Sam Altman has said the quiet part out loud more than once this year. Rising AI costs are a “huge issue,” he said. Ads in ChatGPT, something he once called a last resort, are now live for free-tier users. The numbers behind those comments are bigger than most people realize.
Audited financials reported by the Financial Times put OpenAI’s actual 2025 spending at roughly $34 billion against $13 billion in revenue, well above the $22 billion figure that circulated in earlier reports. Internal projections The Information first reported in 2024 pointed to a $14 billion loss for 2026 alone. Real 2025 losses, roughly $21 billion in core operating losses or $38.5 billion including a one-time restructuring charge, have already exceeded that number. Analysis from Klover.ai puts cumulative cash burn through 2030 as high as $218 billion. This is not a company quietly building toward profitability in the background. It’s a company spending faster than almost anything in corporate history, betting that the payoff arrives before the money does not.
OpenAI’s revenue jumped from $3.7 billion in 2024 to $13.07 billion in 2025, beating the company’s own $10 billion internal target. That’s real growth. It’s also nowhere close to covering what OpenAI spent to get there.
Total costs hit $34 billion for the year. About $19 billion went to research and development, close to $6 billion to sales and marketing, and $7.5 billion to what’s called “cost of revenue,” the direct expense of serving every ChatGPT query, up from $2.65 billion the year before. Microsoft alone collected $17.2 billion of that spending, including $10.59 billion for R&D compute, more than half of OpenAI’s entire research budget.
Strip out the one-time accounting charge from OpenAI’s 2025 conversion to a for-profit structure, and the core operating loss still widened from $8.78 billion in 2024 to $20.92 billion in 2025. Include the charge, and the net loss attributable to OpenAI comes to $38.5 billion, roughly 7.5 times what it lost the year before.
Here’s the part that should worry anyone betting on a turnaround. Back in October 2024, The Information reported internal OpenAI projections showing a $14 billion loss for 2026, a figure the report itself called roughly triple early 2025 estimates. OpenAI’s actual 2025 loss already exceeds that number. The company hasn’t just missed its own targets. It’s missed them by billions, a year early.
A single ChatGPT query costs somewhere between one and ten cents to run, according to a breakdown by Asia Times. A high-definition image generation runs closer to twenty cents. Small numbers, until you multiply them across the queries OpenAI handles every day.
The hardware behind it isn’t cheap either. Nvidia’s data center chips run into the tens of thousands of dollars each to buy, and several dollars an hour to rent through the cloud. OpenAI runs tens of thousands of them around the clock. Model training adds another layer on top of that, and so does the research talent needed to keep pushing capability forward.
None of this gets cheaper as more people use the product. If anything, it gets more expensive. More usage means more compute, and more compute means more infrastructure spend.
OpenAI now has more than 900 million weekly ChatGPT users. About 50 million of them pay for anything, somewhere around 5 to 6 percent. Every query from the other 850 million still costs OpenAI real money to serve, whether or not a dollar comes back.
That’s the gap advertising is meant to close. Altman spent over a year insisting ads in ChatGPT would only happen as a last resort, a stance Windows Central covered at length. Early this year, that resort arrived. His explanation shifted from resistance to something closer to acceptance: people want to use a lot of AI and don’t want to pay for it, so the company is betting a business model built around that reality can work.
Altman’s framing is that these losses are the price of building infrastructure that outlasts them. It’s the same logic Amazon and Uber ran on for years. Spend aggressively now, let the losses fund market position, and let profitability follow once scale kicks in.
Software got cheaper to distribute as it scaled, because the marginal cost of serving one more user approached zero. Compute doesn’t behave that way. Every additional user still needs GPU time, still needs power, still needs infrastructure that has to be built and maintained. Scale doesn’t erase the cost the way it did for earlier tech companies. It moves the number to a bigger stage.
So why does the money keep arriving?
In March 2026, OpenAI closed a $122 billion funding round, up from the $110 billion figure it had announced a month earlier, at a post-money valuation of $852 billion. Amazon anchored the round with $50 billion, Nvidia and SoftBank each put in $30 billion, and Microsoft continued participating. It’s the largest private funding round in Silicon Valley history.
Then, in early August 2026, OpenAI closed a $7 billion tender offer at that same $852 billion valuation. No increase. It’s the first flat print across the company’s entire fundraising history, and several outlets read it as a sign that OpenAI’s confidential SEC filing, submitted back in June, isn’t about to turn into an actual public listing anytime soon.
That matters more than a footnote. Investors aren’t betting OpenAI is profitable today. They’re betting it becomes too central to the AI economy to walk away from before the losses catch up with it. A flat valuation print is the first real hint that some of them are starting to hedge that bet, even as OpenAI reportedly still targets a valuation of up to $1 trillion whenever it does go public.
OpenAI isn’t the only lab spending far more than it earns. It’s the clearest public example of what running the current AI business model actually costs, because leaked financials have made its numbers public in a way competitors’ haven’t.
Watch what happens next. Price cuts arriving weeks after a model launch. Enterprise customers pushing back on token costs loudly enough for a CEO to call it a huge issue in public. Competitive pressure from cheaper Chinese models, and from Google and Microsoft undercutting on price from the infrastructure side.
The AI industry spent the last two years selling a story about inevitable capability growth. What matters just as much now is whether the economics underneath that story can hold. OpenAI is targeting roughly $600 billion in total compute spend through 2030, against a hoped-for $280 billion in annual revenue by then. Whether that math closes, or whether OpenAI runs out of runway before it does, is arguably the defining financial question of this entire era of AI.
The next signal to watch isn’t another funding round. It’s whether the next one, if there is one, prices higher than $852 billion, or flat again.
Why Is OpenAI Losing Money? Inside the $34 Billion Bet on ChatGPT’s Future was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.