$965bn, $1.5tn or $2tn: What Would Anthropic Need to Deliver?

Anthropic’s S-1 is public. What do $965bn, $1.5tn or $2tn ask of the business behind it?

The sector right now

Frontier AI is now priced like the largest listed companies on earth. Anthropic is the first of the US leaders to open its books as a standalone company.

•      OpenAI raised $122bn in March at an $852bn post-money valuation. Bloomberg reports it is in early talks to raise again at about $1.4tn. Its chief executive has ruled out a listing this year.

•      xAI took a different route. SpaceX absorbed it in February at a combined $1.25tn and listed on Nasdaq on 12 June at a $1.77tn valuation. The shares closed the first day at $161, then fell below the $135 IPO price within about a month.

•      Anthropic closed its Series H on 28 May at a reported $965bn post-money.

The S-1 names Google and OpenAI as its main competitors (p. 40). The companies are also tied to each other. Anthropic rents up to $120.5bn of compute from SpaceX's AI unit (p. 171), and Amazon and Google are at once its investors, its cloud suppliers and its rivals.

Most revenue figures in this sector are run-rates: a recent month multiplied up to a year. Audited numbers are rarer, and China got there first. Zhipu and MiniMax listed in Hong Kong on 8 and 9 January, raising about $560m and $620m, with audited accounts in their prospectuses; Zhipu priced at about $7bn. xAI's numbers now sit inside SpaceX's accounts. Anthropic's S-1 puts audited figures behind a valuation more than a hundred times Zhipu's, which is why it is worth reading slowly.

Revenue is the story

Anthropic booked $16.3bn of revenue in the first half of 2026, against $1.2bn a year earlier. For the whole of 2025 it was $4.6bn.

Costs are growing more slowly. Compute, the biggest cost, went from twice revenue to 55% of it in a year. The operating loss narrowed to $693m, from $2.1bn.

The $80.7bn loss is mostly paper

The first-half net loss was $80.7bn. Fair value charges explain it: $72.4bn from revaluing convertible notes held by Amazon, and $8.7bn from revaluing commitments to sell shares to Google and Amazon at a fixed price, $81.1bn in total.

The notes have $6.2bn of principal but sit on the balance sheet at $108.6bn, because they convert at $2–5 a share and gain value every time Anthropic's valuation rises.

Why $2–5? 

Amazon lent the money in 2023 and 2024, and the conversion prices were fixed then: $20.93 per preferred share for the 2023 notes and $50.48 for the 2024 notes, or $2.09 and $5.05 per common share after September's 10-for-1 split.

At the IPO, the notes convert at whichever is lower, that fixed price or the IPO price. They are still outstanding only because Amazon's stake is capped at 12%. Any part that would breach the cap rolls into a new note on the same terms

.

The charges show how much value is moving to those investors, not how the business is running. Take them out and the half-year shows a $360m profit.

The $3.9bn of cash flow, cleaned up

Operating cash flow was positive at $3.9bn. But two items flatter it:

•      $1.1bn of interest income, earned on the $131.7bn of cash raised from investors.

•      $2.1bn from drawing down prepayments made in earlier periods, mostly for compute.

That source runs out.

Strip those out and the business generated about $0.7bn in six months. After $0.5bn of capital spending, about $0.2bn is left. That is a business close to cash break-even, and it pays a further $2.2bn of staff costs in shares rather than cash.

What would $965bn need?

The S-1 does not state a valuation; $965bn is the figure under discussion since May. Anthropic pays no dividend, so every dollar of return has to come from the company becoming more valuable. Taking ten years as an example holding period, and an exit multiple of 37.5x earnings:

Annual return	Value needed in 10 years	Net profit needed at 37.5x earnings
8%	$2.1tn	$56bn
10%	$2.5tn	$67bn
12%	$3.0tn	$80bn
15%	$3.9tn	$104bn
20%	$6.0tn	$159bn

The 37.5x multiple is Aswath Damodaran's January 2026 figure for US system and application software, the nearest sector for a mature AI business. It still prices in about 23% expected annual growth - which may be elevated for AI in 10 years? The US market excluding financials trades at 28.9x.

The same sum works for any holding period: value needed = $965bn × (1 + annual return) ^ years.

For context, the US stock market as a whole is priced to return about 9% a year. The ten-year Treasury yielded 4.96% on 22 October 2026. Aswath Damodaran strips out a 0.23% US default spread, which leaves a risk-free rate of 4.73%. Add his 4.46% equity risk premium for the US and the market's cost of equity comes to about 9.2%. His own implied figure for the S&P 500 at the start of August was 8.97%. A single company still building its earnings would normally be expected to offer more than the market average.

What would $1.5tn or $2tn need?

The $965bn is only the first benchmark. It is the price private investors paid in May, not the price the market will set. If the IPO comes in at $1.5tn or $2tn, the bar for every buyer at that price moves up with it.

Annual return	Entry at $965bn	Entry at $1.5tn	Entry at $2tn
8%	$2.1tn ($56bn)	$3.2tn ($86bn)	$4.3tn ($115bn)
10%	$2.5tn ($67bn)	$3.9tn ($104bn)	$5.2tn ($138bn)
12%	$3.0tn ($80bn)	$4.7tn ($124bn)	$6.2tn ($166bn)
15%	$3.9tn ($104bn)	$6.1tn ($162bn)	$8.1tn ($216bn)
20%	$6.0tn ($159bn)	$9.3tn ($248bn)	$12.4tn ($330bn)

Each cell shows the value needed in ten years, with the net profit that value implies at 37.5x earnings in brackets. No dilution is assumed.

The jump is easy to miss. A buyer at $2tn who wants only 10% a year needs a $5.2tn company in ten years. That is more than the $3.9tn a buyer at $965bn needs for 15%. Going from $965bn to $2tn doubles the price before the first trade, and the business has to carry all of it - so will it?

The sector test: does the revenue fit?

The maths can work and the market still be too small.

So we ran a simple check: how much revenue would Anthropic need, against what the whole world spends on technology?

Three inputs

• The 37.5x exit multiple above

• A 25% net margin. That is close to the 26% adjusted net margin Anthropic reported for Q2 2026 (p. 160). It is an illustration, not a forecast.

• Global IT spending of $4.3tn in 2025, growing about 7% a year, as cited in the S-1 (p. 185). At that pace it reaches roughly $9tn ten years from now.

Annual return	Entry at $1.5tn: revenue needed in 10 years	Share of global IT spending	Entry at $2tn: revenue needed in 10 years	Share of global IT spending
8%	$345bn	4%	$461bn	5%
10%	$415bn	5%	$553bn	6%
12%	$497bn	5%	$663bn	7%
15%	$647bn	7%	$863bn	10%
20%	$991bn	11%	$1.32tn	15%

For scale, the S-1 says the ten largest listed technology companies earn just over $2.5tn of revenue between them (p. 185). Grown at the same 7% a year, that is about $5.3tn ten years from now. At a $1.5tn IPO, the 15% case asks one company to reach about an eighth of that; at $2tn, about a sixth. Add 4% annual dilution and the same case needs about $958bn of revenue at $1.5tn (11% of global IT spending) and $1.28tn at $2tn (14%).

The test has a blind spot, and Anthropic points straight at it. The company argues its market is the more than $40tn of knowledge work done each year (p. 185), not IT budgets. If AI is paid for out of salaries rather than software budgets, the ceiling moves up. If it has to win share from existing IT spend, it does not.

Who is selling?

Founders and employees sold $4.7bn of shares in April at $25.91 a share, and the IPO includes more sales by existing holders (S-1, Related Person Transactions, p. 284; The Offering, p. 18).

Who is selling, and who is buying?

This article is commentary on public information and is not investment advice.