OpenAI and Anthropic: the earnings a trillion-dollar IPO must deliver

The valuation question for an OpenAI or Anthropic IPO is how much future profit and cash investors must receive to justify the price paid today. OpenAI has greater commercial scale; Anthropic offers rapid enterprise-channel expansion. Both must pay for compute before shareholders receive the hoped-for returns. A higher entry price makes the cost of financing that wait more consequential.

This comparison uses public reporting available on October 3, 2026. Reuters said it reviewed confidential Anthropic IPO materials; its OpenAI projections cite company materials seen by the Financial Times. Complete public prospectuses were unavailable. Historical figures remain attributed to reporting, and projections remain forecasts.^[1][2]

Revenue needs a common period and definition

A comparison of scale needs revenue measured over the same period. Completed-year sales, a forecast and the annualized pace of recent business answer different questions. In a fast-growing business, valuing the company on its exit run rate effectively gives it credit for sales that have not yet persisted for a full year.

On September 29, Reuters cited a person familiar with the matter as saying that OpenAI's annualized recurring revenue was approaching $70 billion. An earlier September 18 report, drawing on company materials, put the 2026 revenue forecast at $36 billion. Both can be true: if business accelerates substantially in the second half, the year-end annualized pace will exceed revenue actually earned across the full year. The $70 billion cannot therefore be described as realized 2026 revenue, nor as cash already collected.^[2][3]

Reporting on Anthropic provides a different perspective: revenue of nearly $4.6 billion in 2025 is a historical annual figure. Dividing OpenAI's $70 billion annualized measure by Anthropic's $4.6 billion historical revenue would combine differences in business scale, timing, and measurement. That ratio cannot establish the true competitive distance between the companies.^[1]

The table draws on media disclosures from different dates. All amounts are in US dollars; funding negotiations and listing targets involve different transaction terms.^[1][2][3]

Investors should ask both companies to present the same reconciliation during their roadshows: the latest complete year, the trailing twelve months, quarterly revenue, and the period-end annualized measure, followed by an explanation of the subscription and usage-based components. For usage-based businesses in particular, a large customer's training project or a one-off workload can change the monthly pace. Annualizing that figure still leaves the question of how long the demand will persist.

Which losses consume cash?

Anthropic’s reported 2025 net loss of roughly $42 billion included about $34 billion of financing-instrument fair-value charges, while its operating loss exceeded $8 billion.^[1] Read together, these figures show both a large accounting effect and substantial operating losses. None is a substitute for the cash-flow statement.

IPO analysis should separate the income statement into three layers. Revenue and service costs determine the gross profit generated by customer business; research and development, sales, and administrative expenses determine operating profit; financing valuations and other items affect net income. The fact that financing-related accounting charges are not operating expenditure does not justify ignoring operating losses. Equally, a large net loss does not prove that cash has been exhausted.

OpenAI's pressure is more directly visible in its forward cash budget. The materials relayed by Reuters project cumulative negative free cash flow of $278 billion from 2026 through 2030, with revenue reaching $350 billion in 2030. This is a company forecast, not losses already incurred over the next five years. It implies that management is willing to commit substantial capital to achieve much greater revenue scale, and that investors must also assess the terms on which subsequent funding could be raised.^[2]

A simple calculation illustrates the dilution risk. Suppose a company currently has an equity value of $1 trillion and later needs to issue $200 billion of new shares at the same pre-money valuation. Existing shareholders would retain 1 ÷ 1.2, or 83.3%, a dilution of 16.7%. If the pre-money valuation instead fell to $500 billion and the company raised $200 billion, existing shareholders would retain 5 ÷ 7, or 71.4%, a dilution of 28.6%. This is an equity-financing scenario, not an established funding arrangement for either company. It explains why IPO investors need to examine cash budgets beyond the listing itself.

Debt, customer prepayments, supplier arrangements, or lower investment could reduce the need for equity financing. None is free: debt adds fixed payment obligations, prepayments may require discounts, supplier terms may sacrifice purchasing flexibility, and compute cuts may constrain growth. Any claim that funding is ample should also explain who bears the cost.

Cloud distribution earns its share of the economics

Amazon and Google cloud marketplaces help Anthropic reach existing enterprise procurement budgets, but that access has a cost. Reported 2025 channel revenue was about $2.16 billion, or 47% of sales, with $351 million in distribution fees. The company reportedly presents revenue gross and records those fees in operating expenses.^[4] The sales multiple therefore needs to be read alongside the economics retained after distribution.

Using those reported figures, $351 million ÷ $2.16 billion is approximately 16.25%. That is the distribution-fee ratio for the relevant channel revenue. It is neither Anthropic's company-wide selling-expense ratio nor its gross margin after inference costs. The calculation highlights that a dollar of revenue earned directly may leave a different amount of profit from a dollar earned through a cloud channel.

Gross or net revenue recognition depends on accounting judgments such as who controls the service and bears responsibility for performance. This article does not infer from media debate that the treatment is incorrect. Valuation comparisons nevertheless require additional adjustment: if one business reports some channel revenue gross and another reports it net, a direct price-to-sales comparison may mistake a presentation difference for a lower valuation.

A useful comparison starts with what each dollar of business leaves for research and development and shareholder returns after distribution fees, inference-service costs, and customer concessions. That requires more detailed financial disclosure. We do not simply subtract $351 million from total revenue and call the result “true revenue,” because one adjustment cannot reconcile the two companies' complete accounting policies.

Cloud partners' overlapping roles as investors, suppliers, and distributors also complicate negotiations. A channel can bring customers while its platform promotes competing in-house models. Long-term procurement can secure capacity while reducing flexibility to cut spending. Over the next few quarters, growth in direct customers, distribution-fee rates, and unit service costs will be more informative for investors than a broad claim about ecosystem partnerships.

Work backwards from the offer price

For a company without stable earnings, work backwards from the price: first estimate the revenue required, then test how much can become profit. The $1.2 trillion financing valuation and $2 trillion lower boundary of the reported listing target are equity-value scenarios. Incomplete net-debt disclosure prevents treating the resulting ratios as enterprise-value multiples.

Calculation: assumed equity value ÷ revenue for the same year. All revenue figures in the table are sensitivity assumptions, not forecasts for either company. Future dilution and changes in net debt are excluded.

The same $2 trillion valuation represents 20 times revenue at $100 billion of sales and 10 times at $200 billion. A falling multiple does not automatically make the stock cheaper: shareholders must wait for the revenue to materialize while bearing competition, cost, and additional-financing risk. Time must enter the valuation. Profits achievable only five years from now are worth less today than profits already being generated.

Consider another purely illustrative case. Assume fifth-year revenue of $200 billion and a net margin attributable to shareholders of 20%, producing $40 billion of earnings. At a market-assigned price-to-earnings multiple of 30, terminal equity value would be $1.2 trillion. Discounted at a required annual return of 10%, its value today would be approximately $745.1 billion: $1.2 trillion ÷ 1.1 to the fifth power. The example does not predict that either company will reach those figures and excludes interim cash flows and share issuance. It shows that even if a company becomes exceptionally successful, the investment return still depends on the entry price.

Leadership in model capability and leadership in shareholder returns therefore require separate arguments. A high valuation can be supported by faster growth, higher margins, or a longer period of competitive advantage. It cannot be called conservative if all three are simultaneously set to optimistic assumptions. IPO pricing should make clear which assumption matters most and how much room remains if one of them fails.

Growth and concentration can rise together

The debate over Anthropic addresses several different investments. Reuters quoted Morningstar’s Michael Field on growth and an 18–20 times revenue range, AJ Bell’s Dan Coatsworth on customer concentration, and XTB’s Kathleen Brooks on the support that investment provides to semiconductor and infrastructure demand.^[5] These observations are neither a common valuation model nor a market consensus.

These views are not mutually exclusive. A company can be fast-growing, customer-concentrated, and capital-intensive at the same time. Field's multiple requires clarity about the revenue year being used. Dividing $2 trillion by 18–20 implies a revenue denominator of approximately $100 billion to $111.1 billion. This reverse calculation is our arithmetic interpretation, not a revenue forecast supplied by Field, and the multiple cannot simply be applied to the $4.6 billion reported for 2025.

Customer concentration affects the stability of growth. Expansion by a few large customers can lift revenue quickly, but can also amplify volatility when contracts are renegotiated, internal budgets change, or customers switch models. Investors need to distinguish direct customers, collections through cloud platforms, and ultimate enterprise users. A platform's channel share cannot simply be treated as demand from a single end user.

Brooks' infrastructure perspective reveals another distinction: model developers' spending may first become revenue for chip, server, and networking suppliers. Supplier gains do not ensure that model-company shareholders profit over the same period. If developers continue to raise capital, the supply chain may realize earnings earlier while returns at the end platform remain further away. The next article on funding, followed by the NVIDIA, AMD, and optical communications studies, will examine financing and profit distribution along this transmission chain.

The roadshow must connect growth to funding needs

The first roadshow test is whether customers keep buying. A useful revenue bridge would separate new paying customers, existing-customer usage, price changes and large projects. It would show how much business can recur without additional discounts. A larger annualized number alone does not establish the profit available at renewal.

The second is improving profitability. When unit inference costs decline, does the company retain the savings or pass them to customers through lower prices? In a competitive market, technical efficiency does not necessarily produce an equivalent improvement in gross margin. Investors should examine unit revenue alongside unit cost, rather than the cost curve alone.

The third is capital discipline. How much demand covered by long-term compute contracts is already supported by visible revenue? Are procurement payments mismatched with customer collections? If new financing slows, can the company adjust spending without materially damaging service and growth? The answers determine whether shareholders are bearing expansion risk or the risk of persistent dependence on external capital.

Before complete disclosure, we would not reduce this IPO competition to a single price target. A more useful assessment is that OpenAI must demonstrate that greater revenue scale can shorten the wait for cash returns, while Anthropic must demonstrate that faster growth can cover distribution and compute costs. Sustained improvements in gross margins and cash flow, coupled with less reliance on new financing, would provide a firmer foundation for higher valuations. If revenue accelerates while cash consumption and shareholder dilution rise alongside it, listing enthusiasm alone cannot justify the purchase price.