OpenAI<\/a> in 2021 and built a company around a claim: that frontier AI is powerful enough to be hazardous, that the people building it should say so plainly, and that a lab willing to slow itself down is a lab worth trusting with the hard stuff.He has said versions of this on podcasts, in essays, in front of lawmakers, at his own company’s Builder Summit in Bengaluru in February. Last week Anthropic published a substantive risk report, per The Information, laying out what its researchers are finding about models that scheme and work against the goals assigned to them. That is not a document most companies publish weeks before the largest IPO in history.And it worked. Not despite the warnings. Because of them. The safety framing is what made Anthropic the vendor cautious enterprises and regulated industries could sign with, and the numbers are the proof.CNBC has confirmed the annualised revenue run rate hit $65 billion at the end of July, a sevenfold jump in a year, up from roughly $9 billion at the end of 2025. Second-quarter revenue crossed $11.5 billion against $787 million a year earlier. OpenAI, over the same three months, booked $6.7 billion.Then the prospectus lands, and every one of those warnings changes function.They stop being conviction and become legal artefacts, restated in the flat defensive register of securities law. Risks related to our models. Risks related to our regulatory environment. Once a warning sits in that section it is no longer a signal of seriousness, it is a shield against being sued for the thing you warned about.The audience changes with it. For five years Amodei was talking to governments, researchers and the public. From the day of the filing he is talking to shareholders, a constituency with exactly one question, asked quarterly.The financials make that question sharp. Anthropic posted a net loss of almost $42 billion in 2025, roughly five times the year before, on revenue of about $10 billion, according to documents seen by Bloomberg. Reuters has the second quarter delivering a first operating profit of $559 million, which is a real turn.But the spending behind those losses has not slowed. One compute agreement with SpaceX alone could be worth tens of billions of dollars over three years. A company burning at that scale has very little room to make an expensive decision on principle, and even less once the burn is public.<\/p>\nThe men warning about concentrated power are about to concentrate some<\/p>\n
Anthropic knows all of this. You can tell, because of what it is building to defend itself.The company has been preparing a class of stock with extra voting power for Amodei and his six co-founders, The Information first revealed, which would be the first time Anthropic’s leaders held voting rights beyond their ownership. The mechanism is ordinary in Silicon Valley. Meta, Snap and SpaceX all did it.What makes it unusual here is the arithmetic. Amodei owns about 2 percent of the company he founded, a person close to it told the same publication, his stake diluted by hundreds of billions in outside capital. That may be the smallest founder holding of any chief executive taking a company public in decades.Sitting alongside it is the Long-Term Benefit Trust, created in 2023, holding Class T shares with no economic value and one real power: electing the majority of Anthropic’s seven-person board. Ben Bernanke is a trustee. The trust is currently down to three members from its usual five, after Mariano-Florentino Cu\u00e9llar left this month to become Anthropic’s chief global affairs officer.Read that structure plainly and it is an admission. Anthropic is inviting the public markets in and bolting the doors from the inside on the way. Supervoting shares insulate the founders from shareholder pressure. The trust keeps the board.The company also stays a public benefit corporation, legally permitted to weigh society against returns, which will make it by an enormous margin the largest such entity on any US exchange. The next biggest is Veeva Systems, at $40 billion.Amodei’s defence writes itself, and it is not a bad one: the point of the shares is to protect the mission from precisely the short-term pressure the listing invites. But a man who has spent years warning that AI risks concentrating power in too few hands is about to be handed more of it.You also cannot argue that outside shareholders will push a company toward decisions it should not make and simultaneously argue that selling them the company is fine. Both cannot be true. The governance structure is Anthropic conceding the first one.<\/p>\n
The forecast and the warning describe the same two years<\/p>\n
The valuation math tells you what investors are actually buying, and it is not caution. Anthropic is projecting 2028 revenue of roughly $190 billion to $200 billion, Reuters has learnt, with bankers valuing the company off that number two years out. The comparison set includes Palantir, which trades at 53 times this year’s expected revenue, and Cloudflare and SpaceX, both around 41 times.Apply anything like those multiples and a $2 trillion debut stops being a fantasy. That is roughly where one investor landed when the news agency asked him, his question being not whether Anthropic could get there but whether it would stay.Hold that forecast against the safety argument. To go from $65 billion annualised to $200 billion booked inside two and a half years, Claude has to be inside almost everything, including healthcare and biology, which is where the company has told investors it plans to push next.That is not a scenario with a pause in it. That is maximum deployment at maximum speed, in exactly the window Amodei has described as the one where models get capable enough to be dangerous.He has already shown what a pause costs. In June, Anthropic temporarily disabled access to Claude Fable 5 and Mythos 5 to comply with a US export control directive citing national security authorities, and got them back roughly two weeks later after tense negotiation and some concessions.Earlier this year the Pentagon blacklisted the company after talks over military use collapsed on the limits Anthropic wanted, and the two sides are still in court. Investors have been pressing executives on exactly this in pre-IPO meetings, the Journal found, alongside cheap Chinese models and the backlash to data centre construction.Anthropic absorbed all of it as a private company, because the only people owed an explanation were investors already committed and reachable in a room. Do the same fortnight as a listed company and it is a material event, a stock move, a set of analyst notes, and a class of shareholders asking why management picked a fight it could have avoided.The willingness to lose money over a limit is what made the safety position credible. After the listing, that willingness has a price, it is visible to everyone, and somebody will eventually argue in court that paying it was a breach of duty.<\/p>\n
The contradiction does not get resolved, only priced<\/p>\n
Both things stay true, and one of them starts trading. The safety mission is not being abandoned, and the people running Anthropic are not cynics who stopped believing their own case. But an IPO does one thing above all others. It converts a private conviction into a number that trades, and hands the number to strangers.Anthropic has spent five years telling the world this technology is too important to be governed by people chasing quarterly returns. In September or October it will sell that world a large stake in itself and invite exactly those people to the table.The supervoting shares, the trust, the benefit corporation charter are all attempts to take the capital without the consequence. They will hold, or they will not. Nobody knows yet, because the first real test has not arrived.But it will. There will come a quarter when the safe decision and the profitable one point in opposite directions, and unlike every previous instance, the whole market will be watching which way Anthropic goes. Every buyer will have been warned before that quarter arrives, by the people who stand to make the most from being wrong.Anthropic said it out loud for five years. From the filing onwards, it says it on page 47.<\/p>\n","protected":false},"excerpt":{"rendered":"Anthropic accounts for around 60% of the combined revenue estimate of Starbucks and McDonald’s Every company that goes…\n","protected":false},"author":2,"featured_media":196856,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[152],"tags":[14682,88045,587,6283,51310,54567,1164,586,613,88046],"class_list":["post-196855","post","type-post","status-publish","format-standard","has-post-thumbnail","category-dario-amodei","tag-ai-revenue-growth","tag-ai-technology-risks","tag-anthropic","tag-anthropic-ceo-dario-amodei","tag-anthropic-ipo","tag-anthropic-ipo-2026","tag-claude","tag-dario-amodei","tag-openai","tag-safety-in-ai"],"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@people\/117134237591036723","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/posts\/196855","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/comments?post=196855"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/posts\/196855\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/media\/196856"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/media?parent=196855"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/categories?post=196855"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/people\/wp-json\/wp\/v2\/tags?post=196855"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"Anthropic](\"https:\/\/www.europesays.com\/people\/wp-content\/uploads\/2026\/08\/anthropic.jpg\" "\\"Anthropic")
Anthropic accounts for around 60% of the combined revenue estimate of Starbucks and McDonald’s Every company that goes public has to write down all the ways it might destroy itself.It is called the risk factors section. It runs for dozens of pages, it is written by lawyers, and almost nobody reads it. Bankers tolerate it. Investors skip past it to the growth charts. It exists so that when something goes badly wrong three years later, the company can point at page 47 and say the warning was always there, in the document you signed off on.Anthropic has been reading its risk factors section out loud for five years.That is the peculiar thing about what is coming. Anthropic expects to match or beat the size of SpaceX’s record IPO, according to Bloomberg, and could file publicly as soon as the end of this month. SpaceX raised $75 billion at the outset, rising to $86.2 billion once the overallotment was exercised, the biggest first-time share sale ever recorded.Anthropic is running the numbers on doing that or better, with Morgan Stanley, Goldman Sachs and JPMorgan Chase on the deal. The Wall Street Journal puts the debut at September or early October. The last private round, in May, valued the company at $965 billion. It was founded in 2021.None of that is the contradiction. Enormous IPOs happen. Fast growth happens, though rarely at this speed. The contradiction is that Anthropic’s whole reason for existing as a separate company from the one its founders walked out of rests on the argument that this technology is dangerous enough to require restraint. And restraint is the one thing a public market does not know how to price.<\/p>\n