Anthropic's IPO Filing Will Expose AI's Real Cost Structure

The Claude maker confirmed on Monday that it had submitted the draft Form S-1, according to a statement published on the company's website. Anthropic did not disclose the size, timing or valuation of the proposed offering.

"The proposed initial public offering will depend on market conditions and other factors," Anthropic said in its statement.

The filing follows a $65bn (£48bn) funding round that valued the AI safety company at $965bn post-money, as reported by City AM, placing it ahead of OpenAI's last reported valuation. Anthropic was founded in 2021 by siblings Dario and Daniela Amodei alongside a group of former OpenAI researchers who left to build a rival focused on AI safety and enterprise adoption.

What the S-1 filing actually means

A confidential S-1 is a draft registration statement filed with the SEC under the JOBS Act. It allows a company to begin the regulatory review process without making its financial disclosures public immediately. The document will eventually be made available to investors ahead of any roadshow, and it must contain audited financial statements, risk factors, executive compensation details and a full breakdown of revenue sources.

For Anthropic, the S-1 will be the first time its financials face independent, public scrutiny. Until now, revenue projections and profit targets have circulated through press reports and investor briefings, none of which carry the legal weight of an SEC filing. According to recent reports cited by City AM, Anthropic expects revenue to more than double this year and is targeting its first operating profit. Neither figure has been independently verified in public filings.

The number of shares to be offered and the price range have not yet been determined, according to the company's statement. That means the timeline remains uncertain; confidential filings can sit with the SEC for months before a company decides whether market conditions justify proceeding.

Enterprise revenue under the microscope

Around 80 per cent of Anthropic's revenue is understood to come from business products rather than consumer subscriptions, according to City AM's reporting. That ratio makes the S-1's revenue disclosures especially significant for procurement teams and finance directors at companies already using Claude.

Enterprise-heavy revenue is, in principle, stickier and higher-margin than consumer subscription income. But it also concentrates risk. A small number of large contracts, or a heavy dependence on a single distribution channel, can make headline growth figures look more fragile once the detail is laid bare.

The S-1 will need to disclose customer concentration. SEC rules require companies to identify any single customer accounting for more than 10 per cent of total revenue. Given Anthropic's distribution model, that threshold could implicate one or more of its cloud-platform partners, not as end users but as intermediaries.

For operators building on Claude, the critical question is margin. If Anthropic's enterprise gross margins are substantially lower than those of a typical SaaS business, because compute costs per query remain high or because cloud partners retain a significant share of the sale price, that has implications for long-term pricing stability.

Cloud-platform dependency: risk or moat?

Anthropic's Claude models are sold to developers and corporate customers through cloud platforms run by Amazon Web Services, Google Cloud and other providers, as City AM reported. Amazon and Google have both made major financial commitments to Anthropic. Microsoft remains OpenAI's most important backer.

This distribution model gives Anthropic reach it could not have built alone. AWS and Google Cloud have existing billing relationships, compliance certifications and enterprise sales teams that would take years and billions of pounds to replicate. In that sense, cloud-platform dependency functions as a distribution moat.

But it also introduces structural risks that the S-1 will need to address. First, revenue recognition. Anthropic is understood to recognise some cloud-platform sales on a gross basis, while OpenAI reports certain cloud sales on a net basis, according to City AM. The difference matters enormously. Gross recognition books the full sale price as revenue, with the platform's cut appearing as a cost of revenue. Net recognition books only Anthropic's share. Two companies with identical economics could report wildly different revenue figures depending on which method they use.

Public-market analysts will normalise for this, but the distinction complicates any casual comparison between Anthropic and OpenAI. It also means that Anthropic's reported revenue growth rate could look very different once the SEC requires consistent, audited disclosure.

Second, there is the question of contractual terms. Cloud-platform agreements typically include minimum commitments, exclusivity windows and revenue-sharing arrangements. The S-1 will need to summarise the material terms of these agreements, giving the market its first clear view of how much of each pound spent on Claude by an enterprise customer actually flows to Anthropic.

OpenAI's parallel path

OpenAI completed its own conversion from a capped-profit structure to a for-profit corporation in 2025 and has signalled IPO ambitions. A direct financial comparison between the two companies remains complicated by the differing revenue-recognition methods described above. Both companies are competing for the same enterprise customers, the same cloud distribution slots and the same pool of investor capital. The order and timing of their respective listings will shape how public markets price the entire generative AI sector.

What UK operators should watch for

For UK and European businesses already consuming Claude or GPT through their cloud providers, the Anthropic S-1 will contain several disclosures worth reading carefully.

Compute cost per query. The filing will need to detail cost of revenue, which will include the price Anthropic pays for GPU capacity. If that cost is rising faster than revenue, it signals that current API pricing may not be sustainable.

Channel economics. The split between direct enterprise sales and cloud-platform resale will indicate how much pricing power Anthropic retains. A company that sells predominantly through intermediaries has less ability to raise or lower prices independently.

Geographic revenue breakdown. SEC filings require revenue to be broken out by geography. UK and European businesses will be able to see, for the first time, how much of Anthropic's revenue comes from outside the United States and whether international markets are growing faster or slower than the domestic base.

Customer concentration. If one or two cloud partners account for the majority of revenue, that is a risk factor for any business building critical workflows on Claude. A contractual dispute or a shift in a cloud provider's AI strategy could disrupt access or pricing with little warning.

Capital expenditure trajectory. Anthropic's spending on training and inference infrastructure will indicate whether the current generation of models is approaching cost efficiency or whether each successive model requires exponentially more capital. That trajectory affects how quickly AI tool pricing will fall, and therefore how aggressively UK firms should be locking in current rates.

None of these disclosures will arrive immediately. The confidential filing starts a process that could take months. But when the S-1 is eventually made public, it will be the most detailed document ever published about the economics of selling large language models to businesses. For any operator whose technology stack now depends on generative AI, it will be required reading.