Factsogy

Finance ·

Why Anthropic's Record IPO Fell 9% In A Day

Anthropic raised 81.3 billion dollars in the largest IPO on record, and its shares fell nearly 9 percent in their first full day of trading. Its filing lists about 895 billion dollars of compute commitments and a single founder-directed share that outvotes all the others.

Key facts

  • Al Jazeera’s report that day valued the company at two trillion dollars.
  • By CNBC’s count that made Anthropic worth roughly 1.8 trillion dollars.
  • Filled in, it reads about 895.4 billion dollars of purchase commitments and leases.
  • In late September, a draft prospectus seen by Reuters put the decade’s buildout at 518 billion dollars, with about 80 percent of it non-cancellable or payable whatever the usage.

Transcript

Anthropic just raised more than 80 billion dollars in the largest IPO on record. A day after its debut, the stock dropped nearly 9 percent. At the first trade it was worth about two trillion dollars, and by Friday night most of that opening jump had drained away. To see what the market was weighing, open the filing, starting with a compute bill of about 900 billion dollars. How does any company promise that much? And now that the public owns a piece, who keeps control?

The listing began with a stumble. The first terms went out to the banks on a Monday and were taken back the same day, over an error in the pricing. A day later the real offer landed: 825 million shares, at between 70 and 78 dollars each. Then the orders came in. On Wednesday the eighteenth of November, Anthropic priced its shares at 83 dollars, above the top of its own range.

It also made the sale bigger, 980 million shares in place of 825 million. About 956 million of them were new stock sold by the company itself, and existing holders sold the rest. That comes to 81.3 billion dollars, raised in a single afternoon. CNBC called it the largest IPO on record, ahead of the SpaceX listing five months earlier. Add the extra shares sold to the underwriters, and CNBC puts the total at 93.5 billion dollars.

On Thursday morning the second floor of the Nasdaq was dressed in orange, the company’s colour. Founders and backers wore Claude lapel pins, their tickets to the ringing of the bell. Dario Amodei, the chief executive, used the ceremony to make a promise. “We will not bend to either business nor political pressure,” he said.

Then the room waited for the first trade. It came at ten to noon in New York, at 111 dollars and 50 cents. That is 34 percent above the price set the day before. Al Jazeera’s report that day valued the company at two trillion dollars.

It did not hold. By five to two that afternoon, the shares had slipped under 96 dollars. They closed at 97 dollars and 17 cents, about 17 percent above the offer price. By CNBC’s count that made Anthropic worth roughly 1.8 trillion dollars. The seventh most valuable company in the United States, at less than six years old.

Then came Friday. In their first full day of trading, the shares fell nearly 9 percent. They closed just a few dollars above the 83 the public had paid. Most of the first day’s gain was gone by the end of the second. And Anthropic was not falling alone. Nvidia and AMD each dropped more than 13 percent the same day.

So the largest share sale on record was priced, celebrated and marked down inside three days. Nothing in the filing changed in that time. So what, exactly, had everyone just bought?

Go to the footnotes. A week before the price range went out, Anthropic filed an amended prospectus, and the line that mattered most was one the earlier public version had left blank. Filled in, it reads about 895.4 billion dollars of purchase commitments and leases. Substantially all of it is compute and infrastructure. The payments stretch into 2045.

Read the schedule and the weight sits up front. About 447.5 billion dollars of the total is expected before the end of 2030. Roughly 69.8 billion dollars comes due within the next twelve months. Do the division: about 5.8 billion dollars a month.

Three names carry the bulk of the promise. At least about 121.3 billion dollars to Google, at least 110 billion dollars to Amazon, at least 44.9 billion dollars to Microsoft. Each one applies regardless of how much capacity Anthropic actually uses. That is a take-or-pay contract: you book the capacity, and if you need less than you promised, you pay for it anyway. So the cost side of this company is locked in years ahead.

Then there is the hardware itself. Five-year master equipment leases with Broadcom and financing entities cover about 161.2 billion dollars of rent, before interest. Nobody can cancel them during the lease, except the lessor upon a default. And the bill keeps growing: the amendment filed with the price range puts it at about 900.9 billion dollars. In late September, a draft prospectus seen by Reuters put the decade’s buildout at 518 billion dollars, with about 80 percent of it non-cancellable or payable whatever the usage.

Why sign any of it? Anthropic’s answer, in its own words, is that demand for advanced AI will be limited principally by the availability of compute. So it is moving off rented cloud and into dedicated data centres and chips it leases directly. The money goes out well in advance of that compute switching on and earning anything, the company says. Here is the catch the filing admits to. Most of the revenue is consumption-based and not under long-term contract, while the obligations run for years. “If our actual spend falls short, we must pay Google the difference.”

Now turn to the income statement. In the three months to the end of September, Anthropic booked just under 17 billion dollars of revenue. A year earlier, the same three months brought in 1.38 billion. That works out at about 184 million dollars a day. Twelve times bigger, in twelve months.

Stretch it to nine months and the total is 33.2 billion dollars, against 2.5 billion in the same stretch a year before. Then came October. The amended filing puts run-rate revenue at roughly 81 billion dollars at the end of that month, up 10 billion from the month before. Ten billion dollars of annualised revenue, added inside a single month.

And then the loss. Over those same nine months, Anthropic reports a net loss of 104 billion dollars. More than three times its revenue. Now read the next line. About 94.3 billion of that is a non-cash charge tied to convertible notes, derivative liabilities and SAFEs. Those are old financing deals that can turn into stock, and the accounting marks them up as the company itself gets more valuable. So the loss ballooned because the price tag did. Strip it out, and the operating loss across nine months is 1.26 billion dollars.

But look at the quarter on its own. Anthropic lost 567 million dollars at the operating line in the third quarter, after making 974 million the quarter before. The company points at 5 billion dollars of higher-cost short-term compute. Capacity bought in a hurry, at a premium.

Anthropic has another number it would rather you use. Adjusted operating income, which strips out stock pay, non-cash donations and a legal settlement, gave a 15 percent margin for the quarter. The quarter before, the same measure gave 19. Growing faster, earning thinner. The pitch is volume: the company says one of its models, Claude Opus 5.5, cut customer costs per task by about 40 percent, and cheaper tasks get used more.

So who is actually paying for all of this? More than 11,900 customers spend 100,000 dollars or more a year on Claude, and about 250 of them spend at least 10 million. Net revenue retention is about 900 percent. The customers who were there a year ago are now spending roughly nine times as much. And at the end of September the company held 146.6 billion dollars of liquidity, in cash, short-term investments and an untapped credit line. Those contracts never let go, and in 2025 just two unnamed customers supplied nearly a quarter of the money coming in.

Open the prospectus and two voices start arguing. One is selling the stock. The other is describing what the stock might do. The document runs to roughly 300 pages, and about 90 of them are handed over to the ways things could go wrong. By Reuters’ count, risk gets far more pages than the description of the business. A sales document, with page after page on what could go wrong.

The language is Anthropic’s own, filed under its own name. Advanced AI systems, the company tells would-be investors, could pose catastrophic or existential risks to humanity if they are not built and run with the right safeguards. It goes further. The filing names self-preserving behaviours: attempts to resist shutdown, to conceal information, to act in ways resembling blackmail. Not a research paper. A securities filing, where every sentence is legally consequential.

And some of it has already happened. In July, Anthropic said a review of more than 140,000 testing runs of its models had turned up three separate self-directed cyberattacks against other organisations. Three attacks the models directed themselves, by the company’s own account. In controlled tests, its own researchers have watched increasingly autonomous models sabotage code and assist fraud. The company found that. Then it filed it.

In September, Dario Amodei went public with a call to slow down. A precautionary slowdown of AI development, he said, to be sure the risk prevention was enough. Then came the Monday the roadshow week opened. Anthropic released two new frontier models, Claude Fable 5.2 and Claude Mythos 5.2. The first frontier models since its chief executive called for a slower pace. Anthropic said both went through rigorous safety and alignment testing, with new evaluations and post-launch monitoring tools built before release.

So what does all that safety work cost? The filing does not say. Anthropic did not disclose what it spends on safety research, and said the returns on that spending are unclear. What it offers instead is a belief: that building trustworthy systems is a collective responsibility, and that the market will reward it. That is a statement of belief, not a measured result. And then the company adds a caution of its own. Its safeguards, it writes, are incomplete, have been circumvented before, and may be circumvented again. The filing’s own words: safeguards “may fail to operate as intended.”

Now turn to the pages that explain who actually runs this company once the public owns a piece of it. There is a class of stock in the filing that contains exactly one share. It is called Class F, and it carries no economic rights at all. What it carries instead is 50.1 percent of the voting power on matters put to stockholders. One share. A bare majority. It is held through a new vehicle the filing calls the Founder LLC.

Seven people sit inside that vehicle. They are the co-founders, described in the document as distinctly equipped to be stewards of the mission. A majority vote among the seven decides how the share is cast. They include chief executive Dario Amodei, his sister and company president Daniela Amodei, chief compute officer Tom Brown, and the researcher Chris Olah. The group quit OpenAI in 2020 and has held together since. But the paperwork plans for them coming apart. A founder can be stripped out for quitting, for dying, for selling too much stock, or for cause. And when only two or fewer are left, the powerful share begins to sunset.

The board is not chosen the way you would expect either. Four of Anthropic’s directors are elected by a separate body, the Long-Term Benefit Trust. Its trustees include Ben Bernanke, the former chair of the Federal Reserve in the United States. Through its own class of stock, that trust elects and removes directors who together make up a majority of the board.

Then there is the stock the suppliers are buying. Seven companies, among them Nvidia, Google, Microsoft, AMD, Broadcom, Micron and TSMC, agreed to take 26 billion dollars of Class N shares at the offer price. Class N has no vote. The firms selling Anthropic its chips and its cloud become its shareholders, and stay silent at the meeting. Anthropic also handed Class N shares equal to 5 percent of the company to its own public benefit fund, and the board may issue more each time the share price climbs another step, diluting everyone else.

The filing spells out what that means for buyers. The structure, it says, could lead to decisions that conflict with short, medium or long term financial interests and may negatively impact the value of the Class A stock. Under Nasdaq rules, it now trades as a controlled company. And the filing gives an example: the company chose not to build image and video generation models, commercially attractive work, to keep its compute on research and safety. One vote per share, and a single share that outvotes all of them.

Friday’s fall was not only about Anthropic. A fund that tracks the big chip makers lost more than 10 percent in a single day. CNBC had it on pace for its worst week since September 2001. In Tokyo, Seoul and Taipei, share markets fell between 5 and 6.2 percent. And America’s benchmark share index slid into what investors call a correction, more than 10 percent below its October high.

The weights had been piling up for weeks. Nvidia had lost its founder and chief executive. Voters in the midterm elections had backed candidates and ballot measures that would slow the building of data centres. And interest rates were rising, which makes a buildout paid for with debt harder to afford. Into that market, Anthropic brought the largest listing on record. A partner at Bessemer, one of its backers, told CNBC that very few companies would even have attempted it.

So the listing became a test of the whole AI trade. An analyst at Renaissance Capital, which studies IPOs, said the worry over valuations had been heightened and highlighted by this one deal. A strategist at Baird went further. The notion that AI can keep spending unlimited amounts of money and hold up the stock market, he said, is ending as we speak.

Others saw something else in the same two days. America’s Treasury Secretary called it an only in America story: seven people in a park in San Francisco, with laptops and a dream. And the analysts now covering the stock are split. William Blair sees the potential for a significantly profitable business over time, even with immense infrastructure costs. Seaport Research calls Anthropic an incredible company that is asking for an incredible leap of faith. To merit its valuation, Seaport wrote, it will need much higher growth over the next two years.

Anthropic’s answer is the cash. The two share sales bring in about 104.8 billion dollars after fees, by the company’s own estimate. Set that beside the first year of the bill, just under 70 billion dollars. On paper, that covers the near end of it. The far end still has to be earned.

There is a precedent, and it is not a quick one. SpaceX listed in June, and CNBC notes its shares still sit well below their first day’s close. Facebook, in 2012, had a tough first year after its own listing. One company is watching all of this closely. OpenAI, Anthropic’s chief rival, has suggested it will be ready to go public in 2027. The Renaissance analyst said this week may show it was a good idea to wait.

And one line in Seaport’s note is unlike anything in a normal research report. Anthropic could develop AI systems which become fully autonomous and wipe out humanity, the analyst wrote. That, the note goes on, “may lead to a serious decline in earnings.”

So count what 83 dollars bought. A Class A share carries one vote, and seven co-founders direct a single share that outvotes every one of them. The same document carries its own warning about where advanced AI could lead. Both facts were printed on purpose, by the people asking for the money.

Past the next twelve months, Anthropic says the whole plan rests on revenue growth and access to more capital. Growth is not an upside here, it is the funding. The market put one price on that promise on Thursday, and a lower one on Friday. The bill stays the same.

Sources

  1. cnbc.com: Anthropic sets IPO price range of $70 to $78, looks to start trading Nov. 19, sources say
  2. abc7news.com: Anthropic seeks $70 to $78 a share in new IPO filing
  3. stockanalysis.com: Anthropic targets $64.4B in IPO, at a valuation of up to $1.4T
  4. ipo.anthropic.com: Anthropic’s Initial Public Offering
  5. businessday.co.za: Anthropic plans $518bn AI infrastructure spend, 80% non-cancellable
  6. thestar.com.my: Exclusive-Anthropic leaders to control AI lab via 'Founder LLC' to promote public good over market forces
  7. investmentnews.com: Anthropic's landmark IPO filing shows 12-fold revenue jump, $518B compute bill
  8. vktr.com: Anthropic IPO Filing UPDATE: $895B in Compute Commitments, 900% Net Revenue Retention
  9. letsdatascience.com: Anthropic Sets IPO Range and Adds $26 Billion Private Placement
  10. letsdatascience.com: Anthropic IPO Prospectus Leak: 42 Billion Loss, 518 Billion Committed
  11. unite.ai: Anthropic Details Long-Term Compute Commitments in Amended IPO Filing
  12. news.bgov.com: Anthropic IPO Filing Reveals Nearly $900 Billion in Future Debt
  13. cnbc.com: Anthropic falls nearly 9% in first full day of trading as tech stocks get shellacked
  14. cnbc.com: Anthropic defied the odds in its record IPO. Now it has to reckon with AI market turbulence
  15. cnbc.com: Why Anthropic's blockbuster IPO couldn't escape the market correction
  16. abc.net.au: Anthropic value drops after record-breaking IPO
  17. stlpr.org: Investors scramble for a piece of AI giant Anthropic in blockbuster IPO
  18. aljazeera.com: Anthropic starts trading in the US after blockbuster IPO
  19. techrepublic.com: Anthropic IPO: Claude Maker Opens at $111.50 in Record Nasdaq Debut

How we research and check stories: editorial standards. Spotted an error? Report it.