OpenAI to Burn $278B by 2030; Anthropic Expects a Profit
Saturday, September 198 Min Read
A presentation seen by the Financial Times puts OpenAI's compute bill through 2030 at $856B — more than the $840B of revenue it expects over the same five years. A day later the Wall Street Journal reported Anthropic is targeting a November listing at roughly $2 trillion.
$856B
OpenAI Compute Spending Through 2030
$840B
Total Revenue Expected Over the Same Five Years
$278B
Free Cash Flow Gap Across Those Five Years
$2 Trillion
Valuation Discussed for Anthropic's Listing
In This Piece5 Sections
On September 18 the Financial Times was shown a presentation OpenAI had given to investors.
The next day the Wall Street Journal reported that Anthropic's IPO had moved to November.
The two companies are in the same business. The two documents say opposite things.
Two Sets of Books in Twenty-Four Hours
According to the presentation described by the FT, OpenAI expects negative free cash flow of $278 billion between 2026 and 2030.
Over the same five years it plans to spend $856 billion on computing power and infrastructure. In the presentation that is the company's single largest expense line.
The revenue side is there too: $36 billion this year, $350 billion in 2030. The five-year total comes to $840 billion.
Reuters reported that the company raised $122 billion in March at an $852 billion valuation, and that the money is on track to run out by 2028. OpenAI did not comment.
The WSJ reported that Anthropic moved the offering from October to November so it could show third-quarter figures to investors. The valuation under discussion is $2 trillion, and the raise could reach $100 billion.
The same report put annualized revenue above $65 billion at the end of July, up from roughly $9 billion at the end of 2025.
Accounts of the second quarter put revenue at $10.9 billion against an expected operating profit of $559 million. The second figure is about 5% of the first.
Where the Difference Comes From
The two products are close. The two sets of books are not.
The difference is not model quality.
It is who pays the data-center bill, and when.
Compute capacity is not bought on the day it is used. The building, the grid connection, the chip order and the lease are committed years ahead of the demand.
A company that takes that capacity onto its own books pays for a facility sized to 2030 traffic out of 2026 revenue. Cost lands early, revenue lands late.
A company that buys cloud by the unit does not carry that mismatch. Cost rises alongside revenue, and in exchange a share of that revenue goes to the provider.
Bloomberg, citing the FT, reported that Anthropic's gross margin runs above 80% before the revenue share paid to partners such as Amazon and before model training costs.
That does not make renting free. Anthropic carries multi-year cloud commitments of its own, and the revenue share sets a ceiling on its margin.
What changes is which balance sheet holds the commitment, and when the bill arrives relative to the revenue. More: What Is Cash Flow?
The November Calendar and the Record
Anthropic filed confidentially for an IPO on June 1.
The Biggest IPO Ever
SpaceX sold 555.6 million shares at $135 in June to raise $75 billion; the total reached $86.2 billion once the overallotment was exercised. The record before that was Saudi Aramco's $25.6 billion in 2019.
A $100 billion offering rewrites that list. Part of that money comes out of shares already trading: some of the cash that would have bought something else that week goes into a single listing instead. More: What Is an IPO?
Timeline
- June 1Anthropic files confidentially for an IPO.
- June 12SpaceX takes the record with $86.2 billion.
- September 13The FT reports Anthropic expects an adjusted operating profit this quarter.
- September 18The FT reports the $278 billion cash gap in OpenAI's presentation.
- September 19The WSJ reports the listing has moved to November at a discussed $2 trillion.
The Part That Does Not Fit
None of these figures comes from an audited statement.
OpenAI's numbers come from a presentation shown to a newspaper. Anthropic's come from what the company told shareholders and from unnamed sources. Both IPO filings are confidential.
Anthropic's profit is not quite a profit either. The measure the FT described is adjusted operating profit, and model training — the industry's largest expense line — can fall outside that measure.
Even the revenue run rate is disputed: the WSJ put July at $65 billion, while ARK Invest's August 17 report gave $74 billion.
On OpenAI's side, the $350 billion of 2030 revenue is a plan, not a result. The cash gap is the arithmetic consequence of that plan; if revenue falls short, the gap widens.
What Is Left
Neither company's shares trade anywhere. Their bills trade anyway.
An $856 billion compute budget shows up in the income statements of companies like Nvidia, Microsoft and Alphabet.
Here is how the single largest supplier of that budget has been priced over the past three months.
Over the same three months, the tech-heavy index fund traced this line.
The November listing will produce the first audited statement in this chain. Until then, every figure in hand is told on a yardstick the companies chose themselves.
This article draws on the Financial Times' September 18 report on OpenAI's investor presentation and on the Reuters and Bloomberg accounts of it, the Wall Street Journal's September 19 report on Anthropic's IPO and secondary accounts of it, Bloomberg's September 13 report on Anthropic's operating profit, CNBC's reporting on Anthropic's second-quarter revenue, and published comparisons of IPO sizes. Because both IPO filings are confidential, the figures come from documents described to reporters rather than from audited statements.
Companies in This Piece
Since The Event