On Feb 8, 2024 the Wall Street Journal broke the news that our favorite AI wunderkind, OpenAI CEO Sam Altman, was making the global rounds again (primarily to China and Arab States this time… interesting), this time trying to raise an alleged $7 trillion. (that’s 7,000 billion, or seven million million dollars!) So what in the hell is he thinking? What are the dirty details on Sam Altmans 7 Trillion AI bet? We dish:
First, let us assume that Sam & Team actually did some research, and have some spreadsheets to support that number. I mean, if it was just an all out SWAG, or blustery swagger — you figure he’d ask for an easy round number… you know: one trillion, or ten. but no. Seven is a very specific number, and implies some sort of logic / assumptions / justification. We can hope.
Aaand: He figures he’s just the man to make it happen, and is seeking $7T of “funding” from individuals, sovereign wealth funds, corporations, global banks, and nation-states / govts (see breakdown below on who might actually be practically capable of contributing… hint: it ain’t VCs…)
So: what the hell is all that money for? I mean, that number is 10 times the entire annual budget for the US Military… and that bloated budget (~800 billion) funds not only salaries for roughly 1,000,000 personnel, but also all the new equipment purchases and existing equipment operational and maintenance costs… (11 air craft carrier groups, 1,000 jet fighter planes, nuclear submarines, all our nuclear missiles…)
From my understanding, there are three really big buckets that we expect this $7 Trillion AI bet to be dumped into (these multiples of existing figures are my own personal educated guesses):
- Global Energy generation infrastructure
that is: new nuclear / solar / fossil power plants
Goal: double existing global electrical power generation capacity by 100%
see related: How much global electricity does AI consume?
. - new AI chip fabs
production capacity, and new chip architectures / design
Goal: increase global AI chip fab production by 10x… i.e. 1,000%
this is currently dominated by nVidia (see separate story)
Microsoft, AMD, Intel and Arab/China G42 all have designs on getting pieces of this pie
TSMC wins no matter what
. - data center build-outs.
Once you have all those shiny new chips, you need places to put them in.
In modern parlance, these places are known as datacenters.
This is also known as “cloud compute services,” “global compute grid”, or “cloud supercompute”
current leaders are the usual suspects:
AMZN, GOOG, AAPL, MSFT, META, TSLA
1. More Electricity : Power Plants
A new nuclear power plant costs roughly $8 billion
and takes 5 to 10 years to bring online.
Famously, an AI engineer for Meta recently told the US Congress: “If you’d just reserve 2 nuclear power plants for us, that would effectively solve the energy needs of our AI training for the next year or two…” Yes, its true: I absolutely love the humility of AI people. I wonder if he said “Please”?
“If you’d just reserve
2 nuclear power plants for us,
that would effectively
solve the energy needs
of all our AI training
for the next year or two…”
An alternate to this plays right into Altman’s hands: nuclear fusion power. Potentially far safer and more efficient than nuclear fission plants, fusion energy generation has been the holy grail of the energy R&D community for well over 50 years now. No one has come even close to achieving a “breakeven” reaction (something that generates more power than it took to start the reactor). But it looks like we might be getting close.
Altman has personally invested more than $100 million into a leading fusion company, Helion.
Demonstrating a successful fusion reactor, and building out global plants, could truly be an answer to both humanity’s, and AIs, voracious need for cheap electrical power in the 21st century.
2. Faster Chips : New Fabs
A new chip fab is generally a $2 to $6 billion ($0.005T) investment,
and takes 2-4 years to build the physical factory.
All we need to do is take one glance at nVidia stock (and quarterly financial reports) to understand why this matters. The company — which doesn’t actually own a single fab — has seen its fortunes soar, as many see its A100 and H100 GPU chips as the only game in town to train the world’s Foundation / Frontier AI models (Google and AMD beg to differ… Intel prays it can make a difference in the coming decade).
3. More Compute : Global Datacenters
Data Centers are the new supercomputers.
A new data center can cost up to $2 billion,
and can be brought online in 12 months or less.
MSFT alone has already committed an unprecedented $0.25T — $250B — to data center buildout across the next 5 years. This should tell you how seriously corporate America is taking AI. In fact, that kind of puts their outrageous $13B investment into OpenAI into context: its peas compared to their big bet: building out the global supercompute cloud.
You can bet that Google (Alphabet), Meta (Facebook), Apple (Apple), Amazon (AWS), and Tesla (Dojo) are not going to be left out to dry in this furious race. Google has had the infrastrusture for years — they just about singlehandedly wrote the book on how to build power- and cost-efficient data centers — and Amazon has led the world in cloud services with its hyper-profitable AWS compute-for-rent services.
Meta really only became a player recently, in a big way, with CEO Zuckerberg boldly announcing a $20 billion purchase order to nVidia to ship them 250,000 of their H100 GPUs during FY2024. There was some subtext that he did this half in the interest of Meta AI buildouts, and half simply to prevent competitors from getting the chips (thus Silicon Valley’s latest gossip term: “pity the GPU Poor”).
Is there precedent for a 7 Trillion AI bet?
In a word: NO.
For comparison:
The Manhattan Project (the 1945 nuclear bomb) cost: $25 billion
(0.025 trillion, adjusted for 2024 dollars),
invested across 6 years (1939-1945).
The Apollo Project (putting men on the moon) cost: $160 billion
(0.160 trillion, adjusted for 2024 dollars),
invested across 10 years (1961-1971).
So there you have it. The two largest, most ambitious projects ever completed in the modern era. Neither is 1/30th the scope of Altman’s 7 trillion AI dream. Okay… that’s government projects. How do capitalist projects compare, especially in this modern era? Well…
Billion dollar funding rounds are already the rarest of air.
This table lists the 10 largest funding rounds that have ever been accomplished in the last 100 years:
Get my drift?
Nothing comes even close.
Thus, we have to rule out the usual suspects for “pre-IPO funding” — VCs (Venture Capitalists), who generally have total fund balances ranging between $100 million and $2 billion, earmarked for specific industries (i.e.. e-commerce, AI, blockchain). dedicating, for instance, an entire fund to Altman’s folly (which never happens, FYI), would be… well 2B divided into 7000B = 0.03% of the ask.
Yes, you are reading that right. three-one-hundreths of one percent. You would need more than 3,000 two-billion dollar VC fund accounts to go all-in in order to come even close to fulfilling Altmans request. Ah ha.
So… Who is going to fund this?
Which leads us to ask: who exactly is Altman targeting with this 7 Trillion AI Ask?
Here are the only entities which I see even remotely capable of contributing:
- Global financial institutions
(BlackRock, J.P. Morgan Chase, Bank of America, etc) - Nation States
(top 10 wealthiest countries: see chart) - Sovereign Wealth Funds
(i.e. Saudi Princes) - global billionaires
i.e. high net worth individuals - global corporations
who would, btw, have to leverage their physical assets:
chip fabs, real estate, forward revenues… yikes.
That’s about it. Any lesser entities (VCs, friends and family) wouldn’t even make a dent / scratch on the surface of that massive figure.
Drilling down
A quick look at the largest fund reservoirs on planet earth, c. 2024:
Okay, so some — again, IF they devoted a substantial portion of their entire managed asset base — could feasibly contribute to this 7 Trillion AI Dream Bet.
How about corporations… specifically global megacorps?
Ha! So, Altman’s 7 trillion AI Ask is greater than the sum total market capitalizations of the largest three corporations in the entire world. So again: it is possible that these entities could substantively contribute to his vision… but, (and this is a massive caveat), they’d have to put their physical and cash assets in jeopardy as collateral (effectively the corporate equivalent of we normal humans taking out a second mortgage on our home). Prediction: not likely.
It’s also important to note ranks 6 (NVIDIA) and 10 (TSMC) on this list: both those entities are already primarily valued based on their AI chip sales. N-vidia designs them (well, actuall, nVidia’s AI software designs them), and TSMC (Taiwan Semiconductor Manufacturing Company) actually manufactures them.
Finally, lets put the 7 trillion AI dream into context.
How does a $7 Trillion AI compare?
(shout-out to @vensy for the first 4 comparables)
- The total estimated market value of all real-estate in New York City — both commercial and residential — is estimated to be $1.5 Trillion.
- The total cost of the US’s War in Afghanistan — spent across 20+ years — was $2.3 Trillion
- The total annual budget of the US Federal Government is $6 Trillion
- The US GDP (total annual economic output) is roughly $23 Trillion.
- The total GDP of all humans in the world (the total value of all work produced by every living human in year 2021) was $96 Trillion
- The sum total market cap of the largest 10,000 companies in the world is ~$100 Trillion.
So there you have it.
Think Sam deserves our trust? Because, as you can see: the real people who will foot this bill are the global citizens — you and me and our tax dollars — so, do you think its worth it?
.
engine: OpenAI DALL-E3
prompt: “make me an image of a nerd in cool tennis shoes with his hand out, holding a cardboard sign that reads “Will Build AGI for $7 trillion”. behind him sits a shiny green lamborghini.”