OpenAI vs Anthropic--How is the financial report of the "most powerful AI" really looking?

Two of Silicon Valley’s hottest AI unicorns are rushing to go public within the year, but rare disclosed financial data reveals the same tough problem: the astronomical compute costs of training AI models are eating into both companies’ profit margins.

According to financial filings recently obtained by The Wall Street Journal, OpenAI expects its AI compute spending in 2028 to reach $121 billion. Even if revenues nearly double by then, the loss for that year will still be as high as $85 billion—a figure that would surpass the record for losses by nearly all publicly listed companies in history.

Meanwhile, Anthropic expects spending to be far less than OpenAI’s, but its most optimistic projections also reflect a scenario in which compute costs continue to climb. In addition, according to a Bloomberg report Tuesday, Anthropic’s latest annualized revenue (Run Rate) has already surpassed $30 billion, up sharply from $9 billion at the end of 2025.

Taken together, the companies’ financial data outlines the real logic of the AI arms race: revenue is growing at high speed, but the rate at which training costs are ballooning is equally astonishing, and the path to profitability remains long. For investors considering participating in their IPOs, this financial picture not only shows tremendous growth potential, but also clearly flags where the risks lie.

Revenue skyrockets, but losses are just as alarming

Both OpenAI and Anthropic expect this year’s revenue to more than double, driven mainly by corporate customers’ accelerated adoption of AI tools.

OpenAI’s revenue sources span consumer subscriptions, enterprise services, and new products (including hardware); Anthropic relies almost entirely on enterprise customers, and it will recognize sales achieved through cloud partners as revenue.

However, behind the eye-catching revenue growth rates is an equally alarming scale of losses. OpenAI expects that even if revenues grow significantly by 2028, it will still lose $85 billion that year. The company expects to reach overall break-even only by 2030, while Anthropic expects to hit that milestone earlier.

It’s also worth noting that both companies use a two-pronged approach to disclosing profits: after stripping out “research compute” spending, OpenAI is expected to achieve a small pre-tax operating profit this year, and Anthropic is the same under its most optimistic scenario; but once training costs are included, both are mired in losses.

Compute arms race: cost is the biggest variable

Runaway training costs are the most core source of pressure in both companies’ financial structures. With each generation of models, improvements in intelligence require compute investment far exceeding that of the prior generation, and both companies are currently iterating and releasing new models at unprecedented frequencies.

OpenAI expects AI research compute spending in 2028 to be as high as $121 billion. By comparison, Anthropic’s training spending is smaller, but its financial forecasts also show a trend in which compute costs keep rising.

Inference costs (that is, the expenses generated by processing user queries) also represent a major burden. Currently, inference costs account for more than 50% of both companies’ revenue; although this proportion is expected to decline gradually as technical efficiency improves. ChatGPT’s paid users make up only a tiny share, meaning a large portion of inference costs cannot be covered by revenue.

On the cash-flow side, over the next few years both companies will continue to consume cash at large scale, and IPO proceeds are seen as a key source of funding to keep the business running.

Anthropic annualized revenue tops $30 billion, locking in new compute allies

According to Bloomberg, Anthropic’s annualized revenue has already surpassed $30 billion, up more than two times from $9 billion at the end of 2025. Currently, the number of enterprise customers spending more than $1 million per year on the Anthropic platform has exceeded 1,000, and this figure has doubled since February of this year.

To support this growth momentum, Anthropic has signed major compute collaboration agreements with Broadcom and Google. According to filings Broadcom submitted on Monday, the three parties will expand their strategic cooperation, allowing Anthropic to receive about 3.5 gigawatts of compute resources starting in 2027. Broadcom is developing chips based on Google’s Tensor Processing Units (TPUs), providing an alternative to NVIDIA, and the two have already signed a long-term supply assurance agreement extending through 2031.

Anthropic CFO Krishna Rao said the collaboration with Broadcom and Google will help the company build the “compute foundation that is necessary for a significantly growing customer base.” After the news was released, Broadcom’s stock price briefly rose 3.6% after the market close.

In addition, to support what could be record-level IPO financing needs for the two companies, Wall Street is looking to push past existing rule constraints. Bankers are lobbying major index providers to relax inclusion standards. Nasdaq recently announced it will allow newly listed companies to join its indices more quickly, giving them support from a broader pool of capital. For OpenAI, the company said it is currently prioritizing growth over profit; while it could cut training spending, it expects the related investments to bring substantial returns.

Risk warning and disclaimer

        There are risks in the market; invest cautiously. This article does not constitute personal investment advice, and it does not take into account any specific investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article align with their specific circumstances. Invest at your own risk.