OpenAI CFO Says Annualized Revenue Crosses $20 Billion In 2025 24
According to CFO Sarah Friar, OpenAI's annualized revenue surpassed $20 billion in 2025, up from $6 billion a year earlier with growth closely tracking an expansion in computing capacity. Reuters reports: OpenAI's computing capacity rose to 1.9 gigawatts (GW) in 2025 from 0.6 GW in 2024, Friar said in the blog, adding that Microsoft-backed OpenAI's weekly and daily active users figures continue to produce all-time highs. OpenAI last week said it would start showing ads in ChatGPT to some U.S. users, ramping up efforts to generate revenue from the AI chatbot to fund the high costs of developing the technology. Separately, Axios reported on Monday that OpenAI's policy chief Chris Lehane said that the company is "on track" to unveil its first device in the second half of 2026.
Friar said OpenAI's platform spans text, images, voice, code and APIs, and the next phase will focus on agents and workflow automation that run continuously, carry context over time, and take action across tools. For 2026, the company will prioritize "practical adoption," particularly in health, science and enterprise, she said. Friar said the company is keeping a "light" balance sheet by partnering rather than owning and structuring contracts with flexibility across providers and hardware types.
Friar said OpenAI's platform spans text, images, voice, code and APIs, and the next phase will focus on agents and workflow automation that run continuously, carry context over time, and take action across tools. For 2026, the company will prioritize "practical adoption," particularly in health, science and enterprise, she said. Friar said the company is keeping a "light" balance sheet by partnering rather than owning and structuring contracts with flexibility across providers and hardware types.
Profits when? (Score:2)
I tried applying for jobs there trying to convince them on a better, faster, and less compute-intensive approach to ML using physics (like I have done for fintech), but they never responsed. I guess they are ok with hiring useless people to progress agendas and waste money and resources on pretending to be legitimate.
Re: (Score:2, Funny)
I too encourage companies to use quarks rather than electrons for ML They're smaller.
It doesn't have to be profitable (Score:2, Insightful)
People don't realize how much the billionaires resent and hate us. But both as consumers and employees they despise us.
Imagine if you are a borderline God because of how much wealth and influence you have but somewhere in the back of your mind you know all of that is completely dependent on other people's work and on talking people into buying your products.
That's going to eat away at you and you're going to hate it.
AI offers the possibility of the ruling c
oh good for them (Score:2, Funny)
only * checks notes * one seventieth of their required revenue to be profitable.
a cause for celebration.
Re: (Score:2)
GW vs flops? (Score:1)
why measure GW instead of flops? wouldn't using flops as a metric incentivize more energy efficiency?
wouldn't GW measure usage whereas flops would measure capacity?
Re: (Score:2)
GW lets you include the air conditioners, making the resulting number much more impressive!
Re: (Score:2)
It's an odd metric to use (why not FLOPs, or tokens consumed and generated?), but perhaps it makes sense if you want to convince investors that "to double growth we need another N GW datacenter)?
Re: (Score:2)
It lets us do some napkin calculations though. Lets say this stuff runs at 50% load, then thats about 3.5 GW. OpenAI probably truly runs 24/7 without much load changes so that is 8760 hours per annum.
That is about 30.6TWh. Unless my math is completely off this early in the morning, we're talking about annual cost of 300 million if we assume just one cent per kWh in cost.
You can now extrapolate power cost for any price you wish to assume. At 20 cents per, we're looking at 6 billion, no?
And that doesn't inclu
Re: (Score:2)
Re: (Score:3)
They're explaining that they don't understand why flops were important before, they were just big number go up.
They don't understand that there might not be a fixed or linear relationship between flops and watts. They don't want you to think about it either, and I'm not joking when I say that I've had several supposedly notable people in the AI space have claimed to be unfamiliar with the term 'computational complexity'. I still don't know whether they were serious or being rhetorical to preserve their payc
Re: (Score:2)
The relationship between FLOPS and watts is a function of GPU generation, and will certainly change (and be very disruptive - increased power density may require entire datacenter cooling to be redone), but what may change even faster is tokens per FLOP as the models get tweaked for efficiency, and this is what counts since customer pricing is in tokens. The production capacity (tokens/sec) of the "factory" is certainly far from fixed and defined by the power it is consuming.
GW really is an odd metric to fi
Re: (Score:3)
with respect
I think you have missed my point. What you're saying isn't wrong, but what I mean is that any operation performed by an LLM will necessarily be orders of magnitude less efficient per watt than just doing it directly. Necessarily.
This is what computer science is about, and no amount of business guys trying to rewrite the (spoken, purely semantic) language will change that. No amount of token generation or discarded hashes will ever be necessary to perform basic, defined, known arithmetic... let a
Re: (Score:2)
Yes, when there is an alternative it is almost certainly more efficient, and people no doubt are sometimes asking LLMs to do trivial things like math where when they could have just used a calculator instead, but surely you don't believe that the majority of paying LLM users are stupid and using it for things like this? The excitement about LLMs is because they can do things, like writing code, where there is no alternative (other than doing it by hand).
Re: (Score:2)
what?
No I think it's much worse than that. I think by volume the average active use is somewhat literally the act of pushing pixels around a bitmap so that it looks more like a naked lady.
I understand that a functional positive use of LLMs is to enable people to grapple with ideas they wouldn't necessarily have access to... And as such my net experience of them has been that I have undergraduate researchers trying to log into HPC systems with cursor and wondering why there's a problem, and also my friends w
Re: (Score:2)
You seem to be completely out of touch with what people are using LLMs for.
Incidentally, if you want to use an LLM to do everyday computer stuff like logging into a system (some people use it as their system administrator) then what you need is an agentic client that actually runs on the system you want to control, such as Claude Code, Gemini CLI, etc. You seem to be blaming the stupidity of your undergraduate researchers on LLMs, when it is in fact their own stupidity and lack of having bothered to underst
Re: (Score:2)
I see.
I have this nice bridge over here for sale, and you seem like a canny investor.
Imaginary numbers (Score:4, Informative)
"Annualized" just means that if you take the highest month and multiply by 12, you'd get $20 billion. It doesn't mean they had $20 billion in revenues in 2025.
Re: (Score:3)
Surely it'd be latest month, or quarter, being extrapolated, not highest.
It's a bit of a strange quibble though given that by same logic (growing revenues), they most like have 2026 revenue well in excess of $20B.
Their problem, evidentially isn't revenue, it's spending, and AFAIK they are not expected to break even until 2030 (assuming they last that long).
Re: (Score:3)
It probably is, as you suggest, the most recent month. But while the trend is up for now, it's difficult to predict how long the trend will continue to go up. Many businesses are already pushing back against skyrocketing AI costs, not seeing the promised returns. https://finance-commerce.com/2... [finance-commerce.com]
Who is paying for it? (Score:3)
If you are an end-user, Gemini is way better, more functional, and comes with your Google subscription
If you are a company - come on. Claude is WAY WAY beyond anything OpenAI does.
Old Chinese saying (Score:2)
$20 billion income, $20.1 billion expenses, man sad
$20 billion income, $1 trillion expenses, Altman is so fucked