OpenAI and Anthropic, leading global AI startups, are preparing what could be record-setting initial public offerings this year, but concerns persist that their internal structures are fragile. Soaring costs needed to train new AI models are also cited as an Achilles heel.

The Wall Street Journal, citing confidential financial documents the companies shared with investors before finishing funding rounds early this year, reported OpenAI expects to spend $121 billion on computing power needed for AI research in 2028. It could burn $85 billion in cash flow in 2028 even if revenue nearly doubles in 2027. It is a structure in which money going out is still far greater than money coming in.

The WSJ said that would overwhelm losses at almost all other listed companies in history. It added that both companies are expected to consume vast amounts of cash over the next several years and are hoping money raised through IPOs will sustain their businesses.

Anthropic may not spend as much money as OpenAI, but it is in a similar situation in the narrative of rising computing costs. The two companies are releasing new versions of AI models at a faster pace than ever, and there is no sign this race will slow.

The WSJ said, "As it becomes increasingly difficult to raise AI performance by one level, costs are also higher than for previous models." It added, "OpenAI plans to spend far more money than Anthropic on model training over the next few years."

With costs so high, OpenAI and Anthropic are presenting profitability metrics in two ways: including and excluding training costs. Excluding research computing costs, OpenAI is on track to achieve a small pretax operating profit this year, and Anthropic could as well in an optimistic scenario, the WSJ said. Including research computing costs, OpenAI is expected to reach breakeven only after 2030, while Anthropic is projected to reach breakeven sooner than that.

Venture capital firms have tolerated huge losses because OpenAI and Anthropic are among the fastest-growing companies in the history of the tech market. As corporate investment in AI increases, both companies expect revenue to more than double this year.

The Information reported in early March that OpenAI's annualized revenue had surpassed $25 billion as of February. That was up about 17 percent from $21.4 billion at the end of last year.

OpenAI is ahead of rival Anthropic in revenue, but the gap is narrowing. Anthropic's annualized revenue also topped $19 billion. Strong sales of coding AI models drove revenue growth.

The Information said OpenAI's 2025 revenue was about three times Anthropic's, but the gap is narrowing quickly.

But it is difficult to compare the two companies' revenue on the same basis at the current point, the WSJ said. It reported that Anthropic books as revenue technology sold through cloud partners, while OpenAI does not.

OpenAI and Anthropic are also investing tens of billions of dollars a year in inference, the stage of using trained models. At the current point, that is at a level that consumes more than half of revenue.

Unlike Anthropic, which has a large share of enterprise customers, only a tiny portion of ChatGPT users pay, so OpenAI has yet to recoup a significant part of inference costs through revenue. But as the cost of running AI becomes cheaper, that share is expected to gradually decline.