does three things at once. It refuses. It dates the refusal. And it converts every subsequent month of Gullichsen's inaction into additional laches evidence. Every month of delay is another brick Nvidia keeps for free.
There is a fourth function nobody is discussing, and it is the one I find most interesting. Disclosure management.
If Nvidia settles privately, it is a line item. If Gullichsen sues, it becomes a public record. A material legal proceeding triggers disclosure obligations. A billion-dollar contingent liability has to be assessed under ASC 450 — accrued if probable and estimable, disclosed if reasonably possible. Against Nvidia's balance sheet, $1.05 billion is a rounding error. Against its narrative, it is not. "Public company squeezes early technical advisor out of vested equity" is a recruiting problem in the middle of the most competitive AI talent market in history.
A lawsuit converts a quiet problem into an indexed, searchable, permanent one. The rational strategy is therefore not to win the case. It is to prevent the case from existing. Delay here is not a legal tactic. It is a disclosure shield.
It is worth killing one meme. Some coverage has framed this as a regulatory matter — Rule 701 exemptions, disclosure compliance, SEC oversight. That is a category error. Rule 701 governs the issuance of options under a written compensatory plan. It does not adjudicate whether a plan was followed thirty years ago. The SEC does not open investigations into a private vesting dispute between a corporation and one advisor. The agency's bandwidth goes to fraud, insider trading, and disclosure violations. Framing a contract dispute as a regulatory failure is how you manufacture noise and call it analysis.
One more variable deserves a name. In the 1990s, venture-backed startups routinely used NVCA-style equity documents. Under those templates, four-year vesting with a one-year cliff is the standard. A one-year, cliff-less, fully-vesting grant for a part-time technical advisor is an outlier — not impossible, but anomalous. That anomaly cuts both directions. It supports the implicit Nvidia position that the signed agreement may have been a drafting artifact. It also means that if the discrepancy was systemic rather than singular, Gullichsen is not a plaintiff. He is a class. Nvidia's 1993 cohort of early contributors, advisors, and contractors is finite but not small. The real tail risk is not $1.05 billion. It is the archeology.
Now let me earn the blockchain half of this, because it is not decoration.
Vesting schedules are a primitive. They are one of the few financial constructs crypto actually industrialized rather than merely tokenized. Sablier streams compensation by the second. Superfluid runs continuous vesting as a state. Merkle distributors let thousands of claimants prove entitlement against a single root. Hedgey, Tokensoft, and the on-chain experiments coming out of cap-table vendors exist because paper vesting decays over long horizons.
The Nvidia dispute is structurally impossible in that architecture. Not harder to have. Impossible. A vesting schedule encoded as a state machine produces an event log. The event log is the course of performance. The 1993 grant, the board resolution, and the 1996 reconciliation all resolve to one state, or the state fails loudly at the first checkpoint. Code doesn't forget. Paper does.
But I am not here to sell you on-chain vesting as a solution, because it is not one.
Immutability cuts against you in precisely the scenario this case presents. If the 1993 grant had been a smart contract with the wrong cliff — four years written into code when the deal was one — there would be no remedy at all. No ambiguity to litigate. No offer letter to introduce. Paper fails slowly, and slowly means someone eventually finds it. Code fails absolutely, and absolutely means the wrong answer is final.
Then there is the upgradeability problem. Most vesting contracts in production sit behind proxies with an admin key, because every real organization needs to amend schedules, handle departures, and respond to corporate actions. Which means the immutable ledger is ultimately governed by a multisig. Which means the corporate secretary with the Safe is functionally the CFO with the letter. Same architecture. New syntax. Different failure mode, identical failure.
Add the proof cost. Proving entitlement without revealing grant terms is exactly what zero-knowledge proofs are built for, and exactly where the economics turn ugly. The marginal proving cost per beneficiary is tolerable at ten claimants and brutal at ten thousand. It is the same math that leaves ZK rollup operators bleeding on proving costs unless gas is back at bull-market levels. Elegance on a whiteboard. Margin destruction in production.
The consensus reading of this story is David against Goliath. Sympathetic older contributor, trillion-dollar corporation, procedural dismissal. Morally unsatisfying, legally predictable.
That is the wrong frame, and it is the frame that produces no information.
The interesting failure here is not Nvidia's conduct. It is the document architecture — the fact that a company capable of executing thirty years of splits, buybacks, index inclusions, and a market cap in the trillions cannot produce a single authoritative vesting record for a board-level advisor. That is not a legal defect. That is a cap table running on institutional memory.
The 1996 CFO letter is the most damaging document in the case, and Nvidia wrote it. Not because it establishes vesting, but because it proves the company maintained a vesting ledger and reconciled it. Which raises the only question that matters: what happened to that ledger between 1996 and 2024? Paper archives do not evaporate. They get migrated. Every migration is a chance to drop a row. Every system replacement is a chance to lose a column. Corporations lose their own history with more discipline than they lose anything else.
And then there is the rhetoric. "In a free country, you only have to honor your word for a short time." That is a moral argument wearing a legal coat. Statutes of limitation exist to protect evidentiary integrity and legal stability — not to reward forgetting. When a party moralizes a procedural defense it usually signals it lacks a substantive one. Here, the moralizing is coming from the plaintiff.
Which is itself a signal. When the argument migrates from the document to the country, the document has stopped helping. The chart is a symptom, not the cause.
Nothing has been filed. Watch the archive, not the docket.
The forward question for anyone building token vesting in 2026 is whether this decade produces a counterpart to the 1996 CFO letter — an independently verifiable vesting attestation, a published Merkle root, a signed schedule that a stranger can check in 2056 without asking permission. A handful of projects already do this. Most do not, because attestation costs money and optimism is free.
If they don't, they are writing the 2056 lawsuit today, in a jurisdiction that does not exist yet, against a defendant who will not exist either.
Signal over noise. Always.