OpenAI vs Anthropic--What is the financial report of the "most powerful AI" really like?

Investing in stocks relies on Jin Qilin Analyst Reports, authoritative, professional, timely, comprehensive, helping you discover potential thematic opportunities!

(Source: NetEase Tech)

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 challenge: the astronomical computing power costs for training AI models are eating into the profitability of both companies.

According to recent financial documents obtained by The Wall Street Journal, OpenAI expects its computing power expenses to reach $121 billion by 2028, even if revenue nearly doubles by then, the annual loss will still be as high as $85 billion—this figure will surpass almost all historical IPO loss records.

Meanwhile, Anthropic expects its expenses to be much lower than OpenAI’s, but its most optimistic forecast also reflects the rising costs of computing power. Additionally, according to Bloomberg on Tuesday, Anthropic’s latest annualized revenue (Run Rate) has surpassed $30 billion, a significant jump from $9 billion at the end of 2025.

The financial data of these two companies jointly outline the reality of the AI arms race: rapid revenue growth, but the exploding training costs are equally staggering, and profitability remains a long way off. For investors interested in participating in their IPOs, this financial outlook demonstrates tremendous growth potential while clearly highlighting the risks involved.

Revenue surges, but losses are equally alarming

Both OpenAI and Anthropic expect their revenues to more than double this year, driven mainly 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 clients and will include sales through cloud partners in its revenue.

However, behind the impressive revenue growth is equally shocking loss scales. OpenAI projects that even with substantial revenue increases by 2028, it will still lose $85 billion that year. The company expects to break even only around 2030, while Anthropic anticipates reaching this milestone sooner.

It’s worth noting that both companies use a dual-profit disclosure approach: after excluding “research computing” expenses, OpenAI may achieve slight pre-tax operating profit this year, and Anthropic under the most optimistic scenario; but once training costs are included, both are deep in the red.

The AI arms race: costs are the biggest variable

The out-of-control training costs are the core pressure in both companies’ financial structures. Each generation of models with higher intelligence requires far more computing power than the previous one, and both companies are iterating and releasing new models at unprecedented speeds.

OpenAI estimates that its AI research computing expenses will reach $121 billion by 2028. In comparison, Anthropic’s training expenses are smaller, but its financial forecasts also show a continued rise in computing power costs.

Inference costs (the expenses incurred when processing user queries) are also a major burden. Currently, inference costs account for over 50% of both companies’ revenues, although this ratio is expected to gradually decline with technological efficiency improvements. Only a tiny proportion of ChatGPT’s paid users, meaning a large portion of inference costs cannot be covered by revenue.

From a cash flow perspective, both companies will continue to burn cash on a large scale over the next few years, with IPO fundraising seen as a key source of capital to sustain operations.

Anthropic’s annualized revenue exceeds $30 billion, locking in new computing power allies

According to Bloomberg, Anthropic’s annualized revenue has surpassed $30 billion, more than doubling from $9 billion at the end of 2025. Currently, over 1,000 enterprise clients spend more than $1 million annually on the Anthropic platform, a number that has doubled since February this year.

To support this growth, Anthropic has signed major computing power cooperation 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 approximately 3.5 gigawatts of computing resources starting in 2027. Broadcom is developing chips based on Google’s tensor processing units (TPUs) as an alternative to Nvidia, and the two have signed a long-term supply agreement extending to 2031.

Anthropic’s CFO Krishna Rao stated that cooperation with Broadcom and Google will help the company build the “computing power infrastructure necessary for significant customer base growth.” Following the announcement, Broadcom’s stock rose by 3.6% after hours.

Additionally, to meet the potentially record-breaking IPO funding needs, Wall Street is seeking to break existing rules. Bankers are lobbying major index providers to relax inclusion standards, and Nasdaq recently announced it will allow new listings to join its indices more quickly, broadening access to capital. OpenAI has stated that the company currently prioritizes growth over profit, and while it can cut training expenses, it expects the related investments to yield substantial returns.