That's a ridiculous sum. Even if you divide it over a decade.
Global semiconductor sales are about 0.6 trillion annually. Semiconductor industry is projected to invest $0.5 trillion 2024 into chipmaking.
Gulf states are full of dumb money that can be extracted with nice Powerpoint and hype talk, but that's too much even for them.
---- edit ----
In talks with TSMC, Altman has said he wants to build dozens of chip-fabrication plants in the next few years, the people said. His vision would be to raise the money from Middle East investors and have TSMC build and run them. https://archive.is/tyZq2#selection-6231.180-6234.0
The whole arab league GDP (which is whole arab countries not only golf) is about 3.5T [1] which is a little bit over half of this amount. Too much is understatement.
Gulf states sovereign wealth funds have about $3.7T in total assets. They would have to sell 50% of the assets to get 25% share in Sam Altman's new project.
At some point, someone usually wants cash instead of shares in Neom (usually quickly), and you need to liquidate those shares for cash, affecting the valuation of said speculative assets.
Slightly less than related, but sovereign wealth funds can pay absolutely insane amounts of money for the talent that they want. FAANG is far from the end game for truly talented developers.
> FAANG is far from the end game for truly talented developers.
I’ve spent the last 18 months in UAE, pretty well networked into Oxford and Cambridge university alumni networks, and never heard any suggestion of this. The team behind Falcon in AD seem to pay pretty well, but not FAANG.
Even if the gulf states had the money, they still wouldn't do it. The main thing they care about is decreasing their risk of having their assets frozen by the U.S., which is why they are happy to make money-losing investments. Whereas investing this much into a single venture would massively increase their exposure.
> Feb 8 (Reuters) - OpenAI CEO Sam Altman is in talks with investors, including the UAE, to raise funds for a tech initiative to boost the world's chip-building capacity and expand its ability to power AI, among other things, the Wall Street Journal reported on Thursday.
> The project could require raising as much as $5 trillion to $7 trillion, the report added, citing people familiar with the matter.
I don't know why I'm reading Reuters report WJS report someone's words. Where is the actual source of this number? Did Sam Altman really say "$7T" or it's just someone's wild guess?
"People familiar with the matter" gave the story to the WSJ reporter, so likely people that were in some of the meetings. The WSJ story is more comprehensive:
$7T sounds like a lot of money but reading between the lines seems to be the projected total cost for not just chip design and fab but also infrastructure to support such an industry over a bit of time. That seems reasonable to me. It's not like you can easily spend $7T very quickly, so this total value would likely be spent over many years across many different companies/investments in order to grow the entire AI industry. It's hardware, software, facilities, engineers, scientists, power plants, power distribution networks, silicon fabs, etc.
> this total value would likely be spent over many years across many different companies/investments in order to grow the entire AI industry.
Sure, in total across many companies, ventures and years as part of an ecosystem. But one entity raising the entire sum? That seems ludicrously unrealistic.
It's effectively a Manhattan Project for fab capacity. The US spent $1.9b in 1941 dollars on the Manhattan project, which is about $400b today. On that scale, the 'trillions' number is only tenfold, which ... is still outlandish, and can't possibly be managed by a single private entity, but is order-of-magnitude-correct for "we're going to force the development of a civilizationally determinant technology by overwhelming application of capital"
this is insane. like, more than the market cap of nvidia, intel, and amd combined. And thats ignoring that the challenges of manufacturing are different from software
A lot of naysayers missing the point in this thread.
Is the most likely outcome that we shortly 10x the market for AI chips? Probably not. But, if OpenAI get AGI, or even just strong agents or expert tools, you can easily imagine the demand skyrocketing.
If you are a founder or VC, you are into low probability, high payoff bets. Think of this investment in those terms.
Let’s say Altman thinks there is a 20% chance of OpenAI getting AGI in 10 years. (Maybe you think it’s lower, I suspect he thinks it is higher) In the possible world where we do get AGI, the value that OpenAI can create will be hard-capped by chip production, so placing a complementary bet on chips means he can prevent this from limiting OpenAI’s growth in the happy case.
If this bet pays of 100x in 10% of worlds, it’s a great bet. From Altmans perspective the chip business could break even as long as it unlocks that 100x growth for OpenAI. He’s basically doubling down on his existing portfolio.
Another obvious issue in the imaginary AGI world is energy production; and indeed, Altman has big investments in nuclear too.
The first company that gets AGI will be the world's first $10T company. Work backwards from that.
(None of the above requires particularly high probabilities of actually achieving AGI to make sense as a bet.)
AI chips will be ~5% of TSMC's revenue in 2024. A very bullish outcome would be ~8% in 2025. TSMC spends ~$50b on capex per year. This is a completely ridiculous proposal.
(a) someone's source is off by a whole ,000 and the Journal ran with it
(b) OpenAI is going to assume nation-state levels debt become the most valuable and powerful company since the Dutch East India Company,
or (c) OpenAI is talking to investors and governments around the world about public-private investments in fabs, electricity generation, attendant cities, infrastructure, etc. that might require a total investment over X years and all parties involved of $N trillion dollars.
Occam says (a), but (c) seems far more likely than (b).
i think they would do b if they could, but c is more likely. they will have to hit checkpoints to guarantee revenue in order for all parties involves to continue.
Interesting sales tactic though. Just floating this absurd number out there is classic highball, perhaps will make an investment of “only” 100 billion feel more reasonable to Abu Dhabi or SoftBank.
Although I do know this is ridiculous for now, I love that we're entering the age of talking in terms of trillions for innovation and investments. Onwards to climbing the Kardashev Scale as a species!
Global semiconductor sales are about 0.6 trillion annually. Semiconductor industry is projected to invest $0.5 trillion 2024 into chipmaking.
Gulf states are full of dumb money that can be extracted with nice Powerpoint and hype talk, but that's too much even for them.
---- edit ----
In talks with TSMC, Altman has said he wants to build dozens of chip-fabrication plants in the next few years, the people said. His vision would be to raise the money from Middle East investors and have TSMC build and run them. https://archive.is/tyZq2#selection-6231.180-6234.0