Anthropic and OpenAI's Revenue-Accounting Split Becomes an IPO Flashpoint

When $47 Billion Might Not Mean What You Think

Two of Silicon Valley's most valuable private companies are racing toward public markets — and they can't even agree on how to count their money.

Anthropic filed a confidential draft S-1 with the SEC on June 1, 2026, days after OpenAI did the same. On paper, the numbers look staggering: Anthropic's run-rate revenue hit approximately $47 billion as of late May, according to data firm Sacra — up from roughly $9 billion at the end of 2025. Reuters reported a more conservative figure north of $30 billion as of April. OpenAI, meanwhile, is targeting a September debut at a valuation exceeding $1 trillion.

But buried inside both companies' financials is an accounting divergence that Wall Street analysts and SEC watchers are beginning to flag as a potential flashpoint: Anthropic and OpenAI book cloud-partner revenue in fundamentally different ways, and that gap could reshape how investors value both companies when the roadshows begin.

The Gross-vs.-Net Divide

The dispute centers on how each company treats revenue generated through hyperscale cloud platforms — AWS, Google Cloud, and Microsoft Azure.

Anthropic treats itself as the principal in transactions made through these partners. Under that framework, it books the full sale price as gross revenue, then records whatever cut goes to the cloud provider as a cost of sales or marketing expense. The result: every dollar a customer spends on Claude through AWS hits Anthropic's top line.

OpenAI takes the opposite approach. For sales through its Azure partnership with Microsoft, it records only its net share after Microsoft's cut. Revenue that passes through to the partner never appears on OpenAI's income statement at all.

The practical effect is enormous. Bank of America estimated that Anthropic's revenue-sharing payments to hyperscale partners could reach up to $6.4 billion in 2026 alone. Under OpenAI's net-reporting method, none of that would count as revenue. Under Anthropic's gross method, all of it does.

Across a five-year horizon the stakes are even higher: Anthropic has committed to paying Amazon, Google, and Microsoft at least $80 billion through 2029 to run Claude on their servers, according to company forecasts reported by Data Center Dynamics. How those flows are classified will materially affect every revenue multiple investors apply at IPO.

"Not Sure How the SEC Is Going to Let This Happen"

The accounting divergence has not gone unnoticed. Ethan Choi, a partner at Khosla Ventures — one of Silicon Valley's most prominent early-stage AI investors — put the regulatory question bluntly: "If they both IPO in the coming quarters, not sure how the SEC is going to let these two companies have different accounting treatment for essentially the same type of revenue."

That question is not merely academic. SEC rules under ASC 606 require companies to determine whether they act as principal or agent in a transaction — a distinction that drives the gross-vs.-net choice. Reasonable arguments exist for either classification, depending on how much control a company exercises over the service being sold. But the SEC has historically pushed for consistency across comparable business models, and having two directly competing AI companies present their top-line figures using incompatible methods creates obvious comparability problems for public-market investors.

If the agency forces Anthropic to restate gross revenue to a net basis before its listing, analysts estimate the headline annual run-rate figure could fall by 20 to 40 percent in a single filing — a significant headline hit even if underlying economics are unchanged.

Why IPO Pricing Turns on This

Revenue multiples are the primary lens through which high-growth tech companies get valued in public markets. A company reporting $47 billion in gross revenue commands a very different multiple-based valuation than one reporting, say, $30 billion on a net basis — even if cash generation and margins are identical.

Anthropic is targeting an IPO as early as Q4 2026, with bankers expecting the raise to exceed $60 billion. That ambition is backstopped by the company's June 2026 close of a $65 billion Series H at a near-$1 trillion post-money valuation. The gross-revenue methodology helps sustain a narrative of explosive growth; a forced restatement, even one with no impact on profitability, could cool investor enthusiasm at exactly the wrong moment.

For OpenAI, the dynamic runs in reverse. Its net-reporting approach looks more conservative by comparison, but if the SEC allows Anthropic's gross treatment to stand, OpenAI may face pressure — or choose voluntarily — to revisit its own methodology before its own listing to avoid appearing disadvantaged.

What to Watch: SEC Harmonization

Several signals will indicate how this plays out over the coming months.

First, watch for SEC comment letters. Confidential S-1 filings trigger a standard SEC review process that includes written questions to the company. Comment letters on revenue-recognition methodology are common for cloud and SaaS businesses, and Anthropic should expect pointed questions on its principal-vs.-agent determination.

Second, look at how both companies publicly describe their revenue in pre-IPO interviews and analyst briefings. If either company begins voluntarily disclosing both gross and net figures — as some dual-reporting companies have done — it may signal internal recognition that the current approach won't survive regulatory scrutiny unchanged.

Third, track whether the SEC issues any new guidance on AI platform revenue specifically. The gross-vs.-net question has arisen before in cloud-marketplaces and app-store contexts; the agency could use the Anthropic and OpenAI filings as an opportunity to clarify the framework for the AI era.

Until then, investors parsing the Anthropic and OpenAI revenue figures should do so with a clear understanding: the same dollar of Claude usage may appear very differently depending on which company's filing you are reading.