OpenAI kept the nonprofit and deleted the cap: Brockman's stake and the missing guardrail

The capped-profit arrangement had one property a counterparty could actually use: an arithmetic limit on how much the entity across the table could want. When OpenAI created OpenAI LP in March 2019, investor returns were capped at a predetermined multiple and the nonprofit parent kept governance control over the subsidiary ^[2]. The multiple was the safeguard people cited, and it was checkable in a way that intent is not.

Removing it was the stated purpose of the conversion. The account of Brockman's testimony describes the public benefit corporation as the route to conventional equity fundraising, with the ceiling lifted because the ceiling was what made some institutional investors hesitant ^[5]. A Delaware PBC puts a formal commitment to a stated public benefit where that ceiling used to be ^[4]. One of those is a number, the other a clause.

The arithmetic sits oddly against the founding story. OpenAI launched in December 2015 on a pledge of $1 billion ^[1], and by this publisher's own account the money never fully arrived ^[11]. The stake a single founder disclosed under oath is roughly thirty times that entire original pledge ^[10]. Set against a valuation already in the hundreds of billions ^[12], the same holding works out to about 6 percent of the company if it is worth $500 billion and about 15 percent if it is worth $200 billion ^[13]. Either way it is a founder-scale position, not a token.

Two things in the testimony can hold at once. Donation funding probably could not finance frontier compute, which is Brockman's argument ^[7]. The cure also pays him. The record as supplied gives no way to price the necessity separately from the windfall, and it was produced inside litigation that Elon Musk brought against Sam Altman and OpenAI ^[9], which is not a neutral venue for anyone's account.

What the source does not establish matters as much. It says the nonprofit OpenAI Foundation oversees the PBC ^[8] without specifying what oversight means in practice: which decisions need Foundation consent, who appoints whom, what happens when the public benefit and the equity disagree. Under the 2019 design part of that answer was economic, because the cap bound the outcome whatever the board did ^[2]. Under the current design the answer rests entirely on people and paper.

For anyone buying from OpenAI or underwriting it, that is the change worth marking. The safeguard being pointed at in 2026 is a stated purpose plus a board, and stated purposes are enforced only by parties with standing. So far the party pressing the question in court has been a hostile litigant, not a customer ^[9].