Is it sustainable?
“…the US AI companies remind me a lot of 1970s era Detroit with the Chinese companies today playing the role of Toyota, Honda and so on. At least Detroit had mostly depreciated the majority of its factories and the like and wasn’t in the middle of building massive new ones.”
Not clear to me what this implies for the viability of space data centers.
Good article, gives you lots to think about.
Are ai companies like an early days Amazon? Can they generate revenue to minimize burn rate without causing a death spiral? Can the cost of local ai compete with hosted ai?
Aside from the worries about the big ai companies, ai is useful and isn’t going anywhere, some companies will fail (which is to be expected), and how ai is adopted is an open question.
It’s a good article as far as it goes, but it does not, IMHO, go very far at all. It is entirely about LLM AI and use cases. Given that those are what generate the vast majority of the current aggregate AI revenue stream, this focus is perfectly understandable. But it completely leaves out pretty much everything else AI will be used for not too far down the road – things like Elon’s cars and robots and use cases like drug discovery, materials science, genetic modifications to organisms including humans, and many others.
For example, from a strictly “show me the money” standpoint, the use of AI to develop actual cures, as opposed to current Big Pharma palliatives, presents the probability that more and more of the trillions now spent on healthcare can go to AI companies rather than the healthcare establishment. The transfer of all of this money would be via health insurance underwriters. Even a very expensive one-time curative regime for, say, MS or ALS would be worth it to insurers who would, thereafter, not be looking at paying open-endedly for even more cumulatively expensive palliatives.
Still going by my call from June: collapse of the AI market by the end of the first year of the next presidential term.
As I see it, there is value in AI, but probably an order of magnitude shy of enough to cover the expectations of today. That’s bubble markets.
Frankly, I think the whole thing would collapse merely if we used common sense accounting like requiring the reporting of Enron-style hiding of liabilities on the primary company’s balance sheet or depreciation of computing assets on the same time scale as the assets’ lifespan (3-5 years). This isn’t the only book cooking going on. So much of the supposed value of all of these companies is based on a false picture using bad accounting.
Just on the accounting shenanigans alone, I would bet on a large bubble. One doesn’t need gimmicks like that with healthy, (near) profitable business.
And I will reiterate my previous case that the bubble-ish aspects of AI are pretty much all of the LLM slop – analogous to the Cambrian Explosion, then Die-Off, of thousands of narrow-gauge e-commerce outfits that constituted the Dot-Com Bust. E-commerce was a hugely useful technology, but people didn’t want a million little virtual storefronts, they wanted Amazon. AI will more than pay its way when it starts doing big and useful things. That’s not far off.
Dick, the problem is that after the explosion comes the cull. In Earth’s history, there were several times when all the big groups partially or completely died out. A market equivalent is just not that hard to come by, especially when you consider that bankruptcy court provides ways to shed shareholders efficiently.
It’s very possible that present shareholders may end up with little to no portion of that bigger future just because their shares were diluted or discharged in bankruptcy court for nothing. Among many other things, that hasn’t been priced into current market prices.
Could be, like the dot com bubble.
There is a lot of exuberance while there is also long term value to AI in all aspects of society.
I view it a bit like Amazon, who pumped all their money back into the company and the people who owned their stock were mocked. During that era, a lot of companies went down but here we are with the internet as the backbone of society.
Right now, a lot of the focus seems to be on LLM, image generation, and coding but there are some things like CAD where AI would be very useful but can’t be done well right now. It will take a lot of compute to train AI to work on certain problems but when it can, then the compute to do the work could be much much smaller.
A lot of companies will want control over the data and AI tailored to their specific use case. There is a market for open weight and local AI. It could be that the current big AI companies maintain their frontier models but also push out products that include hardware, software, and services but there will be lots of new companies that can do that last bit.
There is a race to build out compute. There will be some winners and losers there. There is also a race to provide consumers with what that compute produces/ed along with hardware and services. There will be winners and losers there too but the impact of any one company going under wont have the same impact on the economy as an Open AI going under.
In most cases, people don’t need the “best” ai. They just need one that is good enough for their application but on the backend, the big companies still need to push advancements and some of them will be in efficiency, meaning that large data centers or always expanding compute might not be as important.
It could be data centers are over built, or soon will be, that would leave a lot of excess chip capacity that could be used for other products that use AI and maybe the cost of using that excess capacity would come down.
Like other innovations, it probably isn’t going to be a straight line going up and moving to the right but over time, that line will be higher than it is today.