OpenAI Asks Congress to Legalize an AI Slowdown — Its $280 Billion Forecast Can't Afford One

OpenAI is not publicly traded yet, so it has no ticker, no price, and no way for you to buy a share directly. But it does have a number it already expects Wall Street to believe: more than $280 billion in annual revenue by 2030. And in the same two weeks its chief scientist published a memo arguing the industry should slow down, people close to the company told WIRED that OpenAI has been asking Congress whether orchestrating an industry-wide slowdown would even be legal. Those two things sit awkwardly on the same balance sheet. This article is about why.

The receipt and its grade

Start with the document that actually exists. On September 6, 2026, OpenAI chief scientist Jakub Pachocki published a blog post, "An Alien Mind," arguing that no lab has solved alignment well enough to keep scaling at maximum speed, and that AI labs should adopt "voluntary slowdowns" until shared safety bars exist — which he expects to become not voluntary but "widely mandated safety bars" enforced by third-party auditors or government bodies. Four days later, WIRED reported that OpenAI had approached members of Congress for guidance on whether coordinating such an industry-wide slowdown would run afoul of antitrust law. Grade the leak: the blog is a filed, quotable primary source; the Congress request rests on anonymous people "close to the company" — a lead, not a verified finding.

The blog is a safety argument. The antitrust request is a legal question. But they are the same sentence, because the law that governs them is the same. When competitors agree to restrict output — and slowing the deployment of frontier AI is, whatever its motive, an agreement to restrict output — they walk into the Sherman Act's restraint-of-trade provision. You can be sincere about the motive and still trip the mechanism.

That is why a bill introduced on July 23 matters. The bipartisan, bicameral Collaboration on Adversarial Threats and Security Risks Act (CATS) would create a "narrowly tailored antitrust safe harbor" letting American AI companies share threat information and coordinate to delay or limit deployment of high-risk models, with written notice to bodies like the NSA's AI Security Center, and explicit guardrails that keep price-fixing and market allocation illegal. Put the before and after side by side:

• Before CATS: a lab coordinates with rivals to slow deployment and absorbs Sherman Act risk.
• After CATS (if it becomes law): the same coordination has a statutory safe harbor, with notice and good-faith conditions.

Read that table twice, because the person who benefits is the point. A bill that lets the industry coordinate its own pace is a mechanism by which the market leader can stop the race. And the market leader is the one spending the most to hold its lead.

The forecast that argues against the slowdown

Now the number that complicates the story. OpenAI closed a $7 billion employee tender in August 2026 that reaffirmed its $852 billion price tag, and its annualized revenue run rate passed $40 billion in July — roughly double where it ended 2025. The company is projecting revenue to exceed $280 billion by 2030, and it has predicted a $14 billion loss for 2026. It filed confidentially for an IPO on June 8, 2026, with Goldman Sachs and Morgan Stanley leading, and its CFO has said the IPO comes in 2027.

Read those together as one statement about incentives. A $280 billion forecast does not get built by slowing down. It gets built by shipping, scaling, and deploying compute as fast as possible every single quarter. A coordinated, industry-wide capacity limit is, in that framework, a threat to the very growth line the $852 billion valuation is anchored to — unless you assume the person who sets the pace can set it to favor themselves.

There is a reading of this that is pure sincerity: executives who believe their own product could become dangerous are asking a country to help them pump the brakes, even at cost to their own valuation. And there is a reading that survives a current-mechanics test: a company burning $14 billion a year to stay ahead of rivals is the one that most wants a legally sanctioned agreement not to outspend each other. The same clause that protects "safety" also protects the leader's margin by freezing out faster-moving challengers. Sincerity and self-interest are not mutually exclusive, and the dossier does not need to pick one. But an investor should notice that the newest, loudest voice for the industry throttle is the incumbent with the biggest cash burn.

The analogy laps at a known shape: competitors agreeing to limit supply to manage a market is a cartel, whether the commodity is oil, freight, or trained compute. That mapping explains the incentive structure cleanly. It misfires on enforcement, because you cannot easily meter "AI progress" the way you can count barrels — no auditor can verify that a lab "slowed down" except through the one thing that is checkable: its money. So the fuse on the analogy is attached to the same ledger the forecast is written on.

The innocent reading and the break condition

There is a genuine counterargument, and it is worth taking seriously. John Schulman — an OpenAI co-founder now at Thinking Machines — dismissed the antitrust excuse as "fake," arguing that antitrust does not prevent companies from jointly developing a pacing proposal. The broader point has merit: legal ambiguity itself is a form of throttle, because labs that fear a Sherman Act suit will quietly avoid coordination regardless. In that light, asking Congress is less a scheme than a plea for cover to do something everyone already wants to do on their own.

What would confirm the innocent reading? Passing the bill and then showing, in audited disclosures, that OpenAI's own capital spending and cash burn actually decelerated as a result. What would break the self-protective reading is the same evidence, moving the other way: OpenAI's internal forecasts have reportedly doubled its projected cash burn through 2030 even as it argued for a slowdown. If the public growth story keeps accelerating while the public safety story asks everyone to brake, the divergence is the finding.

Here is the honest end state, with each claim graded. That OpenAI's chief scientist called publicly for voluntary slowdowns: filed, quotable. That OpenAI asked Congress whether coordinating such a slowdown is legal: reported, anonymous-sourced, worth treating as a lead rather than a fact. That a bill exists to make the coordination legal: introduced, referred to committee, not yet taken up — and on current timing it may not clear Congress until after the next general election. That OpenAI is worth $852 billion on a trajectory to $280 billion in revenue by 2030: the company's own disclosure, and the premise its IPO will be priced on.

Nothing has changed legally as of this writing. A bill introduced is not a bill enacted, and a safe harbor that does not exist cannot protect anyone. So the near-term question for anyone watching the OpenAI IPO is not whether the slowdown becomes law. It is which direction the money moves while the question sits in committee — because whatever Congress does, the company's own forecasts are the one piece of paper it has already signed.