OpenAI continues to grow rapidly. In the second quarter, the company generated $6.7 billion in revenue, up 18 percent from the beginning of the year. At the same time, its operating loss widened from $9.3 billion to $12.3 billion. Rival Anthropic is moving in the opposite direction: revenue surged to more than $11.5 billion, while the company reported a positive adjusted operating result for the first time. The figures mark more than another round in the competition between ChatGPT and Claude. They suggest that a different question is becoming increasingly important for the leading AI companies: How much economic value can actually be extracted from the enormous amounts of computing power they consume?
OpenAI Earns More – and Spends Much More
At first glance, OpenAI’s latest figures remain impressive.
Revenue rose from $5.7 billion in the first quarter to $6.7 billion in the second, an increase of around 18 percent. The longer-term growth is even more striking: according to company figures, OpenAI’s current annualized revenue run rate has risen above $40 billion, roughly twice the level recorded at the end of 2025.
The business is being driven by ChatGPT subscriptions, enterprise applications, Codex and the first advertising revenues.
But costs are rising even faster.
OpenAI’s operating loss widened to $12.3 billion in the second quarter, compared with $9.3 billion in the first. According to available information, this figure also includes stock-based compensation.
The company therefore continues to spend considerably more money than it earns from its products.
That is not fundamentally surprising in the AI industry. Training new models costs billions, while every model request consumes computing resources. Unlike conventional software, serving an additional user does not necessarily come at negligible marginal cost.
But the scale of the losses illustrates how demanding OpenAI’s business model has become.
Anthropic Turns the Equation Positive
At rival Anthropic, the picture currently looks different.
The developer behind Claude generated preliminary second-quarter revenue of more than $11.5 billion. That compares with $4.73 billion in the first quarter and just $787 million in the second quarter of 2025.
Quarterly revenue has therefore increased more than tenfold within a year.
Even more remarkable is the figure further down the income statement.
Anthropic reported a positive adjusted operating profit for the first time in the second quarter. Earlier internal projections had pointed to roughly $559 million. The final figure may still be revised.
For a company that itself requires enormous amounts of computing capacity, this represents an important milestone: at least for one quarter, Anthropic has turned its operating business profitable.
The word “profitable”, however, requires an asterisk.
Profit Is Not Always Profit
Anthropic has not reported a confirmed net profit after all expenses.
The positive figure refers to adjusted operating profit. According to reporting on the company’s accounts, it includes costs associated with training new models but excludes stock-based compensation.
For rapidly growing technology companies, such compensation can be substantial.
It would therefore be too simplistic to conclude that Anthropic is making money while OpenAI is merely burning billions.
A direct comparison of revenue is equally difficult.
The two companies account for parts of their businesses differently. Anthropic reportedly records revenue generated through cloud partners on a gross basis and subsequently treats the partners’ share as a cost. OpenAI reports parts of its business differently, including the treatment of Microsoft’s revenue share.
As a result, the headline comparison of $11.5 billion versus $6.7 billion does not automatically establish a clean winner in the revenue race.
Anthropic’s extraordinary growth remains real. But the figures are not perfectly comparable.
Anthropic Makes Its Money Differently
The more important difference may lie in the companies’ business models.
With ChatGPT, OpenAI has created an enormous consumer market. A significant share of those users, however, pay nothing. Their activity still generates computing costs.
Anthropic has focused more heavily on companies, developers and professional applications.
Claude has been particularly successful in software development. Claude Code has quickly emerged as one of the company’s most important growth drivers, while a large share of revenue comes from businesses and software developers accessing Anthropic’s models through APIs and cloud platforms.
That changes the economics.
A company using AI to develop software faster and save employee hours has significantly greater willingness to pay than a consumer occasionally using a chatbot for free.
Anthropic’s current strength may therefore be less about simply offering “better AI” and more about where its AI is already being used.
A larger share of its customers operate in environments where the economic value of the technology can be measured directly — and where users are therefore willing to pay substantial prices.
$65 Billion Is Not Annual Revenue
Another headline figure also requires careful interpretation.
Anthropic’s revenue run rate reportedly exceeded $65 billion by the end of July. At the end of 2025, it had stood at only around $9 billion. OpenAI is currently above $40 billion.
But a revenue run rate is not the same as annual revenue already earned.
It extrapolates the company’s current pace of sales over a twelve-month period. If a business were to generate $5 billion during one particularly strong month, for example, that would imply an annualized run rate of $60 billion.
It does not mean the company has already earned $60 billion, nor does it guarantee that the current pace will continue for an entire year.
The distinction matters particularly for Anthropic because the company is currently experiencing extraordinary growth.
Its run rate is primarily a measure of momentum, not an audited annual result.
Once again, there is another complication: Anthropic and OpenAI may not calculate their annualized revenue on precisely the same basis.
Even these figures therefore allow only a limited direct comparison.
The AI Race Gets a New Discipline
For years, the competition between leading AI companies was primarily measured in technical terms.
Who has the most capable model? Who reaches the largest number of users? Who has access to the greatest amount of computing power?
The latest financial figures introduce another category: efficiency.
In the long run, generating billions in revenue is not enough if producing those revenues requires spending even more billions.
That becomes increasingly important as competition intensifies.
Alongside OpenAI and Anthropic, Google, Meta and numerous Chinese companies are investing heavily in advanced models. At the same time, prices for many AI services are falling. What can be sold today as a technological advantage may become a largely interchangeable product only months later.
Cost structures are therefore becoming strategically important.
Companies that can train models more cheaply, use computing resources more efficiently or sell their services to customers with a high willingness to pay may ultimately build stronger businesses even with fewer users.
Anthropic’s Advantage May Not Last
Anthropic’s first positive adjusted operating result does not mean that the company is now sustainably profitable.
It is expanding its computing infrastructure aggressively. Long-term agreements for additional capacity involve enormous financial commitments, and growing demand can drive infrastructure spending higher again.
The company itself had previously expected sustainable profitability to arrive much later.
The second quarter therefore demonstrates something narrower but still significant: a frontier AI company can apparently generate enough operating revenue to cover at least its adjusted operating costs.
Whether that develops into a consistently profitable business model remains unproven.
Ahead of an IPO, Visions Become Financial Metrics
The development comes at a crucial moment.
Anthropic is preparing for a possible initial public offering and could reach the stock market before OpenAI.
At that point, exponential growth alone will no longer be enough to tell the story.
Public-market investors will have to determine how much economic value that growth actually creates.
For both companies, the available financial data remain unusually difficult to verify. OpenAI and Anthropic are privately held and are not yet subject to the regular disclosure obligations imposed on listed companies.
Many of the figures currently known come from investor documents or people with access to internal financial information.
An IPO would change that.
Investors would finally receive regular financial statements with more standardized definitions, cost categories and risk disclosures.
Only then may it become possible to determine with much greater confidence which of the two companies has built the economically stronger AI business.
Until then, the latest numbers already reveal one important shift:
Winning the artificial intelligence race is no longer merely about having the most powerful models or the largest number of users.
The next phase may also be decided by who can turn every dollar of computing power into more business.
SK
