Investors react to news about OpenAI’s and Anthropic’s revenue almost as if it were the financial reporting of public companies. But the financial metrics of these AI startups are difficult to compare directly: it is unclear what expenses they bear, how much cash they are burning, and how close they are to profitability. Even the revenue figures that regularly appear in the media may be calculated using different methodologies.

This affects not only the valuation of the companies themselves, but also sentiment in the stock market. Tens and hundreds of billions of dollars are being invested in AI development, and investors are trying to understand whether future income will justify the current scale of spending, Axios reports.

Why revenue news sends stocks tumbling

Recently, Financial Times reported that OpenAI’s annual revenue figure could be about $20 billion lower than some market participants had assumed. After the publication, tech stocks fell sharply.

However, as Axios notes, the discrepancy apparently was not caused by a sudden deterioration in OpenAI’s business or by concealment of its financial results. The reason lay in an attempt to calculate the company’s revenue using the same methodology employed by Anthropic.

This episode shows how cautiously headline-grabbing figures should be treated. If companies calculate metrics differently, comparing their financial results can create a misleading impression of who is growing faster and how successfully each is turning the popularity of its models into money.

Venture investor and blogger Paul Kedrosky compared what is happening to absurdist theater. In his view, the current revenue figures of the largest AI startups depend largely on arbitrarily chosen calculation methods.

What’s wrong with annualized revenue and ARR

Publications about AI companies often cite two closely related metrics: annualized revenue calculated on the basis of the current pace of receipts, and ARR — annual recurring revenue from subscriptions and other repeat payments.

Both metrics help assess the scale of a business, but they are not equivalent to the revenue actually received over a year. An annualized revenue run rate, for example, can be obtained by multiplying one month’s income by 12. Such a calculation assumes that receipts will remain at the same level, although in practice they can change noticeably.

Imagine a person who in December received a regular salary and a large annual bonus. If you multiply December income by 12, you get a sum that significantly exceeds the person’s real annual earnings. Something similar happens in business when the chosen month is atypical.

ARR is useful for software companies with recurring subscriptions and relatively predictable demand. But developers of advanced AI models operate in a rapidly changing environment: demand for products, usage volume, and receipts can fluctuate substantially even within a few weeks.

Kedrosky believes that with such volatility, the result depends too heavily on exactly which period the company uses for its data. And even if the metric is calculated correctly, it says nothing about spending on computing infrastructure, model development, personnel, and other business needs.

High revenue by itself does not mean a company is profitable or has a sufficient cash cushion.

Why public companies are in a different position

If OpenAI and Anthropic were already publicly traded, investors would receive regular reporting on recognized revenue, expenses, and financial results. Public companies must comply with standardized accounting rules, and their annual reports undergo independent audits.

This does not make financial information absolutely transparent: even large public companies face questions about the disclosure of liabilities and other aspects of reporting. But uniform rules make it possible to compare the metrics of different companies with a much lower risk of error due to methodological differences.

For now, however, a significant portion of the information about the finances of the largest AI startups comes from company statements and media reports. The details of how metrics are calculated may remain unclear, while the full picture of expenses and profitability remains inaccessible to outside observers.

Investors will probably get more information from the documents companies are required to disclose when going public.

The stories of Groupon and WeWork are reminders of the risks

Inflated expectations regarding startups’ financial results are nothing new. When a company is preparing to go public, it must disclose more information and bring its metrics into line with established standards. Sometimes this significantly changes investors’ understanding of its business.

In 2011, Groupon, then considered one of the most promising technology companies, was forced after preparing its IPO documents to cut its previously reported revenue in half in order to bring its reporting into compliance with generally accepted accounting principles in the United States.

Another example is WeWork. Before its planned 2019 IPO, the disclosure of financial documents revealed massive losses and significant long-term lease obligations. The public offering ultimately collapsed.

These stories do not mean that OpenAI or Anthropic will necessarily face similar problems. But they are a reminder: valuations of private companies built on limited financial information can differ substantially from the picture that emerges after full disclosure of financial reporting.

The next important milestone is Anthropic’s IPO

According to information cited by Axios, Anthropic may go public in mid-November. If those plans materialize, investors will have an opportunity to examine the company’s more detailed financial documents and check how closely its public metrics match market expectations.

Until then, the revenue figures of OpenAI and Anthropic should be seen as standalone reference points rather than as a full report on the state of the business. To assess the resilience of these companies, it is necessary to understand not only the volume of receipts, but also the methodology used to calculate them, the cost structure, capital requirements, and the ability to turn growing demand for AI into profit.

In a period of enormous investment in artificial intelligence, the difference between fast-growing revenue and an economically sustainable business becomes especially important. And as long as the financial data of the largest private developers remain incomplete, the market risks making decisions based on figures that look more precise than they really are.