Musk Lost on a Calendar. The AGI Clause Already Died.

On May 18, a unanimous nine-member advisory jury in the Northern District of California found that Elon Musk filed his lawsuit against OpenAI, Sam Altman, and Greg Brockman too late. Judge Yvonne Gonzalez Rogers accepted the verdict immediately. The breach-of-charitable-trust claims were time-barred under a three-year limitations window; the unjust enrichment claims were barred under two. MIT Technology Review reported that OpenAI argued Musk had been aware of the conduct since before 2021 for the trust claims, and before 2022 for enrichment; the jury agreed in roughly ninety minutes. Musk’s lead attorney Marc Toberoff called the decision “a calendar technicality” and pledged appeal to the Ninth Circuit.

Toberoff is right that it was a calendar finding. He is wrong that it was the only calendar that mattered. Three weeks before the verdict landed, on April 27, OpenAI and Microsoft rewrote the contract that had given the original lawsuit its underlying structure. The clause Musk most plausibly wanted unwound, the one that made the for-profit conversion a corporate-control question rather than a fiduciary one, was deleted before he could put it in front of a jury. The trial Musk got was not the trial his complaint was originally drafted against.

The verdict was about a date, not the merits

What the jury found was narrow. They were not asked whether Altman and Brockman had in fact breached fiduciary duties; they were asked whether Musk could be heard on the question this late. Statutes of limitations exist to prevent indefinitely deferred grievances, and the relevant California limitations periods are short. Musk donated $38 million to OpenAI in its early years, withdrew his support in 2024, and filed the original complaint in February of that year. Al Jazeera reported that Judge Gonzalez Rogers, accepting the advisory finding, noted that a “substantial amount of evidence” supported the limitations conclusion and that Musk faces “an uphill battle” on appeal.

The structural facts the case was built on are not in themselves frivolous. OpenAI was founded as a nonprofit in 2015. The first for-profit subsidiary discussions began in 2017. The capped-profit subsidiary was created in 2019. Microsoft took an exclusive GPT-3 license in 2020 and announced a $10 billion investment in 2022. OpenAI is now valued at $852 billion and is preparing for what would be one of the largest IPOs in history. A founder who put in $38 million when the entity was nonprofit and watched it convert into something on that scale has, at minimum, a colorable interest in being heard about the structure of the conversion. The jury did not say that interest was unfounded. The jury said he waited too long to assert it.

What the case was actually about underneath

The lawsuit’s surface claim was nonprofit-mission betrayal. The remedy Musk sought tells you what the litigation was actually structured to obtain: the unwinding of a 2025 restructuring and the removal of Altman and Brockman from their positions. That is not a damages action. That is a control action. The damages framing was instrumental, the structure was the prize.

What made the structure worth fighting over was a specific contractual provision that no longer exists. The original OpenAI-Microsoft partnership included what became known inside the industry as “The Clause”: a term that gave OpenAI the right to pull its technology back from Microsoft if and when OpenAI determined it had achieved artificial general intelligence. As Spyglass documented in its analysis of the renegotiation, the original clause created an asymmetric power dynamic favoring OpenAI, with the nonprofit board holding the authority to declare AGI unilaterally. A 2025 amendment narrowed that authority. The April 27, 2026 amendment replaced the clause entirely with a date: 2032. AGI declaration now triggers nothing. The contract runs on a calendar instead of on a definition.

Whoever held governance over the OpenAI nonprofit board between 2019 and April 2026 held a real piece of optionality on the future of the partnership. That is the thing Musk was suing to claw back into a structure where someone other than Altman and Brockman could exercise it. The lawsuit’s mission-restoration framing was the public-facing reason. The contract-clause framing was the operative reason. The April 27 deletion did not just defuse the AGI question commercially. It defused the entire premise on which Musk’s structural remedy depended.

The trial record itself made it harder to read the litigation as primarily mission-driven. OpenAI’s lead trial counsel William Savitt put on testimony and emails showing Musk recruiting OpenAI personnel for Tesla and Neuralink during the period he claims to have been working to preserve OpenAI’s nonprofit mission. MIT Technology Review reported that Judge Gonzalez Rogers pointedly noted xAI now operates in the identical competitive space Musk has spent the trial criticizing.

Then the distillation moment. Asked under oath whether xAI used distillation techniques on OpenAI models to train Grok, Musk did not deny it. He said “partly,” and characterized the practice as standard across the industry. Yahoo’s syndicated reporting described distillation as systematically prompting established models through their public APIs and using the response data to train competing systems — the technique OpenAI and Anthropic have both publicly opposed when Chinese labs have used it against them. The implication of Musk’s answer was not lost on the courtroom. The plaintiff arguing that OpenAI had abandoned its commitment to building AI safely was simultaneously confirming that his own AI company was using OpenAI’s models as training substrate.

What the compute math actually says

The deeper question the lawsuit was reaching toward is whether OpenAI’s current trajectory is rational on any frame other than the AGI race. The numbers say something specific. HSBC Global Investment Research projects total OpenAI compute costs of $792 billion between 2025 and 2030, with data-center rental alone at $620 billion. The bank’s revenue model puts OpenAI at $213 billion by 2030, on the assumption that paid penetration of consumer chat goes from roughly 10% today to 44% of the global adult population in five years. Even with that assumption, cumulative free cash flow stays negative through 2030. The funding gap relative to compute obligations is $207 billion.

The compute commitments themselves are concrete. OpenAI is on the hook for $250 billion to Microsoft, $38 billion to Amazon, multi-decade Oracle and Broadcom infrastructure spend, and a stated $1.4 trillion compute-capacity target by 2033. The 2026 operating loss is projected at $14 billion. The company is preparing this IPO into a P&L where compute commitments dwarf revenue and the path to profitability runs through penetration assumptions that no consumer software vendor has ever achieved at this scale.

There are two ways to read the spend. One reads it as a race to AGI: build the capacity, achieve the breakthrough, and the revenue model justifies itself retroactively because the product becomes economically irreplaceable. The other reads it as consumer- and enterprise-market capture: build the capacity to be the default infrastructure of the AI economy, and let AGI remain a marketing posture rather than a P&L event. The first read requires you to believe that AGI is plausibly within the IPO investor’s time horizon. The second only requires you to believe that AI inference at scale will continue compounding into a category that supports $200 billion in annual revenue by 2030.

Until April 27, the first read had a corporate-mechanics dimension to it. AGI declaration meant something contractually; it triggered structural consequences for the partnership; it was a thing to race toward as a definite, dated event. After April 27, AGI declaration triggers nothing. The contract runs to 2032 regardless. What this leaves OpenAI with is the second read by default: the spend has to make sense as market-capture economics, because the AGI corporate lever is gone. AGI is now a brand position, not a contractual instrument. That is a quieter story than the AGI race, but it is the story the contract amendment makes the spend defend.

The calendar Musk lost on, and the calendar he didn’t

Musk filed when AGI was a sword. He litigated after Microsoft helped sheath it. He lost on the statute of limitations on May 18. He lost the underlying lever on April 27. The two calendars he was operating against converged in a way his complaint was not drafted to survive.

The appeal will turn on whether the Ninth Circuit reads the limitations question more favorably to Musk than the trial court did. Even if it does, the trial he would get on remand is not the trial his original complaint anticipated. The unwinding of the for-profit conversion was structurally about restoring nonprofit governance over the AGI declaration. The thing being declared no longer carries the legal weight it carried in 2024. A founder’s quarrel about who controls a corporate trigger is a different quarrel when the trigger has been removed.

OpenAI walks out of this with the lawsuit shelved, the AGI clause retired, the Microsoft license terms reset to a date, and an IPO calendar that has now been visibly defended against the most prominent counterparty risk in the company’s history. The remaining counterparty risk is the math itself. The math gets harder once AGI stops being a contractual escape hatch and starts being a brand the public-market investor has to price.

AI-generated editorial illustration · TemperatureZero · May 19, 2026