Skip to content
[ 05 ]Journal

The IPO Window Is Open for Exactly One Kind of Company. Anthropic Is It. Firmus Wasn't.

Anthropic is showing investors its Q3 numbers today ahead of a November listing. Firmus pulled a $5 billion IPO last week. The market is not closed, it is selective, and the line runs between realized revenue and forecasts.

All writing
The IPO Window Is Open for Exactly One Kind of Company. Anthropic Is It. Firmus Wasn't.
Listen to this · AI voice8 min

The IPO Window Is Open for Exactly One Kind of Company. Anthropic Is It. Firmus Wasn't.

Today, at its headquarters in San Francisco, Anthropic is showing a room of institutional investors its third-quarter financials for the first time, according to Bloomberg and CNBC. It is the last step before what Reuters reports will be a November listing, and what could be the largest IPO in history.

Last Thursday, Firmus, an Australian AI data centre operator backed by Nvidia, Coatue, Blackstone and Jane Street, pulled its US$5 billion IPO on the Australian exchange. It would have been the second-largest share sale in the country's history. The company cited "market volatility and conditions" and said it would raise private capital instead.

Same week. Same sector. Same rising-rate environment. Opposite outcomes. The difference between them is the most useful thing a founder can learn about capital markets this month.

What Anthropic is selling

Reuters obtained Anthropic's confidential prospectus at the end of September, and the numbers are specific.

Revenue grew roughly twelve-fold in 2025 to nearly $4.6 billion. The operating loss widened to $8.06 billion from $2.98 billion. The reported net loss of about $42 billion included a roughly $34 billion non-cash charge tied to the rising value of financing that converts into shares, not money spent running the business. Compute and infrastructure cost $7.33 billion, triple the prior year and more than half of $12.65 billion in total operating expenses. The company has committed to $518 billion of future cloud, computing and infrastructure obligations. It held $20.28 billion in cash and short-term investments at year end.

Two disclosures matter most for everyone else. Nearly a quarter of 2025 revenue came from two customers. And many of its largest clients are not on long-term contracts and could cut or stop spending.

The company filed confidentially in June. The Wall Street Journal has reported it could raise as much as $100 billion at a valuation around $2 trillion, more than double the $965 billion it was valued at in May. People familiar with the plans told CNBC its revenue could top $100 billion by the end of this year. If it prices there, it passes SpaceX, which raised about $86 billion in June at a $1.77 trillion valuation, popped 19% on its first day, and has since drifted back below its opening-day high while staying above the $135 offer price.

What Firmus was selling

Firmus planned to sell shares at A$11, which would have given it an equity value of about US$30.6 billion, nearly triple the US$10.5 billion it was valued at in a private round in early August. Add roughly US$30 billion of debt and the enterprise value was about US$60 billion.

Against that, the company had two leased data centres running, in Melbourne and Singapore, and plans to build five more. Its draft prospectus projected US$5 billion in annual earnings within five years. It had never built an AI data centre.

Then, two days before the deal fell apart, investors learned that escrow terms would let more than half the stock be sold by existing holders from day one. Orders that had covered the book started disappearing the next day. The company considered cutting the price and chose instead to go private.

Joseph Koh of Blackwattle Investment Partners, who looked at the deal and did not bid, said it was "a very big price tag for what would likely be expected to happen in the future assuming near flawless execution." Jun Bei Liu of Ten Cap called it "an important reality check for the AI investment boom," while adding that she would not read it as the end of the AI trade.

The difference is not AI. It is realized versus forecast.

Anthropic is walking into a room today with a quarter of actual revenue. It is enormous, it is unprofitable, and it is concentrated, but it is a number, and investors can argue about the multiple. Firmus was asking investors to pay triple the price of two months ago for a forecast, with the insiders free to sell on day one.

The market around both of them has turned selective rather than closed. Oura delayed its listing. Holtec Nuclear suspended its IPO citing poor sentiment. OpenAI pushed its own debut to 2027. Last week, confusion over OpenAI's revenue, after it told investors a $50 billion figure that stripped out cloud partner sales rather than the $70 billion that had been widely reported, dragged down chip stocks and the Nasdaq. Broadcom's Hock Tan said on Monday that with rates climbing, access to capital could become the biggest constraint on AI companies. The Fed's next meeting is two weeks away and this morning's inflation report made another hike more likely.

In that environment the window is open for companies that can show the number and closed for companies that can only describe it.

Why this matters if you are nowhere near an IPO

Anthropic's multiple becomes everyone's comp. Whatever it prices at, every AI company's next round will be negotiated against it. If the largest, fastest-growing company in the category gets a certain multiple on realized revenue, companies with slower growth should expect a discount from that, not a premium on a forecast.

If the biggest lab has to disclose customer concentration, so will you. A quarter of revenue from two customers is a risk factor in a $2 trillion prospectus. It is a deal-breaker in a Series A. Fix it before someone asks.

The compute commitment is your tailwind. Half a trillion dollars of contracted infrastructure spending means the labs need volume. The price of intelligence per token is going to keep falling because the people selling it have fixed costs to cover. Build on top of that, as we argued in AI Just Moved From Equity to Debt, and let someone else carry the capex.

If you are building infrastructure, Firmus is the lesson. Investors will no longer fund a build-out on a forecast, especially one where early backers are selling. Contracted revenue, built capacity and locked-up insiders are now the price of admission, in public and private markets alike. The IMF said as much yesterday when it listed an AI correction among its top near-term risks and pointed to $650 billion of debt sitting behind circular financing deals.

Plan your exit around a selective market, not a shut one. As we wrote in 51 Funds Still Writing Early Checks, capital is available for companies that meet the bar and scarce for everyone else. The same is true of exits. Strategic acquisitions and secondaries will carry most companies through 2027. An IPO is for the handful that can show a quarter like the one Anthropic is showing today.

The honest version

Anthropic could price at $2 trillion and trade down, as plenty of large IPOs have. Today's meeting could go badly and push the timeline again. The company's own prospectus devotes pages to the risks of the technology it sells.

None of that changes the lesson. In a rising-rate market, investors pay for what you have done and discount what you say you will do. Build the number first.

Sources

[ Next ]Keep reading