OpenAI's Computing Deals Top $1 Trillion (ft.com) 40
OpenAI has signed about $1 trillion in deals this year for computing power to run its AI models, commitments that dwarf its revenue and raise questions about how it can fund them. From a report: Monday's deal with chipmaker AMD follows similar agreements with Nvidia, Oracle and CoreWeave, as OpenAI races to find the computing power it thinks it will need to run services such as ChatGPT.
The deals would give OpenAI access to more than 20 gigawatts of computing capacity, roughly equivalent to the power from 20 nuclear reactors, over the next decade. Each 1GW of AI computing capacity costs about $50bn to deploy in today's prices, according to estimates by OpenAI executives, making the total cost about $1tn. The deals have bound some of the world's biggest tech groups to OpenAI's ability to become a profitable business that can meet its increasingly steep financial obligations.
The deals would give OpenAI access to more than 20 gigawatts of computing capacity, roughly equivalent to the power from 20 nuclear reactors, over the next decade. Each 1GW of AI computing capacity costs about $50bn to deploy in today's prices, according to estimates by OpenAI executives, making the total cost about $1tn. The deals have bound some of the world's biggest tech groups to OpenAI's ability to become a profitable business that can meet its increasingly steep financial obligations.
Prior art (Score:4, Informative)
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fen.wikipedia.org%2Fwiki%2F... [wikipedia.org]
"The telecommunications industry had experienced significant growth and investment during the 1990s, fueled by the expansion of the internet and the introduction of wireless technology. Companies such as WorldCom, Global Crossing, and Lucent Technologies had achieved enormous market valuations based on expectations of continued growth and profitability.[1] By the late 1990s, the industry had become overvalued and highly leveraged. Many companies had taken on substantial debt to finance their expansion, and investors had poured billions of dollars into the sector based on unrealistic expectations of growth and profitability.[2][3]
"... in the five years after the Telecommunications Act of 1996 went into effect, telecommunications companies invested more than $500 billion, mostly financed with debt, into laying fiber optic cable, adding new switches, and building wireless networks."
Re: (Score:2)
Most completely stupid things have prior art in human history. Most people are completely incapable of learning from history though.
Re: (Score:2)
The difference is these companies aren't necessarily going into debt to keep running. They just keep going back to investors who hand over their money.
Also, those other companies in the 90s used the debt to purchase and install physcal equipment needed for the upcoming expansion. That they sabotaged themselves with high prices and slow rollouts is the main reason for their failure.
Re: (Score:2)
Admittedly I am not deeply knowledgeable on this. While they are not issuing debt like corporate bond style debt, they are still borrowing this money. I's not coming from current revenue or capital they have amassed, rather it is in the form of agreements with other companies. Reminds me of 'swap' derivatives used by financial institutions, hedge funds, etc.
So for example, NVIDA lends Oracle $100 billion in GPUs, then Oracle lends OpenAI $100 billion in data center space, then OpenAI lends NVIDA $100 bi
Re: (Score:2)
Agreed. It's not the same, but similar issues remain. Investors, both from the stock market and direct, are the ones providing the money. So long as companies keep buying the products, everything is fine.
What has to happen is AI, in all its forms, needs to get its act together. While the market can stay irrational longer than you or I, you, as the company, need to show something for the billions being sent your way. Personally, I believe there will be advances from all this computing power. However, it
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
And by 2001, had collapsed.
And STFU, I was working for Ameritech, the midwest Baby Bell, '95- '97, We were a startup division, we were going to be your long-distance provider. 2 fucking years (and my wife semi-serious about suing them for alienation of affection, and me on the verge of burnout), they spent about $750M dollars... and then decided it was too much work, and shut it down.
Too big to fail cabal (Score:2)
Re: Too big to fail cabal (Score:2)
LED bulbs (Score:2)
I see that compute power has gone the way of light bulbs: compute capacity now measured in power consumption. "Buying 6 gigawatts of chips."
Required attribution (Score:3)
you know I did the Kessel run in under 6 GigaWatts - HS
Re: (Score:2)
Measuring by the amount of electricity they're going to guzzle down is an easy way to do that. It gives you a good scale
Re: (Score:2)
I think it makes sense. Electricity and water are the two major expenses for these companies. If you're reading these articles as a regular person then you are probably primarily interested in them as in investment opportunity and you want to know what the scale of their costs are going to be.
I would say that by far the primary metric should be ability to do useful work, i.e., what can be accomplished. The costs are important but secondary. Otherwise, we would hype up inefficient processors because they can use more electricity while doing less work. This should be true for both investors and engineers.
Bubble will burst (Score:2)
We've just got to count the seconds until it does.
requires high cost service (Score:2)
For each trillion spent they need 1 Billion subscriptions that provide over $1000 and in like 18 months. Can they reach a billion subscriptions at $60/month or more on average? Good luck.
Re: (Score:2)
Seems a bit far-fetched, doesn't it? Especially as their great shiny thing is not actually that good.
Re: (Score:2)
It is not bad either if you are aware of its limitations. Absolutely worth the current plus subscription price.
So this product isn't for you (Score:2)
The primary goal here is to automate White collar jobs. The product isn't for consumers it's for billionaires that employ white collar workers in the businesses they own.
And it is absolutely worth trillions to them. Because it's not about how much money it brings in it's about breaking the dependency the billionaires have on workers and consumers.
Re: (Score:1)
The prospect of AGI combined with the prevalent Silicon Valley elitist philosophy and the current brazenly unethical lot of politicians? I am quite happy to have reached the last quarter of my life.
Re: (Score:2)
Ok, sure, crazy, where do the profits come in? How do these billionaires expect to not lose money exactly? No matter which angle you take on "rich own and control everything", unless there are only rich people dealing with each other, the rich need to deal with the rest of us on some level. Their plan, as you claim it, fails when everyone starves to death. Money and/or slaves (electronic or otherwise) do not quell ambition, Billionaires don't stop when they reach a billion.
Re: (Score:2)
For each trillion spent they need 1 Billion subscriptions that provide over $1000 and in like 18 months. Can they reach a billion subscriptions at $60/month or more on average? Good luck.
The billion subscriptions at $1000 would balance out the trillion dollars, but why would that return need to come in 18 months? The return does indeed have to materialize at some point in time. However, if it comes in 5, 10, or 20 years, as long as it comes, the time frame is only important in terms of managing the expectations of those doing the financing. OpenAI doesn't yet turn a profit, but it's an exception among the large players, most of whom are self-financed and currently hugely profitable even
Re: (Score:2)
Have you ever started a company and needed outside investors? They expect returns in reasonable timeframes or they find something else to do with their money. Who is investing on a 20 year time frame?
Imagine booking a trillion dollars (Score:2)
And still not turning a profit. Wild times.
Oh crap, it's even worse (Score:2)
That's what they're *paying* for compute time. They're not earning a trillion dollars. Oh wow.
Re: (Score:2)
It's not about the real profit, but the potential.
Re: (Score:2)
And still not turning a profit. Wild times.
You have to read the fine print of these deals. For example, the recently announced deal with AMD is a huge windfall for OpenAI. The minor part of the deal is the purchase of some few billions of dollars worth of AMD GPUs. That's chump change compared to the really interesting part of the deal. OpenAI gets a warrant to purchase 160 million shares of AMD at an exercise price of $0.01 per share. At today's AMD stock price, that would be $33 billion dollars of free money handed to OpenAI. Some of the sha
Re: (Score:2)
The bigger and more interesting question is why AMD is willing to do this. Apparently exercising the warrant should result in the creation of 160 million new shares, which would increase the total number of shares by 10%. That's like a negative stock buyback and would be a huge minus for AMD. Yet, they are desperate enough to strike this deal.
Well isn't it obvious? AMD thinks that being on OpenAI bandwagon would raise their share prices (due to PR, due to increased business, due to AI hype, whatever) - and if the share price really reaches $600, then it will have tripled from what it has lately been. So diluting the shares by that 10% is a price that the stock owner will happily pay after that tripling...
Re: (Score:2)
The only way that would make sense for AMD is if they know OpenAI won't be around much longer after they pay for the GPUs.
Re: (Score:2)
Why would AMD be willing to do this? Because they think selling off 10% of their company to OpenAI for pennies will net them a fuckton of money in the future. And it might, if it manages to crack NVidia's near monopoly.
It's a time of contrasts (Score:5, Interesting)
Real progress is being made on AI systems that will help scientists and engineers solve previously intractable problems.
Immense quantities of energy and computing power are wasted creating useless, annoying and dangerous slop.
Irrational investors make crazy bets, hoping to strike it rich on "the next big thing".
People seem incapable of rationally focusing on making useful stuff.
I expect it to get even more insane
Just like last time (Score:2)
All the good parts of AI will be tainted with the irrational exuberance of "let's spend a trillion dollars because the chatbot can replace all your employees and replace all actors and artists and musicians". Well, I'm sure the parts that are beneficial to marketing departments will remain, because companies are literally willing to spend billions of dollars to increase personalized point-of-sale potato chip conversions 300%. That 300% is impressive. Pay no attention that means they sold three small bags of
Why OpenAI failed. (Score:2)
Don't read this comment now. Bookmark it and come back to it in 5 years. OpenAI failed because they wasted the $1 trillion dollars. They couldn't build the requisite data centers, the software couldn't improve because it got caught up in management quagmire .. they couldn't really scale far beyond what they have in 2025, and most of the CPUs they acquired sit idle.
Note .. we've seen the pattern before for example when Meta suck 10s of billions of dollars into VR with nothing to show for it. When GM, Ford, V
Re: (Score:2)
I would differ in one aspect: "the software could not improve because its not very good performance was the best that was possible and then model collapse hit, starting to make it worse".
Funny money (Score:2)
All those big deals are contingent on certain things happening (things that are highly unlikely to happen). And a lot of them are circular. OpenAI invests in Oracle which invests in NVidia which invests in...OpenAI.
Re: (Score:2)
Of course people (including me) will still be sad to see their 'wealth' plummet.
Re: (Score:2)
Paul Harvey used to say, "When it comes to the stock market roller coaster, it's those who jump off, that get hurt." It's still great advice today, and in the future when the bubble bursts.
So? (Score:2)