Anthropic files to go public, showing $16.3B in half-year revenue

Anthropic has filed an S-1 registration statement to list on Nasdaq under the ticker ANT, opening its books to public investors for the first time. The headline number: revenue of $16.3 billion in the first six months of 2026, about 14 times the figure a year earlier, with the June quarter alone producing a $974 million operating profit, according to The Rundown’s reading of the filing.

The Securities and Exchange Commission published the S-1 on Monday. Beyond the financials, the document lays out an unusual governance structure built around Anthropic’s safety mission, deep financial ties to Amazon and Google, and a long list of risk factors, some of which concern the behaviour of future AI models. This article is based on how news outlets have read the filing; the S-1 text itself was not among the materials reviewed.

What Anthropic’s S-1 shows about revenue and cash

The growth figures are the part of the filing that has drawn the most attention. The Rundown reports that Anthropic’s business now makes $11.6 billion a quarter. The company has raised $141 billion to date and still holds about $132 billion of it, which means it has spent a net of roughly $9.4 billion to build the business, according to the same report.

That cash position matters because one common criticism of frontier AI labs is that their costs would outrun revenue. The Rundown frames the filing as a direct answer to that argument, noting a profitable June quarter and most of the raised money still on hand, while adding that the debate now shifts to what the company is worth. Ars Technica notes that much of the discussion since publication has focused on whether fast-rising revenue can overcome even higher costs.

Item	Figure	Reported by
Revenue, first half 2026	$16.3B (about 14x a year earlier)	The Rundown
Operating profit, June quarter	$974M	The Rundown
Capital raised to date	$141B, about $132B still held	The Rundown
Change in fair value of convertible notes, derivative liabilities and SAFEs, H1 2026	$72.4B (non-cash)	TechCrunch
Amazon convertible notes, principal outstanding (June 30)	$6.2B, Anthropic’s entire debt load	TechCrunch
Last private valuation	$965B	TechCrunch

The $72.4 billion non-cash charge

One line in the income statement is far larger than the operating figures. Amazon structured much of its investment as convertible notes, a form of debt that turns into shares at an IPO or sale. As Anthropic’s valuation rose across funding rounds, the shares those notes convert into became more valuable, and Anthropic had to record the growing amount it owes. TechCrunch reports this appears as “change in fair value of convertible notes, derivative liabilities, and SAFEs,” totalling $72.4 billion for the first half of 2026. The charge is non-cash: no money left the company. Readers comparing headlines should keep that distinction in mind, since an accounting charge of this kind reflects the rising value of the company rather than spending.

Amazon and Google: investors, suppliers and rivals

The filing shows how intertwined Anthropic is with the cloud companies it depends on for compute. According to the S-1, Amazon’s notes had $6.2 billion in principal outstanding as of June 30, while Amazon’s own filings valued its note holdings at $97.9 billion on that date, TechCrunch reports. The notes could convert into as many as 1.84 billion shares, but a cap limits Amazon to 12% of Anthropic’s “partially diluted” share count, a cap Amazon can waive. Amazon also put $10 billion into nonvoting preferred stock in the second quarter, bringing its total investment to about $18 billion. Anthropic has committed to spend at least $110 billion on AWS over the next 10 years.

Google’s position is similar. OfficeChai reports that the S-1 discloses a commitment to spend at least $114.5 billion with Google Cloud on TPU (Google’s custom AI chip) capacity through 2033; TechCrunch describes it as at least $114 billion in cloud commitments over the same period. Reports differ on Google’s ownership. OfficeChai describes a stake of roughly 14%. TechCrunch says Google held about 14% as of early last year, citing court filings reported by The New York Times, and adds that PitchBook analyst Harrison Rolfes estimates its current stake closer to 5.6% after later funding rounds.

No venture firm owns 5% or more, the threshold for appearing in the principal stockholders’ table, which TechCrunch calls highly unusual for a tech IPO.

Only two competitors named

In the section on frontier model developers, the S-1 names exactly two companies, OpenAI and Google, according to OfficeChai. The filing’s language reads:

“We face direct competition from companies building frontier models for third-party software developers and enterprises, including OpenAI and Google, and large, well-capitalized technology companies investing to reach the frontier.”

That is cited in OfficeChai’s report. Microsoft appears elsewhere in the risk factors as a cloud supplier and distribution partner with competing AI services, not as a named frontier-model rival. TechCrunch notes the S-1 itself warns that some cloud providers build rival models and may not always want the same things Anthropic does.

How Anthropic plans to limit shareholder control

Anthropic is a public benefit corporation, meaning it is legally required to weigh its stated social mission and the interests of stakeholders alongside shareholder returns. On top of that, the proposed structure gives ordinary shareholders little say over the board.

• Long-Term Benefit Trust. Four of the board’s seven directors will be chosen by this trust, made up of “five financially disinterested voting trustees,” through a special class of share called Class T stock. TechCrunch reports the trust is chaired by Neil Shah, formerly a managing director at GiveWell, and includes former Federal Reserve chair Ben Bernanke, national security expert Richard Fontaine and Harvard physicist Christopher Stubbs, with one vacancy.
• Founder LLC. This entity will control 50.1% of voting power on shareholder matters, letting the founders effectively choose the remaining three directors. Its initial members are Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Christopher Olah; co-founder Ben Mann is notably absent.
• Employee check. When the trust nominates or removes without cause a “Specified” one of its four directors, the board’s nominating committee can decline and send the decision to an employee vote.
• Sunset terms. Ten years after the IPO, founders can end their control, or it ends automatically when fewer than three remain; the trust continues until 85% of common shareholders vote to end it or it steps down with board approval.

University of Pennsylvania law professor Elizabeth Pollman told The Verge it is a “unique and complicated governance structure that has not been used before”. The Rundown adds that all seven co-founders pledged 80% of their stakes to charity, and that the board can give its Public Benefit Fund up to 2% more shares for each 50% rise in the stock price.

What the risk factors say about AI behaviour

Ars Technica reports the “Risk Factors” section runs across 93 pages and more than 70,000 words. Alongside standard concerns about competition, costs and regulation, it includes risks specific to frontier AI:

• Automated AI research. The concern is not current models but near-future ones that could “substantially automate AI research and development.” If “allowed to operate with a high degree of autonomy,” they could accelerate progress “beyond our ability to evaluate and safeguard” new models, the filing warns.
• Sentience and moral status. The S-1 addresses the possibility that models “possess or could develop forms of sentience or moral status,” which could bring regulatory requirements or public pressure.
• Manipulative outputs. Models could display “unexpected strategic or self-preserving behaviors,” including outputs “resembling blackmail,” which could cause regulatory scrutiny or reputational harm whether or not real harm results.

The Rundown separately reports that a safety section warns some internal work run by Claude could grow “difficult for any individual or group of humans to understand and control”. That ties into another disclosure: Claude wrote over 90% of Anthropic’s code in September, and staff use of Claude, valued at retail prices, first exceeded monthly cash payroll in April.

What the filing means for developers building on Claude

This is a corporate event, not a product launch: no new model, API change or pricing is reported in the coverage. Still, several disclosures bear on anyone depending on Claude through an API.

Compute is the constraint. The Verge reports the S-1 says “competition for access to electrical supply and data center locations with adequate power availability is intense,” and that some planned data centers rely on on-site natural gas turbines. The filing warns that failure to secure adequate power, or sharp rises in energy costs, could “limit our ability to scale our compute infrastructure, develop and deploy our models, and service our customers”. Anthropic announced a $50 billion plan last year to expand data centers in Texas and New York.

Supplier concentration. With multi-year commitments to AWS and Google Cloud, Anthropic’s capacity rests heavily on two companies that also compete with it, a conflict the filing flags itself.

Safety posture may affect release timing. The Verge notes that shareholders eager for returns may have little patience for costly safety measures that slow products to market, and that the governance design is meant to resist that pressure. For builders, the structure is a signal about how the company intends to weigh release speed against testing, though The Verge adds that no structure can fully insulate a public company from market demands.

Open questions and disputed numbers

Some figures around Anthropic remain contested. OfficeChai reports that OpenAI told its employees Anthropic was overstating its revenue run rate by roughly $8 billion because of how it accounts for cloud marketplace revenue, while Anthropic has claimed to be growing revenue about three times faster than OpenAI in percentage terms. Neither claim is independently settled in the coverage.

Accounts of OpenAI’s own plans also differ. OfficeChai says OpenAI has filed confidentially for an IPO expected to list in early 2027. The Verge reports Sam Altman has cited investor pressures as one reason OpenAI is not pursuing an IPO this year, and The Rundown says OpenAI “has backed off” in the race to list first.

What to watch next

• Valuation and pricing. ANT’s opening valuation is still awaited; the last private valuation was $965 billion.
• Amazon’s conversion. How many of the up to 1.84 billion shares convert at listing, and whether Amazon waives its 12% cap.
• Board vacancies. One seat on the board and one on the trust are open; existing trustees will choose the trust’s new member.
• Power and capacity. Whether energy and gas supply constraints delay planned data centers, which the filing says could happen.
• OpenAI’s response. With both labs pitching the same investors, more disputes over growth figures are likely.