The trillion-dollar AI IPOs: what the valuations would have to prove

The short version

A listing at $1.5–2 trillion would price Anthropic as if it were already one of the world’s most valuable companies, before it has shown a full year of profits under standard accounting.

• Anthropic’s reported revenue has grown exceptionally fast, from $4.6 billion in 2025 to $16.8 billion in the third quarter of 2026 alone, as reported from its draft prospectus and investor meetings.
• It reported heavy losses under standard accounting in the first half, and about $450 billion of noncancelable commitments to rent cloud computing over the next decade.
• Simple arithmetic shows that, priced as a mature company today, $1.5–2 trillion implies $240–320 billion of yearly revenue; giving IPO buyers 10% a year for a decade would take $620–830 billion by 2036.
• Past giant listings ran from Google (share price up about 150-fold) to Rivian (down 85%). From $1.5–2 trillion, a tenfold rise, a fraction of what early Google or Tesla buyers made, would mean $15–20 trillion, three to four times Nvidia today.

Where the IPO stands

Most of the numbers are reported rather than confirmed; the table separates the two.

	What we know	Status
Filing	Draft S-1 submitted confidentially to the SEC on Jun 1, 2026	Confirmed (Anthropic)
Public prospectus	Reported for as soon as Oct 23; as of Oct 24, no public S-1 from Anthropic appeared in the SEC’s EDGAR system	Reported; our EDGAR check
Timing	Roadshow from the week of Nov 9, with trading targeted before Thanksgiving	Reported (Bloomberg)
Valuation	About $2 trillion expected; some investors now discuss about $1.5 trillion	Reported (WSJ, Reuters)
Size of raise	Up to $100 billion	Reported (WSJ)
Banks	Morgan Stanley, Goldman Sachs, JPMorgan, Citigroup	Reported (Reuters)
Last private price	$965 billion after a $65 billion round (May 28, 2026)	Confirmed (Anthropic)

What Anthropic’s numbers show

Anthropic is growing extremely fast, spends heavily on computing power (nearly $9 billion in the first half, about 55% of revenue, before other costs), and its growth slowed in the latest quarter.

Measure	Figure
Revenue, 2025	$4.6 billion
Revenue, first half 2026	$16.3 billion (first half 2025: $1.1 billion)
Revenue, third quarter 2026	$16.8 billion, up 45% on the quarter (second quarter: up more than 100%)
Net loss, first half 2026	$80.7 billion, of which about $72 billion is a non-cash charge related to convertible notes and other financing instruments
Computing spend, first half 2026	Nearly $9 billion (all of 2025: $7.3 billion)
Noncancelable cloud-computing commitments	About $450 billion over a decade: $111 billion Google Cloud, $110 billion AWS, $31 billion Microsoft; up to $750 billion including deals in talks
Customer concentration	About 20% of first-half revenue from two customers

What the valuation implies

The question isn’t only how big Anthropic must get, but how fast. Three simple tests show both. None is a forecast: each assumes Anthropic eventually trades like a mature company at 25 times earnings and keeps 25 cents of profit from each dollar of sales, similar to large, established software companies. They ignore new shares and dividends. The 25 is an illustrative long-run multiple; for comparison, the S&P 500 traded at 18.7 times expected earnings on Oct 23 (FactSet). At $1.5 trillion, the first test gives $60 billion of profit and $240 billion of revenue.

Test (our arithmetic)	When	Value	Profit needed	Revenue needed	What it takes
Priced as if already mature, at $2 trillion	Now	$2.0 trillion	$80 billion	$320 billion	About 5× today’s ~$67 billion annualized run rate
An IPO buyer at $2 trillion earns 10% a year for 10 years	2036	$5.2 trillion	$207 billion	$830 billion	Revenue growth of about 29% a year, every year for 10 years
The same at $1.5 trillion	2036	$3.9 trillion	$156 billion	$620 billion	About 25% a year for 10 years

The first test can be compared with today’s market. Gartner expects the world to spend about $2.7 trillion on AI this year, about $1.48 trillion of it on AI infrastructure, a broad category including servers, chips, networks, cloud infrastructure and devices. Spending on AI software is about $462 billion, and on the AI models themselves about $28 billion. Priced as a mature company today, Anthropic would need revenue equal to about 52–69% of that AI software spending, a comparison of scale rather than a market-share estimate (our arithmetic).

Another way to run the numbers

New Constructs, a research firm, runs the numbers differently and sets a much higher bar. To justify a $2 trillion valuation, its model assumes Anthropic keeps 10% of revenue as after-tax operating profit, reinvests heavily (half of each year’s revenue increase at first), treats its $518 billion of commitments as liabilities and discounts future cash at 9.33% a year. On those assumptions it needs revenue of about $3.6 trillion a year by 2035, far above our ~$830 billion, partly because it assumes Anthropic keeps much less of each dollar as profit; its reinvestment, discounting and liability assumptions also differ, so the two calculations aren’t directly comparable. Its low case values Anthropic at $144 billion.

Which markets would have to grow

By one estimate Anthropic already led its main market in late 2025, so its growth would have to come mostly from that market getting much bigger, and from share gains and new products. Menlo Ventures, an Anthropic investor, estimated from a November 2025 survey of US enterprises that Anthropic earned about 40% of their spending on AI models through programming interfaces (APIs), ahead of OpenAI (27%) and Google (21%), and about 54% of coding-related spending.

How fast it would have to grow

The growth tests ask Anthropic to grow about 25–29% a year for 10 years from roughly $67 billion of annualized revenue. The biggest companies, once they reached that size, mostly grew more slowly. Amazon came closest; only Nvidia grew much faster, over two years, and its boom is AI spending itself, including spending by labs like Anthropic.

Company, from about $60–70 billion of revenue (our arithmetic, from annual reports)	Start	Starting revenue	Years	Growth per year
Amazon	2012	$61 billion	10	~24%
Google (Alphabet)	2014	$66 billion	10	~18%
Apple	Fiscal 2010	$65 billion	10	~16%
Microsoft	Fiscal 2011	$70 billion	10	~9%
Facebook (Meta)	2019	$71 billion	6 (to 2025)	~19%
Nvidia	Fiscal 2024	$61 billion	2 (to fiscal 2026)	~88%

What if AI does part of the work people are paid for?

The biggest bull case is that AI will take on part of the work people do, which would make its market far larger than software. Pay is the prize: US employee compensation was about $15.7 trillion in 2025 (BEA). Worldwide, labor gets about 52% of income, by the International Labour Organization’s estimate; applied to 2025 world output of $118 trillion (World Bank), that is roughly $60 trillion a year, our rough approximation.

Estimates of how much work could be automated, and how fast, vary widely. McKinsey estimated in 2023 that current technology could in principle automate activities taking up 60–70% of employees’ time, with half of today’s work automated around 2045, the midpoint of its range. Goldman Sachs estimated that year that AI could expose the equivalent of 300 million full-time jobs to automation and lift global output by 7% over a decade. MIT economist Daron Acemoglu estimated in 2024 that only about 5% of tasks would be profitably automated within 10 years, adding no more than about 0.7% to productivity.

Adoption is still early. Across Census Bureau survey periods from December 2025 to early May 2026, 17–20% of US businesses reported using AI in the previous two weeks. Stanford researchers found no widespread job displacement, but employment of 22–25-year-olds in the most AI-exposed jobs is about 19% below where it would be had it kept pace with peers in less-exposed jobs; the gap was measured through June 2026, and the researchers describe it as descriptive, not an estimate of how much AI caused. The Yale Budget Lab has found no clear evidence so far of AI effects on the labor market.

Even if AI does the work, much of the value can go to customers, as lower costs, or to rivals selling cheaper models (see the next section).

Geography could also shrink the base. Anthropic hasn’t disclosed revenue by region; a public prospectus would normally include it. Claude, like ChatGPT, isn’t offered in mainland China, and since September 2025 Anthropic has also refused companies majority-owned from China. Elsewhere, cheaper Chinese models are winning volume: on OpenRouter they handled 57–67% of tokens in mid-September across its US, European and “Global South” customer groups, and 67% in the Global South, CNBC reported. On the other side, nearly 80% of Claude’s consumer use comes from outside the US, Anthropic said in September 2025. US frontier models still attract more overall spending and companies still want them for complex tasks, CNBC reported, and the falling costs that cheapen rivals can also cut Anthropic’s own.

The table applies the same shares to three bases (our arithmetic). The world and high-income figures are rough approximations: World Bank output ($118 trillion and $77 trillion) times the ILO’s global labor share of 52.4%, or up to about 55% for high-income economies, itself an approximation. Unlike US employee pay, ILO labor income includes the self-employed. These are scale comparisons, not estimates of Anthropic’s market.

Share of labor income that becomes Anthropic’s revenue (assumption; our arithmetic)	World (~$60 trillion)	High-income economies (~$40–43 trillion)	US employee pay only ($15.7 trillion)
1%	~$600 billion	~$400–430 billion	~$160 billion
3%	~$1.8 trillion	~$1.2–1.3 trillion	~$470 billion
5%	~$3 trillion	~$2.0–2.1 trillion	~$790 billion

The tests above need $240–830 billion. On a US-only base, 5% (~$790 billion of US employee pay) roughly matches the top of that range. In the near term, a valuation like this rests on growth and the story; in the long term, it depends on how much of that labor value becomes Anthropic’s revenue and profit, and how fast.

What if AI gets much cheaper?

The tests above need revenue growth and a 25% margin, not steady prices. But if prices keep falling fast, Anthropic has to sell far more AI, or more valuable products, to keep revenue growing; cheaper computing could ease some of the pressure on margins.

Prices for a given level of capability have collapsed: Epoch AI found in March 2025 that the price of reaching a fixed performance level fell between 9 and 900 times a year, depending on the benchmark and performance level, with the fastest drops the most recent. Open-weight models, which anyone can download and run, add pressure. On Artificial Analysis’s intelligence index, as of Oct 24, the best open-weight model scored 46, against 58 for Anthropic’s Claude Opus 5.5; a closed model from China’s Moonshot AI, Kimi K3.1 Pro, scored 55. Rival models are also far cheaper: DeepSeek’s V4 Pro lists output at $3.96 per million tokens at peak hours, about a fifth of Claude Opus 5.5’s $20. These are output list prices, not the cost of the same task, which also depends on input charges and how many tokens each model uses. On the developer platform OpenRouter, Chinese models handled 57–67% of tokens in the week of Sept 14, across its US, European and “Global South” customer groups, up from 6–13% in February, CNBC reported. Token volume isn’t revenue share, though: CNBC noted that US frontier models still attract more overall spending.

In late 2025, businesses had been slower to switch: Menlo Ventures’ survey put open-source models at 11% of US enterprise use, down from 19% a year earlier, which Menlo linked partly to Meta’s Llama stalling; it also found enterprises particularly cautious about Chinese open-source models. On Oct 19 the US made public an order, signed earlier, requiring civilian federal agencies to remove DeepSeek from their systems.

Revenue is price times usage, so falling prices raise the bar. For Anthropic to grow revenue 29% a year (our arithmetic):

If the price per task falls each year by	Usage must grow each year by
10%	~43%
30%	~84%
50%	~158%

The bull case is that cheaper AI means far more use, the “Jevons paradox” Microsoft CEO Satya Nadella invoked in January 2025. Anthropic’s history is consistent with that so far: it cut the price of its top model by two-thirds in November 2025, and its revenue run rate still rose from about $9 billion at the end of 2025 to $65 billion by July. The bear case is commoditization: a lead in quality, which matters most today in coding and agents, has to be won again with every model generation.

How far can a giant IPO grow?

Past listings show that the price you start from limits how far a stock can run. Google’s share price has risen about 151-fold since its 2004 IPO; the same rise from $2 trillion, with no new shares, would mean about $300 trillion, against $5.1 trillion for Nvidia, narrowly today’s most valuable listed company (Oct 23 close; our arithmetic). These are selected listings that illustrate different outcomes, not a representative study.

Value at listing is at the IPO offer price as reported at the time; Palantir and Coinbase listed directly, and their values use Nasdaq or NYSE reference prices on a fully diluted basis, prices at which no shares changed hands. Values now are market values (shares outstanding) at the Oct 23, 2026 close. Price multiples compare split-adjusted closing prices with the same offer or reference prices; they are price ratios, not necessarily returns available to first-day buyers, and they exclude dividends, so they are not total returns. Worst fall is the largest peak-to-trough decline in daily closing prices through Oct 23, 2026.

SpaceX, the closest in size though very different as a business, listed at about $1.77 trillion in June. It jumped 57% in its first three trading days, then fell 49% by August, and now trades about 9% above its IPO price.

Why index funds could swing the price at first

Index rules could force buying soon after listing, though less than the headline value suggests. Under the Nasdaq-100’s fast-entry rule, in effect since May 1, a new Nasdaq listing that ranks among the index’s 40 largest companies by full market value is typically added after its 15th trading day, if it meets liquidity and other conditions, and timing can shift around Nasdaq’s quarterly reviews. The index weights by listed market value, not float, but caps companies with less than a third of their shares trading at three times their float: for a $2 trillion company, it would count as up to $300 billion if the shares available to trade were worth $100 billion. Invesco’s QQQ, the largest fund tracking the index, holds about $499 billion (Yahoo Finance).

SpaceX entered on July 7 at a 1.28% weight, with an estimated $4.3 billion of index buying (MarketBeat). It rose about 5% between the announcement and inclusion, fell 7% on its first day in the index and was 33% below its pre-inclusion price by early August (our calculation). Tesla rose 70% between its S&P 500 announcement in November 2020 and its inclusion in December, climbed another 27% by late January, then fell 36% by March. The S&P 500 requires 12 months of trading and profits under standard accounting, rules S&P declined to relax in June. Index buying can lift a price in the short run; the long run depends on the business.

If OpenAI lists too

Two trillion-dollar AI listings would test how much new stock the market can absorb. OpenAI was last valued at $852 billion in a private round in March, backed by Amazon, Nvidia and SoftBank. Its price talk has come down too: by Oct 23 it was reportedly seeking about $30 billion at about $1.25 trillion, down from $1.4 trillion, with talks ongoing (Bloomberg). OpenAI told investors its September annualized revenue was almost $50 billion, Reuters reported; that isn’t directly comparable with Anthropic’s figure, because it covers a different period and excludes sales through cloud partners.

Together with SpaceX, already listed at $2.0 trillion, the three would be worth roughly $5 trillion, about the value of Nvidia (our arithmetic). Those are values, not cash to be raised: SpaceX raised $75 billion, and Anthropic may raise up to $100 billion (WSJ).

Anthropic and OpenAI are on the revenue side of the AI build-out examined in our Big Picture on AI infrastructure; their listings would turn a private bet into a public one.

What has to go right, and what could go wrong

The case for the price rests on growth continuing for years; the case against rests on costs and commitments that don’t shrink if growth slows.

What has to go right:

• Growth has to stay very fast for years. Unexpected slowdowns can pressure valuations: after quarterly growth fell from more than 100% to 45%, some investors discussed lower prices, Reuters reported.
• Computing has to get cheaper per dollar of sales. Anthropic told investors it was profitable on an adjusted basis in the third quarter, Reuters reported, but that isn’t profit under standard accounting, and its standard-accounting result for the quarter isn’t known.

What could go wrong:

• Large noncancelable commitments reduce flexibility if demand disappoints: about $450 billion for cloud computing, the WSJ reported in October. Reuters reported in September at least $518 billion of broader computing and infrastructure spending, about 80% of it noncancelable or payable regardless of use.
• Rivals such as OpenAI and Google are well funded.

Governance limits outside shareholders at both companies: Anthropic’s co-founders are to keep 50.1% of the votes after listing, TNW reported, which PitchBook analyst Harrison Rolfes argues warrants a valuation discount, and OpenAI’s nonprofit foundation controls its operating company.

What the market and analysts think

Views split on whether the growth is worth the price, and most agree the starting valuation is the risk.

• New Constructs, which puts Anthropic’s 2025 operating margin at minus 175%, argues that open-source models make it hard for closed models to earn a profit.
• Notre Dame law professor Patrick Corrigan, who studies IPOs, told the Associated Press in June, in a report republished by Fortune: “The question here is whether the price investors are going to end up paying is going to match up to the substance and fundamentals of what AI is really going to do in the real economy and as a business.”
• IDC’s Tim Law told the AP: “I thoroughly believe the demand is there and will grow.”
• Dan Ives of Yorkville Ives told Yahoo Finance’s Brian Sozzi on Sep 29 that IPO delays won’t derail the AI trade: “We’re only 15% through the spending, the four to five trillion that’s going to happen.”

This article was researched and drafted with AI tools, including Anthropic’s Claude; Anthropic is the subject of this article. It was fact-checked in separate reviews, including one using an OpenAI model, and approved by our editors.