The documents OpenAI shared with its shareholders in recent weeks tell a story of growth and cash consumption moving in lockstep. In the first three months of the year, the company burned through $3.7 billion in cash while booking $5.7 billion in revenue. Both figures roughly tripled from a year earlier.
The numbers, first reported by The Information, lay out the central arithmetic of the AI boom: demand is exploding, and so is the cost of serving it. OpenAI’s cash burn in the quarter exceeded half of its revenue, a ratio that has barely budged even as the company’s top line has tripled. For every dollar the company earned in the quarter, it consumed more than a dollar of cash, according to people familiar with the shareholder materials.
The company ended March with more than $73 billion in cash and marketable securities, up from about $40 billion at the end of December. The jump came almost entirely from a record funding round that closed in late March, which valued OpenAI at roughly $850 billion, rather than from operations. The round was one of the largest private financings ever assembled, with Amazon, Nvidia, and SoftBank committing about $110 billion among them, according to people familiar with the terms. Amazon’s $50 billion pledge included a $35 billion tranche contingent on an initial public offering or a defined artificial general intelligence achievement, a structure that ties part of the financing to events OpenAI does not fully control.
The balance sheet cushion changes the timing of OpenAI’s next moves. If cash consumption stays at the first-quarter pace, the company can fund itself for years without raising fresh capital, and it has little immediate need to rush its planned public offering. Executives have told investors the company expects to burn about $25 billion this year and $57 billion next year, figures that imply the current pace will accelerate rather than ease. Cash burn can also swing quarter to quarter: a single large compute contract, a training run, or a data center prepayment can move the quarterly number by billions, and the company has cautioned shareholders against extrapolating any single period.
The broader operating picture is more complicated than the cash number alone. OpenAI’s quarterly operating loss was about $9.3 billion, according to the shareholder documents, while its consolidated net loss reached $21.3 billion. More than $12 billion of that net loss came from non-cash accounting adjustments tied to the company’s corporate restructuring, which revalued investor rights and produced paper charges that will not recur in the same form. Analysts who have reviewed the documents say the operating loss is the number to watch, since it reflects the real cost of running the business.
That cost is dominated by compute. OpenAI trains and operates frontier models on hundreds of thousands of accelerators, and the bills for chips, data centers, and electricity grow with usage rather than shrink with scale. The company has told investors it does not expect to be cash-flow positive until around 2030, an admission that the heavy spending phase has years to run. Its own projections of $115 billion in cumulative spending through 2029, reported by The Information, suggest management expects the cost curve to keep climbing even as revenue triples.
The burn rate has not deterred the private market. Investors valued OpenAI at $852 billion in March, and the company confidentially filed a draft registration statement with the Securities and Exchange Commission on June 8, according to people familiar with the matter, setting up a public listing that could value it at as much as $1 trillion. The filing will force OpenAI, for the first time, to publish its financials in a format regulators and analysts can scrutinize line by line. For investors weighing the IPO, the shareholder documents are the closest thing to an earnings preview, and they cut both ways: revenue growth is real, but so is the cost of delivering it.
The question hanging over that filing is whether revenue growth can outrun costs. OpenAI’s annualized revenue run rate now exceeds $22 billion, and its enterprise business has been the fastest-growing part of the company, anchored by deals such as the record ChatGPT enterprise agreement signed with Samsung. But the company’s spending plans imply that unit economics will not improve much as volume grows, a pattern that defies the usual software economics, where margins widen after the initial investment. OpenAI’s consumer business, meanwhile, counts on a ChatGPT user base that has grown far beyond what the company projected a year ago, with weekly active users now above 900 million.
For OpenAI’s rivals, the documents are a useful benchmark. Anthropic has raised more than $100 billion over the past year across three funding rounds, most recently $65 billion at a $965 billion valuation in May, with much of that money flowing into the same chips and data centers. The disclosures from OpenAI’s shareholder materials suggest the entire frontier-model industry is running the same experiment: borrow enormous sums, build enormous infrastructure, and bet that AI revenue catches up before the cash runs out. So far, the private markets have been willing to fund the bet. The public markets, which demand profit or a credible path to it, will be a harder audience.
