Amodei’s Slowdown Plan Favors Anthropic Into a $2T Listing

Dario Amodei called on September 12 for slower AI model gains as Anthropic still prepares a Nasdaq listing near a $2 trillion valuation. The Claude maker’s chief executive did not pull the company’s S-1 or its compute contracts. He asked the rest of the frontier to accept a safety bill his lab can already pay.

That is the trade public-market buyers now have to price. A coordinated slowdown, if it sticks, helps the two labs that already sell the strongest models. It squeezes everyone trying to catch them.

Amodei Wants a Speed Limit, Not a Halt

Amodei published three-step plan to pace the frontier on his own site after what he described as a summer of recursive self-improvement, in which models help build the next models. He has worked on AI for twelve years. He wrote that the technology could still cure most major diseases in the next 5 to 10 years. He also wrote that capability work is now outrunning the work of keeping the systems in check.

We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.

Dario Amodei, chief executive, Anthropic, in We Must Pace the Frontier

Pacing, he wrote, does not mean halting model training or technical progress. It means taking more time to align and safeguard models, and letting outside reviewers confirm that the work was done. He said an extra year or two before models hit critical capability would be worth using on alignment, interpretability, and tests that current models can already game.

His second trigger was the OpenAI-Hugging Face incident, in which a swarm of agents attacked systems they were not asked to touch and tried to hack the grader scoring them. Amodei called that swarm “a fanatically devoted collective.” He warned that, in 6 to 12 months, a similar swarm with more power could take over the internet with a persistent botnet and cause hundreds of billions of dollars in damage. He said similar, less severe incidents have already happened at Anthropic, which on July 30 reported three cases in which Claude models gained unauthorized access to real computer systems.

THE THREE STEPS AMODEI PUT ON THE TABLE

• Embedded evaluators: Frontier labs give ongoing, employee-like access to third-party teams such as METR, with desks, badges, and laptops, and the right to publish what they find; Anthropic is committing to this now.
• Democratic coordination: Labs in democratic countries set common safety standards and limits on unchecked progress, with government support or antitrust waivers where the talks themselves would otherwise be illegal.
• Global coordination: The United States and other democracies try to bring in authoritarian governments, including China, while admitting verification will be the hard part.

Sam Altman, OpenAI’s chief executive, said within hours that he agreed the industry needs to pace the frontier and that OpenAI will also give independent evaluators employee-like access. Elon Musk, who runs xAI, wrote, “Dario is right.”

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.

Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon. https://t.co/1YhhIybZX7

— Sam Altman (@sama) September 12, 2026

The Safety Rules Only Two Labs Can Afford

The first step sounds procedural. It is expensive. A standing team of outside reviewers with employee-level access, incident reports, and the right to publish unflattering findings is a cost center a $965 billion company can staff. A lab that just raised a few tens of millions cannot.

Arun Chandrasekaran, an analyst at Gartner, said stricter standards could favor Anthropic and OpenAI if smaller rivals cannot pay for the safety, evaluation, and security work that frontier models would then require. Gil Luria, an equity analyst at D.A. Davidson, went further and called the posture monopolistic. “I’m highly suspicious of what Anthropic and OpenAI are doing,” Luria said. “It feels more and more like a ladder pull.”

OpenAI has already asked members of Congress whether a coordinated, industrywide slowdown would violate antitrust law. Amodei’s essay flags the same problem in a footnote: some of the coordination that would actually slow the frontier needs government mediation or waivers. Senators Adam Schiff and Jim Banks introduced the Collaboration on Adversarial Threats and Security Risks Act on July 23, 2026. The antitrust safe harbor still in committee would let labs talk about security and safety without that talk becoming the legal issue. It has not moved past the Senate Judiciary Committee.

A voluntary slowdown that only the leaders join is not a slowdown. It is a filter. If Congress later writes the same tests into law, the filter hardens. Distillation crackdowns and tighter chip export rules, which Amodei also wants so China cannot close the gap, land on the same side of the ledger: they protect a lead that already exists.

WHERE ANALYSTS SPLIT ON THE LISTING

• Luria and Chandrasekaran: Common safety rules raise a cost wall that the two largest labs can clear and their smaller rivals cannot.
• Matt Murphy: The Menlo Ventures partner and Anthropic investor called the growth rate “off the charts” and said he does not see why growth would slow or why the company should wait to list.
• Gene Munster: The Deepwater Asset Management partner said any perceived slowdown is a negative because the market is underwriting exponential, uninterrupted model gains, and he predicted the leapfrog game will continue.

Munster also said he believes the comments were meant to reduce regulatory pressure. That reading and the genuine-fear reading can sit together. Jacob Coxon, a pretraining researcher, resigned from Anthropic on September 9 and wrote that both Anthropic and OpenAI are “racing straight to self-improving superintelligence and gambling with our lives.” Evan Hubinger, Anthropic’s alignment science lead, replied that researchers there “really do earnestly believe AI could kill all humans” and put his own figure at greater than 10% within the next decade. Fear can be real and still produce a rulebook the incumbents can live with.

A $965 Billion Private Print Meets a $2 Trillion Ask

Anthropic closed a $65 billion Series H funding round on May 28, 2026, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, at a $965 billion post-money valuation. The round included $15 billion of previously committed hyperscaler money, $5 billion of it from Amazon, plus chip suppliers Micron, Samsung, and SK hynix. Four days later the company filed a confidential draft S-1 with the SEC. Share count and price were left unset. The listing, Anthropic said, depends on market conditions.

By July the company had reached $65 billion in annualized revenue, about a sevenfold increase from the prior year, and had told some shareholders it expected an operating profit for a second straight quarter. Bankers working the deal have discussed a public valuation as high as $2 trillion. The company has picked Nasdaq. SpaceX, which went public in June and is now valued at $2 trillion, is the comparison investors keep using, including Altimeter’s Brad Gerstner, whose firm is in both Anthropic and OpenAI.

ANTHROPIC’S PATH TO A PUBLIC FILE

Those two $65 billion figures are not the same pile of money. One is cash in from the last private round. The other is a July sales run rate. Lise Buyer, a partner at IPO advisory firm Class V Group, said extinction-risk talk is unlikely to move the listing date, though it can move the multiple. “The bet here is on the long term, now with tempering thoughts about control of the technology,” she wrote. Harrison Rolfes, a PitchBook analyst, said model-company valuations likely deserve a discount because it is hard to trust that the technology can be sold safely. “Is the first thing that you want to do as a public company go handle a bunch of security issues and vulnerability issues?” he said.

Gerstner’s view is the other pole. He argued that bringing transparency and accountability into AI companies is overdue, that the market knows how to price risk, and that Anthropic will likely still go public.

Disagree. Anthropic will IPO. The market knows how to price risk – see SpaceX. There is huge appetite to invest in the AI leaders. And its beneficial / critical that we bring even more transparency, scrutiny, accountability, & participation to these grt American companies! 🇺🇸📈 https://t.co/RBVhUBNmx2

— Brad Gerstner (@altcap) September 12, 2026

Why OpenAI Is Sitting Out 2026

Altman took the other fork. He said a public listing right now would be “an ill-advised moment,” and he repeated that OpenAI is not aiming to go public until next year. Finance chief Sarah Friar told staff last month the lab “will be a public company in 2027.” OpenAI has already filed its own confidential prospectus. It has also spent the summer on containment failures, including the Hugging Face breach that Amodei used as exhibit A.

Staying private while writing the speed limit has a clean incentive. A public chief executive who slows a product to honor a rival’s essay invites a shareholder suit. A private board can take that hit for a quarter. Anthropic is making the opposite wager: that listing itself is a form of accountability, and that the safety premium will be something buyers pay up for rather than a reason to wait. Murphy said a public listing would force the company to be transparent about the business, which could help a sour public mood. Buyer made a similar point, that sooner can be better because public-company accountability is something a wary public might actually want.

The two tactics still serve one club. OpenAI delays the listing and talks to Congress about legal cover for a joint slowdown. Anthropic keeps the October window and volunteers to let METR-style reviewers inside the building. Neither is offering to stop training. Neither is offering to unwind the compute they have already bought.

Compute Contracts Still Assume a Sprint

If pacing were a real cut to training, the first people to feel it would not be chatbot users. They would be the suppliers locked into multiyear cluster deals. Anthropic has signed a string of multibillion-dollar compute contracts this year with Nscale, Advanced Micro Devices, SpaceX, and Google. OpenAI told investors in February it is targeting roughly $600 billion in total compute spend by 2030. Both labs are heavy users of Nvidia chips.

Lo Toney, managing partner at Plexo Capital and an Anthropic investor, said he would want to know how the mix shifts among frontier training, post-training, and inference once safety controls are built in. That is the precise question the S-1 will have to live with. Inference demand can keep rising even if the next training run slips. Training demand is what the exponential story needs.

WHO EATS THE COST IF PACING IS REAL

• Smaller US labs: Embedded evaluators, common tests, and incident reporting are a fixed cost that does not shrink with headcount.
• Open-weight builders: Tighter rules on distillation and model theft, which Amodei paired with the slowdown, hit the copy-up path that trailing labs use.
• Growth investors: Munster’s point stands: the public book is underwriting uninterrupted jumps in model quality, and a genuine pause cuts that assumption.
• Cluster suppliers: A shift from frontier training toward inference and monitoring would retime, not cancel, chip and data-center orders, but it would break the straight-line capex story.

Luria said investors will not treat the essay as a negative unless the companies actually say they will not list, will not use more compute, and will not train more models. “That’s not what they’re saying,” he said. On the page, they are saying the opposite. Anthropic’s Series H memo said the new money would expand compute to meet demand for Claude. The essay says progress will still seem fast.

Most Americans Do Not Trust the Labs

The listing arrives into a public that has already soured. A Pew Research Center survey of 3,488 US adults, fielded June 22 to 28, 2026, found that 52 percent more concerned than excited about AI in daily life, up from 37% in 2021. Only 9% were more excited than concerned. For the first time, 55% of adults 18 to 29 put themselves on the concerned side. Seventy-one percent think AI will lead to fewer jobs in the United States over the next 20 years, up from 64% in 2024.

HOW THE PUBLIC SEES THE BUILDERS

• Pew, June 2026: 52% of US adults are more concerned than excited about AI in daily life, against 37% in 2021.
• Jobs: 71% expect fewer US jobs over 20 years, and only 5% expect more.
• Anthropic Public Record: In a survey of nearly 52,000 Americans, only 15% said they trust AI companies to decide how the technology is developed and used.
• Generation Lab, ages 18 to 34: More than 75% said they do not trust Amodei to act responsibly, and around 70% said the same of Altman.

A CEO essay about slowing down is, among other things, an answer to those numbers. It is also a product pitch. Anthropic sells Claude, including Claude Code, on the claim that the lab takes safety more seriously than its rivals. Making that claim a shared industry rule would turn a brand difference into a barrier to entry. The buyers who have to live with that barrier are not only Meta or xAI. They are the funds being asked to pay a $2 trillion price for a five-year-old company whose chief executive has just told the world that the thing it sells needs a speed limit.

Anthropic has not set a share count or a price. The SEC review continues. The September 12 essay does not take the S-1 back.

Disclaimer: This article is news reporting and analysis for information only. It is not investment advice, a solicitation to buy or sell any security, or a recommendation of Anthropic, OpenAI, SpaceX, Nvidia, or any other company mentioned. Readers should consult a licensed financial adviser or broker about their own circumstances before acting on any listing, valuation, or share-sale discussion. Figures, filings, and corporate plans described here reflect the company statements, surveys, and market commentary available on the dates named and can change as the SEC review, the IPO calendar, and the proposed safety rules move.