This is just opinion, could be wrong, discussion is encouraged.
I think the way Anthropic and OpenAI are going, needing to rely ever more on hype and rushing towards IPO before the general populace understands the rouse that’s being played, I can see how they will play out in the years to come. Not mentioning the heat from China.
I thought until recently they would eventually die because the economic reality would simply catch up to them, but thinking of how Apple TV+ works to produce above average television shows at a massive loss to Apple, I think that this perhaps the only viable route ahead. These big giants might just buy up some of these LLM farms, once their real value becomes apparent in the market (before or after IPO) and be incorporated as a proprietary technology for that company. Apple could buy Anthropic for a few billion to provide agentic coding and significantly lower the barrier of entry to new programmers to develop apps for their platforms. The entire business idea by itself is a losing business, but integrated horizontally across their other business arms it could make some sense.
Similar things could happen between other players. What’s very clear is that LLM as a core technology has plateaued and to gain 1% more accuracy or performance at this point is beginning to require an order of magnitude increase in raw compute.
Curious what others think on how this road for these chatbot corps looks 2-3 years down the track.
A thing that gets overlooked often is that being the first in a field is often quite bad for business. Apple learned that lesson. They didn't make the first smartphone, tablet, or mp3 player. They waited for other people to be first and then iterated well on that.
The people who financed laying the train tracks that became the economic arteries of North America (along with robbing more land and pushing out the people who already lived there) didn't become rich. The second generation that ran logistics and passengers was way more likely to become rich. They didn't have to pay for the infrastructure but they benefited from it being there.
OpenAI, Anthropic, Meta, the son of the South African emerald miners, Google, etc. They're just laying tracks right now. The conglomerates that don't rely only on so-called AI have a chance to survive. All the other ones are going to run out of money and/or get bought and taken over. The real winner of the industry probably hasn't even made it to the starting line.
Training models takes a big investment and then becomes difficult to protect against people who will just try to copy and steal it. In a way, I think world history will look at the US as the first generation track layers. You built the data centers at ridiculous cost, both in hardware and negative consequences for people and the environment, and then get out-iterated by all the people who didn't want to jump on to the hype train three years ago or couldn't foot the bill at the time. The cost decreases the longer you wait. US companies think they are investing into future long-tern licensing agreements to use their yet unknown products. I think that's dumb and isn't going to happen.
I think Apple will benefit from their lean back and wait approach. They will look at where this stuff consolidates into something useful and then just buy it.
Edit: Forgot this point: most of the train tracks have become redundant in the age of airplanes. The infrastructure is decaying. Data centers may become overgrown train tracks faster than we expect.
The infrastructure is the number one issue here. Everyone focuses on Anthropic and OpenAI because they generate headlines - and don't get me wrong, both of those companies are built on sand - but the true rot in the industry is in the data centres.
The economics just do not work. You buy GPUs for billions and billions of dollars. Those GPUs are outdated by the time you install them, and then they burn out within 5 years. You pay ungodly amounts for electricity. And you make, well... Not nearly enough. There's literally no way to sell compute on GPUs at or above cost. None. The math doesn't math. The data centre CEOs know this, but they're all convinced that everyone else knows something they don't.
The promise for the data centres, the only thing that offered a glimmer of future revenue, is that OpenAI and Anthropic (the only companies in the world actually buying compute at scale right now) were going to continue to increase their compute spending basically infinitely. As those two companies scale back their spending (because they're running out of money) it's going to become increasingly apparent that there is no one else rushing to fill the gaps. Microsoft, Meta and SpaceX all supply their own compute and have more than enough for the absolutely trivial use their AI products see. The idea was that this would be a new internet; you build the infrastructure and an explosion of new companies flood in to use it. Be the guys selling picks and shovels. But that explosion of new companies didn't happen, and never will.
Probably 95% or more of new data centre projects will be abandoned without ever turning the lights on. Most without ever even breaking ground.
I agree with you. I still think somebody is going to do something with the compute capacity that hasn't burned up in cycles to train models to create Yoda with a six-fingered hand. And somebody will find something worth doing with these models. It probably won't be search, or writing, or making graphics for your PowerPoint. But I think video creation, be it for Hollywood or video games, will be here to stay and be potentially profitable. There will be a use in coding although probably no profit. I'm still hoping for good uses in medicine and pharmacology as well. And as aids to people with disabilities there are probably very good use cases as well. And they are the richest people on this planet so there is a Ferengi level of profit there. /s, of course.
The digitization of books will be useful as well.
In the end, the ROI for so-called AI will be inked in red. And the US will enter a recession, which will give us MAGA 3.0. So invasion of Greenland ca. 2034?
Socialism strengthens from economic hardship just as much as fascism does. Just look at the surging popularity of the DSA in America.
It's going to get harder and harder to make the case for MAGA to voters when they already had their turn at the wheel and only made things worse. Meanwhile Mamdani continues to get more popular every day because he's actually doing things that help people.
Hopelessness only helps the oppressors in the end.
Let me say up front that I'm very much in favor of democratic socialism. When I now try to disabuse you of two points in your comment, it's not for ideological reasons. It's more of a gut feeling about how politics, and especially politics in the US works.
Mandani works in NYC. A politician like that wouldn't work in Wisconsin or Louisiana. Probably not even in California. That is not to say people wouldn't benefit from more left leaning politics in those places. But they wouldn't get elected into high enough offices in sufficient numbers to drive change. There is the US hangover from the cold war that reflexively makes anything with socialism a turnoff for the poorly educated masses (the irony!). And the power of religious belief, especially monotheistic religions, favors right wing authoritarians. The Midwest is too religious for its own good. NYC is the outlier.
MAGAs don't vote on track record. It's vibe based and that vibe is obnoxiously trying to own their opponents. The fact that 47 is ruining the US"s standing in the world, the economy, and all unwritten political norms that had underpinned the republic are not as important as fucking with the libs. 47 may be at the end of his political career, somebody is going to take over and continue on this path. You can't fight feelings with facts. And you can't drive change without getting enough socialists into office everywhere.
There are plenty of examples of politicians like Mamdani winning all over the USA. The right message will resonate with voters. It's just about figuring out how to speak to their needs. People are struggling and they're looking for answers.
As for MAGA, I genuinely mean it when I say they don't matter. MAGA are not the majority of the electorate. They do not win elections. Yes, they'll always vote for Trump or someone like him, but Trump never won with only MAGA. They need to find a way to appeal to people outside of their base in order to win. In 2024 that was because their message of affordability resonated with voters who spent four years under Biden drowning in cost of living.
You're writing the fascists' victory speeches for them. Stop telling yourself why they're strong and start looking for where they're weak.
Very fine points on being the smart follower, yet many a European would take issue with your US centric viewpoint on rail :) Even in the US cargo would come to a standstill without rail.
I think Apple in particular has a high chance of coming out on top once the dust settles. They can’t lose in a fight they didn’t partake in, and they happen to be building the perfect hardware for small scale models.
That's because virtually nobody got rich off of trains in Europe. They would have all been reasonably regulated and their incomes reasonably taxed. And rail was used to connect places in Europe that already existed and often already had roads connecting them. There was no "wilderness to conquer." That's what sets the US railroad apart from their old world counterparts. And as far as cargo is concerned, a lot of it may run on rails. But that's because it's already there. They wouldn't build a new network today. And the infrastructure that survived this far isn't exactly in good shape.