Sam Altman is back tweeting non-stop. That can only mean one thing: something’s cooking. Or rather, was cooking and it’s now ready to be served by some of the world’s best chefs.
One tweet caught my attention:
we believe the world needs more ai infrastructure—fab capacity, energy, datacenters, etc—than people are currently planning to build.
building massive-scale ai infrastructure, and a resilient supply chain, is crucial to economic competitiveness.
openai will try to help!
Oh, that sounds like it’d need $7 trillion or so.
The next day, the Wall Street Journal published the news of the week: “Sam Altman Seeks Trillions of Dollars to Reshape Business of Chips and AI.”
The article mentions interesting details of Altman’s next venture—his relationship with the UAE, conversations with TSMC, the US government’s stance, or Microsoft’s support—but, of course, what caught up was the $7 trillion number.
It’s been a target of jokes, memes, criticism, and hypotheses as to why Altman could want such a large sum—or what it reveals about OpenAI’s plans, their yet-to-be-disclosed achievements, and the future.
Some argue wanting to devote almost 10% of the world’s GDP to building GPUs (or a new generation of AI-specific chips) is a waste given the amount of natural resources that would be required; or claim such an amount of money reveals GPUs are not the right architecture to power the AI systems of the future.
Others applaud that Altman is taking his ambition to new heights to accelerate the future; or make optimistic predictions about what’s going on under the hood at OpenAI: the chip shortage is so profound that Microsoft’s help isn’t enough or they’ve found the path forward and are going all in.
Okay, so $7 trillion is undeniably a lot of money.
Why do we need so many AI chips? Is this a geopolitical move backed by the US government? Is this much money needed to create AGI? Does Altman’s ambitions make any sense at all?
Yes, they do make sense—to him.
To understand what’s going on we have to force our mind frames to overlap with his as much as we can. This article is my five-part attempt at thinking like Altman (I’m not his friend so his public comments are what we’ve got).
(Note to nitpickers: this article isn’t a judgment, excuse, or acceptance of his beliefs, reasons, or goals. It’s an exploratory analysis).
(And, just to be clear, I don’t think this is happening; I don’t think Altman is getting the $7 trillion. I also don’t think this is a PR stunt or a persuasive/deceptive tactic. I believe it’s genuine.)
Altman wants $7 trillion. That’s larger than the value of the current semiconductor industry. It’s more than the market cap of Microsoft and Apple combined. It’s more than the sum of Germany’s and the United Kingdom’s GDP. Humanity has never embarked on a project as expensive as that. By far.
It’s not a stretch to say that amount of money could redefine modern society if it were used to guide it in the right direction (which may not be Altman’s preferred choice). That’s what I think he wants to do: improve our technology to the same extent the Industrial Revolution did.
Humanity’s current best tech (both at the software and hardware levels) can “only” produce GPT-4-level systems (soon GPT-5-level), which already cost north of 100s of millions of dollars—just to train. I don’t know which GPT number, if any, will become AGI, but with a single AI system soon crossing the $1 billion mark, suddenly $7 trillion to back up the efforts isn’t so disparate.
Whether Altman will need to transform those tasty American dollars into human ingenuity—money can buy most things but not scientific breakthroughs—is a different question. Although we can actually answer the question of “Where does he want to invest the money?” because Altman himself has told us somewhat.
He wants to stimulate the construction of new semiconductor fab plants (with the help of TSMC, the world leader) possibly to overcome the limits of current hardware. He wants to improve the energy supply and has invested in fusion initiatives for years in hopes they achieve a breakthrough. He also wants to have a larger datacenter supply (Nvidia GPUs aren’t enough and Microsoft’s Azure Cloud isn’t enough).
So not only AI systems are getting exponentially more expensive but he wants to 10x the entire supply chain that makes them possible. Combined—models, fabs, fusion, cloud—that’s not a one-year project but a 10 or 20-year one. Perhaps more. The WSJ says “The effort could span years and ultimately might not succeed.” (To those who think AGI will happen next year, sorry to disappoint you.)
That’s why the $7 trillion number isn’t crazy at all. It’s not intended to build GPT-6 but to give birth to a new world.
If we run some numbers, it even starts to seem too little.
To give an estimate, let’s say the project is finished in 20 years. In 2044, that amount of money won’t be as impressive (or outright impossible) in relative terms. In the year 2000, the world’s GDP was $33 trillion. Today it’s $100 trillion. Assuming linear growth (conservative if nothing drastic happens) at a constant 3% annually, we could be at $200 trillion in 2044.
Anecdotally, Microsoft was founded in 1975. In 1975, the world’s GDP was $6 trillion, less than what Altman is seeking. Almost 50 years later, Microsoft’s market cap is half that number. Could Bill Gates have predicted that?
Altman is simply allowing the future to be as crazy as it always seems to be when viewed from the present.
I don’t think he expects to get the entire package, though. Like Nvidia CEO Jensen Huang said about the $7 trillion number, “Computers are going to become faster and therefore the total amount that you need is not as much.”
Whatever the case, saying $7 trillion is an absurd number is far from a good analysis.
