Monday, we looked at the weirdest thing going on in the world of artificial intelligence — autonomous bots sharing “skills” on their own no-humans-allowed social media platform — but today comes the sobering reality that the biggest and best known AI startup loses billions of dollars a year pursuing diminishing returns, and its only out is to ask investors for more money.
OpenAI, you have a problem.
Bear with me because we need to go over a bunch of very big numbers in today’s column, and frankly, the scale looks like something out of a Biden Cabal spending bill. I’m kidding, of course — nothing matches the scale of a Biden Cabal spending bill, and besides, unlike anything from those days, there’s real utility in AI.
In the three years that large language models have been genuinely useful to the general public, OpenAI — creator of the popular ChatGPT chatbot — has raised serious cash from investors, banks, other companies, etc. It’s impossible to find firm figures for 2023, but the company completed a $6.6 billion capital raise in 2024 from Microsoft, Nvidia, and SoftBank. They raised an eye-popping $40 billion last year, led by SoftBank with participation from Microsoft, Coatue Management, Altimeter Capital, Thrive Capital, and others.
The company is believed to enjoy a total cumulative funding of somewhere around $57–64 billion by the end of 2025.
There’s nothing new or panic-worthy about a tech startup blowing through oodles of other people’s money, but even last year’s massive funding might not be enough to cover OpenAI’s 2026 operations. The base scenario is that they’ll need another $10-$20 billion in 2026 just to keep the servers humming and paychecks from bouncing. The “stretch” scenario is $30 billion.
From there, things get more difficult to forecast, so I chatted with ChatGPT about its owners’ financial needs, and after tossing around various scenarios, GPT projects another $15-25 billion in fresh investor funds in 2027, and (assuming continued bullish growth) $30 billion or more in 2028. If there’s what GPT called “an arms race” between OpenAI and the competition, 2028’s funding needs could rise even higher.
I know the headline says, “Buddy, Can You Spare $50 Billion?” but the real baseline figure between now and 2028 might be closer to $65 billion.
But let me reiterate that a tech startup in a growth field is expected to burn through sums that normal people can barely imagine. The rub comes when you look a little deeper and ask, “What are investors getting for that money?”
The nice answer is “diminishing returns.”
GPT’s current top-of-the-line model, GPT-5.2, debuted late last year, supplanting 2024 GPT 4o, which obsolesced GPT 3.5. The thing is — and this is where that cash burn comes into question — 4o was 12.5 times more expensive per “token” to operate than 3.5, and 5.2 costs something like 50 times more than 3.5 did. (Tokens are an LLM’s base unit, to put it in the simplest possible terms.)
Despite massive cost increases, actual utility improvement for most users — the kind using the free or $20-a-month plans — is marginal. In fact, I canceled my $20 plan because even though OpenAI now spends an estimated 10 cents per 1,000 tokens (compared to $0.002 for 3.5), ChatGPT can no longer keep up with current events. The breaking point was when I asked it to edit my column about Operation Gideon to arrest former Venezuelan president Nicolás Maduro, and GPT practically shrieked at me, insisting that no such thing had ever happened.
It gets funnier. Writing this column, I had to switch to Grok because GPT 5.2 insisted there was no such thing as GPT 5.2, and again, scolded me for believing such a thing. So then I had to waste time, running the dollar figures back through Grok to confirm accuracy — accuracy GPT no longer seems to possess.
For this, their costs per token have increased 50 times over?
“We lose money on every prompt, but we make it up on volume!”
But let’s get back to those dollar figures.
OpenAI generated about $6 billion in revenues (not profits, just revenues) in 2024, $20 billion in 2025.
Revenues are growing aggressively, but not nearly as quickly as the expense of designing, building, operating, and (most of all) training more advanced models. As a result, OpenAI burned through $5 billion in ’24 and $8 billion in ’25. Spending keeps growing faster than revenues.
And Another Thing: Also, OpenAI openly discusses allowing NSFW content generation (short of pr0n, but we’ll have to wait and see just how short) and using users’ most private data to insert ads for the first time into GPT chats. Who knows, maybe one or both will help turn the corner.
The hope is that last year’s massive capital infusion will result in some kind of gotta-pay-for-it feature or internal efficiency that can accelerate revenue growth even further, or reduce costs. As yet, however, don’t even think about profits. Not for a while yet, and for many firms, never. The real question for OpenAI is whether it can reach positive cash flow before investor patience runs out.
The brass ring is out there, somewhere, but nobody can quite see it yet.
Recommended: Trump Was No Draft-Dodger, and Here’s the Timeline That Proves It
Support Human-Generated Content for as Long as the Machines Let You!
I kid, I kid. But for real, PJ Media VIP members get deeper dives, exclusive columns, and commentary you won’t see on the public side. Best of all, you can get 60% off right now with the promo code FIGHT.
Sign up for PJ Media VIP and help us keep telling the stories that matter.
