OpenAI is still figuring out how to make money, but wants you to believe in it

This week, OpenAI CFO Sarah Friar took to the internet to make a bold pitch for the company's future, which she claims is bright, despite what the current numbers say. If you buy her logic, you must believe some things to be true, regardless of the poorly connected thread of logic that appears to hold them together. In what we …

[Anonymous Coward]

History rhymes

I seem to recall the same with twitter, facebook and the dot com bubble being a very similar situation. They are losing money but it will be worth it. Maybe

[Irongut]

Re: History rhymes

The Internet had already shown itself to be useful before the dotcom bubble. It also didn't require all the data, energy and land it can consume to build datacenters.

[Anonymous Coward]

Re: History rhymes

@Irongut

"The Internet had already shown itself to be useful before the dotcom bubble. It also didn't require all the data, energy and land it can consume to build datacenters."

But social media and the variety of companies in the dot com bubble didnt show themselves useful.

[hedgie]

Re: History rhymes

Even some of the "failures" of the dotcom era were, unlike AI, based upon sound ideas. They just didn't have the leftover cash to survive after building up infrastructure to weather the storm. Webvan and grocery deliveries was great in principle, but they spent all their money getting everything up and running. The big supermarket chains were able to buy all that stuff up at firesale prices and started offering that service. For a while it looked like Amazon was going to end up there as well, but had enough to hold on and keep going.

[Michael Strorm]

Re: History rhymes

Around the time of the dotcom burst- I think it was shortly before, but don't quote me on that- someone said that the Internet *would* ultimately be as big a deal as those companies and many others were claiming... it just wasn't going to happen overnight or as quickly as they needed it to happen.

That was extremely prescient in hindsight, but I'm not sure you could make the same compelling argument regarding gen AI.

[hedgie]

Re: History rhymes

Exactly. Every potentially useful application of "AI" would be in automating *small* routine stuff, and local models would be fine for all of that. These gen AI companies are just spewing nonsense (and now feeding on their own slop) that's only going to get worse. A lot of the dotcom failures were the ones with a silly idea or nonsensical business plan that we all laughed at, but others were just because they didn't have the money to keep going until profitability or because the technology (or connectivity) weren't there yet.

The gen AI seem more like the former, promising the world but failing to deliver, and all at a staggering cost that's being largely borne by the rest of us between the skyrocketing costs of hardware, and likely soon, electricity, job cuts by execs gleeful to replace humans with AI for a bigger bonus, and the environment.

[DS999]

Re: History rhymes

Its easy to say "it will be worth it" when you look only at the big winners like Google and Facebook. You need to also look at the big losers like pets.com and myspace, and there were many more of those than there were big winners. Dating back to the 80s AI has been through hypecycles before where it reached crazy pinnacles of unrealistic expectation, then fell to Earth and filled some roles but didn't remake entire industries. I expect the same is true this time.

The level of investment planned by 2030 requires that it displaces millions if not tens of millions of jobs, but other than call centers I can't see anything that it is really capable of displacing at the current technology level - and those jobs were already outsourced to low wage countries so the savings are limited.

[Anonymous Coward]

Re: History rhymes

Trying to identify the winners from the losers is the trick. Sometimes it’s luck. Remember Friends Reunited, MySpace, AOL or Yahoo! (past finance).

“ If you buy her logic, you must believe some things to be true, regardless of the poorly connected thread of logic that appears to hold them together. In what we can only assume is a pitch to soften up the market for more investment and possibly an IPO, Friar argues that one of the things people should believe about OpenAI is that the more money it spends, the more money it makes.”

Perhaps ChatGPT wrote it.

[BartyFartsLast]

Just guessing

Did the investors ask AI for financial advice by any chance?

[Richard 12]

"revenue followed the same curve"

Which means they can never, ever make a profit. The more customers, the bigger the loss.

It's the classic "yes, we lose money on every single transaction, but we'll make it up in volume".

When will HSBC et al call in their loans?

[UCAP]

Re: "revenue followed the same curve"

The banks won't dare to call in the loans. Doing so will result in OpenAI defaulting & going into Chapter 11, at which point the losses to the banks will be so big that their shareholders are going to start looking for some CxO heads to put on spikes.

[Doctor Syntax]

Re: "revenue followed the same curve"

So the banks go into Chapter 11 or it's equivalent instead.

Somebody has to decide what the level of losses will be. Throwing good money after bad is not a good idea. If the losses are inevitable the sooner they're realised the less they will be.

[Anonymous Coward]

Re: "revenue followed the same curve"

They just have to find a way to insure themselves against it. That's what the credit default swaps were in the 07-09 crash.

[Filippo]

Re: "revenue followed the same curve"

The trick is to let the debt become so big that a default would collapse the entire US economy. At that point, you bail everyone out with taxpayer money. Then, if anyone suggests to deploy regulation to prevent this from happening again, shoot him.

[Anonymous Coward]

Re: "revenue followed the same curve"

Please and let that be the Orange c*nts legacy - ontop of being an poor imitation Orange Hitler

[Pete 2]

Proof of the pudding

This tells us everything we need to know about the state (or potential threat) of AI.

If it was half as powerful as it's detractors make out, it would have worked out how to turn a profit for itself. Since there is absolutely no sign of it ever earning itself a penny, we can safely dismiss any of the silly notions that it is capable of dominating the world.

[Anonymous Coward]

Re: Proof of the pudding

... unless its cunning plan is to dominate the world by bankrupting us all ...

[rgjnk]

Compute in GW

Can we please kill this idea of measuring compute in power consumption terms?

Beyond infrastructure/waste it's an utterly meaningless metric.

[Doctor Syntax]

Re: Compute in GW

It tells you all you need to know. The operation is measured in terms of its inputs because they cannot quantify useful outputs. If they could they'd use that measure instead.

[Michael Strorm]

Current AI "progress" is mainly money and hardware-driven brute force

It's also a reflection of the fact that the vast bulk of continued growth and "progress" in generative AI is now being driven far more by brute force- i.e. simply throwing increasingly ludicrous amounts of computer hardware, electrical power and money at it- than by fundamental improvements in the underlying technology.

That's a stark contrast with the original computer revolution which only happened because of exponential improvements in digital technology and the likes of Moore's Law. One wonders how far they would have got without that- if they'd tried building ever more powerful machines purely by building more and more warehouses full of valve/vacuum tube-based hardware.

We're already at the point where the continued investment required for this approach must be near its limits.

[Paul Herber]

Re: Compute in GW

It's like measuring your capabilities by how many pints you can drink.

We can measure the input.

The output must be similar but we don't tend to measure that!

Useful work? Hmm. well, immeasurable.

[Michael Strorm]

Re: Compute in GW

> "The output must be similar but we don't tend to measure that!"

I suspect that would be taking the piss.

[Filippo]

Re: Compute in GW

"Here is our new car. It consumes way more gas than the previous one, so it must be better."

This actually would make sense to some people.

[VoiceOfTruth]

OpenAI, and 'AI' in general

In a nutshell:

>> OpenAI is clearly hoping something will emerge.

Investors: let's pour money in.

[Michael H.F. Wilkinson]

... new economic models will emerge

Or else we will let AI hallucinate these for us?

Just asking

[m4r35n357]

Re: ... new economic models will emerge

I can only assume that is exactly what is happening.

[Anonymous Coward]

"if expectations about AI gains in productivity and profitability are not realized."

When do they actually make this judgement? They've been saying this for two years already. You can say "Well they haven't materialised...*yet*, but we're sure they will." forever.

[Tron]

AI is not too big to fail.

quote: How this might unravel is yet unclear.

It will fizzle out like a damp squib. The AI features will just stop working.

The tech sector will still have tonnes of cash. Because they all took in a load of investor's money and then paid each other. Clever stuff, eh?

The 'investors' will be thanked for their donations, and there will be much regret that things just didn't work out the way everyone had hoped.

You lost cash? Well, let's hope you followed the golden rule of investing/gambling and only risked money you could afford to lose. If not, don't slam the window on your way out.

[ComicalEngineer]

Surely AI will be able to tell us how it can make money as it is the fount of all human knowledge?

Or not.

[Bluck Mutter]

3 billion ChatGPT users by 2030????????????????

We have 8.3 billion people in the world (8.5 by 2030 ) and as of 2025, an estimated 3 billion people worldwide are not using the internet/don't have access

There are 1.5 billion under 10 years old in the world whom we can safely say won't be using AI (or if you want a tighter data point 800 million under 5's)

So we have a possible population of ~3.7 billion who have access to the internet and are potential AI end users and they believe 75%-80% of these people will use ChatGPT?

And define "use". Occasional free accidental use due to it being built into some web site or 3 billion people PAYING to use it.

Yeah...right.

Bluck

[Bluck Mutter]

forgot to add....

And a large portion of the that ~3.7 billion potential user base reside in China who will therefore be consuming Chinese provided AI services given the hostile nature of the US towards China.

And that same sentiment will play out in places such as the EU whom are wary of US tech and the general "nasty" attitude of the US to the EU.

And lets not mention the peoples of South America who will have a renewed dislike of US imperialism.

Anti US global sentiment will affect US business utilization (eventually)

So "Yeah...right" x 2

Bluck

[jake]

Sounds to me like ...

... Someone is trying to start a new religion.

My advice? Don't drink the Kool-Aid.

[Anonymous Coward]

Re: Sounds to me like ...

Yeah, it's like missionaries of dementia all around imho, tirelessly proselytizing exalted enlightenment of devoted conversion to the cult of nutbag.

Who'd ever bet their GDP growth and retirement income on such nonsense!?

[David Newall]

Do LLMs experience time?

I argue not because they "learn" up to a point, after which they merely regurgitate with no mechanism to learn more. I am stunned by what they can do today, but not one whit by their total and complete absence of actual intelligence. Let the learning continue in production and who knows what will emerge.

[Inventor of the Marmite Laser]

So it's still

1) Get AI

2) Profit

[excession]

Re: So it's still

You missed:

1.5) ????

[T. F. M. Reader]

GW to dollars

I did look at the good CFO's Internet missive, primarily to find out whether she actually provided any useful numbers at all about OpenAI's spending growth. The Reg's piece didn't, and neither did she. She gave the revenue numbers in $ (faithfully quoted by El Reg), but compute spending was only given in GW.

This raises the question: if the money spent is less than revenue (where revenue scales roughly linearly at $10B/GW, according to the CFO herself) then the CFO has something to boast about, but if, as I suspect, it is the other way around then the "3x growth every year" pitch does not look so good, does it?

But I don't really know, so to find out I did the obvious thing: I asked ChatGPT.

Me: "Assume I need 1GW of cloud compute power in the next year, from Microsoft Azure, mostly from GPU. Give an order of magnitude estimate how much the compute resources would cost me in US dollars over the year."

ChatGPT gave a long answer in several items. I summarize below to save space here, I assure you I do not misrepresent anything.

1. "[I]n a recent industry report it was noted that Anthropic’s potential 1 GW compute commitment could cost roughly $20–$25 billion per year on hyperscale cloud infrastructure." [No citation or link was offered, so I chose to believe the bot.]

2. GPU-hour for Nvidia H100 on Azure in some 2025 comparisons costs ~$7 (on demand, the bot notes, spot/discounted proces may be significantly - 25-30% - cheaper). [I didn't check the numbers either here or elsewhere below.]

3. A single NVIDIA H100 GPU might consume ~0.3 kW of power. 1 GW of power capacity ≈ 3,333,000 H100 GPUs running at full utilization. 1yr = ~8760hr if run 24/7. That works out to 29.2B GPU-hours per year at 100% utilization, or 14.6B GPU-hours at 50%, ~3B GPU-hours at 10%, which corresponds at needing a lot of compute (e.g., for training) for only a few weeks in a year. [Lower utilization scaling is mine, the bot is not to blame.]

4. At $7/GPU-hour the 100% utilization case works out to something like $200B/yr (ouch!). At 50% it's $100B/yr (still ouch!). Assume 10% utilization and a 30% discount and you get down to ~$14B/yr. Not as big an ouch, but Mrs. CFO needs a better argument, methinks, especially since I asked the bot to look only at one component of the cost.

Standard disclaimers apply.

[teebie]

"The more you spend, the more you make"

Yes, that is generally true. What needs to happen is for the amount you make to exceed the amount you spend, which doesn't look like it will happen any time soon.