OpenAI vs Anthropic—How are the financial reports of the "strongest AI"?

```

Two of Silicon Valley’s hottest AI unicorns are racing to go public within the year, but a rare disclosure of financial data reveals the same dilemma: the astronomical computing power costs of training AI models are devouring both companies’ profit margins.

According to financial documents recently obtained by The Wall Street Journal, OpenAI projects its computing expenses will reach $121 billion in 2028. Even if revenue nearly doubles by then, the company would still post a loss as high as $85 billion that year—a figure that would surpass the losses of almost any publicly listed company in history.

At the same time, Anthropic’s projected expenses are far less than OpenAI’s, but even its most optimistic forecasts reflect the steadily rising cost of computing power. In addition, Bloomberg reported Tuesday that Anthropic’s latest annualized run rate has exceeded $30 billion, jumping significantly from $9 billion at the end of 2025.

Financial data from both companies paints a clear picture of the AI arms race: while revenue is soaring, the pace at which training costs are expanding is equally staggering, and the path to profitability remains long. For investors eyeing their IPOs, this financial outlook both demonstrates tremendous growth potential and highlights the risks involved.

Revenue Soaring, But Losses Just as Shocking

Both OpenAI and Anthropic anticipate their revenues will more than double this year, chiefly driven by accelerated adoption of AI tools by enterprise clients.

OpenAI’s revenue sources include consumer subscriptions, enterprise services, and new products (including hardware); Anthropic relies almost entirely on enterprise customers and counts sales through cloud partners as part of its revenue.

However, behind the impressive revenue growth are equally eye-catching losses. OpenAI projects that even with substantial revenue growth by 2028, it will still lose $85 billion in that year. The company is not expected to reach break-even until 2030, while Anthropic forecasts it could achieve this milestone earlier.

Notably, both companies report profits using two different metrics: excluding “research computing” expenses, OpenAI may achieve a small pre-tax operating profit this year, and the same is true for Anthropic in their most optimistic scenario; but once training costs are included, both remain deeply unprofitable.

AI Computing Arms Race: Costs are the Big Variable

Uncontrolled training costs are the core pressure point in both companies’ financial structure. Each new model generation requires far more compute resources than the last, and both companies are launching new models at unprecedented frequency.

OpenAI projects that in 2028, its spending on AI research compute will reach $121 billion. In comparison, Anthropic’s training expenditure is much lower, but its forecasts also show a steady upward trend in computing costs.

Inference costs (the cost of processing user queries) also represent a major burden. Currently, inference costs account for over 50% of revenue at both companies, although this proportion is expected to gradually decrease with technological efficiency improvements. Paid ChatGPT users account for only a tiny fraction, meaning a large portion of inference costs cannot be offset by revenue.

On a cash flow basis, both companies are expected to continue burning large amounts of cash for years to come, and IPO fundraising is viewed as a key source of funding to sustain operations.

Anthropic’s Annualized Revenue Tops $30 Billion, Secures New Compute Partners

According to Bloomberg, Anthropic’s annualized revenue has surpassed $30 billion, more than tripling from $9 billion at the end of 2025. Over 1,000 enterprise customers now spend more than $1 million each year on Anthropic’s platform, doubling since February this year.

To support this growth, Anthropic has signed major computing power agreements with Broadcom and Google. According to documents filed by Broadcom on Monday, the three parties will expand their strategic partnership, enabling Anthropic to access about 3.5 gigawatts of computing resources from 2027. Broadcom is developing chips based on Google’s Tensor Processing Units (TPU) as an alternative to Nvidia, and the sides have signed a long-term supply assurance agreement running through 2031.

Anthropic CFO Krishna Rao stated that collaboration with Broadcom and Google will help build the "computing foundation needed for significant customer growth." Following the news, Broadcom shares rose as much as 3.6% after hours.

In addition, to meet the potentially record-breaking IPO fundraising demand of the two companies, Wall Street is seeking to break through existing regulatory constraints. Bankers are lobbying major index providers to relax inclusion standards, and Nasdaq recently announced it will allow newly listed companies to join its indexes faster, thus tapping into a broader capital pool. OpenAI, for its part, said the company currently prioritizes growth over profit; while it could trim training expenses, it expects such investments to return substantial rewards.

Risk Warning and DisclaimerThe market involves risks and investments should be made with caution. This article does not constitute personal investment advice and does not take into account the special investment objectives, financial circumstances, or needs of any individual user. Users should consider whether any opinions, viewpoints, or conclusions in this article are appropriate for their particular situation. Investing accordingly is at your own risk. ```