My napkin-math approach to get a bird's eye perspective on the situation:
A $1T investment needs to produce on the order of $100B in yearly earnings to be a good investment.
Global GDP is about $100T.
So one way for things to work out for the AI companies would be if AI raises GDP by 1% and the AI companies capture 10% of the created value.
This is a good analyst report - lots of data. Conclusion - firms are spending ahead of sustained revenues right now, and a lot of the money is going offshore to TSMC, basically.
I’m not certain of the conclusion - I think a lot depends on amortization schedules - if data centers are fully booked right now, then we don’t need very long amortization schedules at the reported 60+% margin on inference to see this capex fully paid off.
My prior is that we are seeing something like 1/10,000th or so of the reasonable inference demand the world has fulfilled. There’s a note in the analysis that might back this - it says that we are seeing one of the only times ever where hardware prices are rising over time. Combined with spot prices at lambda labs (still quite high I’d say), it doesn’t look like we’re seeing a drop in inference demand.
Under those circumstances, the first phases of this bet, cross-industry, look like they will pay off. If that’s true, as an investment strategy, I’d just buy the basket - oAI, Anthropic, GOOG, META, SpaceX, MSFT, probably even Oracle, and wait. We’ll either get the rotating state of the art frontier capacity we’ve gotten in the last 18 months, or one of those will have lift off.
Of those, I think MSFT is the value play - they’re down something like 20% in the last six months? Satya’s strategy seems very sensible to me - slow hyperscale buildouts in the US (lots of competition) and do them everywhere else in the world (still not much competition). For countries that can’t build their own frontier models, the next best thing is going to be running them in local datacenters; MSFT has long standing operational bases everywhere in the world, it’s arguably one of their differentiators compared to GOOG/META.
If a different architecture to LLMs is invented (that could actually "think", that could potentially reach AGI), then perhaps it would be more efficient than LLMs. Perhaps LLMs can make themselves more efficient. They can't even remember "properly". Hallucinations cripple them for serious, professional uses. If they may hallucinate 5% of the time and you are asking mission critical queries, that's a problem.
Perhaps all of these data centers won't be needed. At least not by some of the current AI companies that won't keep up. If that happens to OpenAI, that would be quite a shock to the financial system (and GDP).
With the talk of Nvidia backtracking and saying they won't invest 100 billion in OpenAI, Oracle in a poor financial position with the loan's for it's upcoming data centres becoming more expensive and dubious (they could fail to pay them)- the picture isn't as positive as you make it out to be. Which makes me think that you have an ulterior motive.
Microsoft's changes to windows have alienated some of their userbase. Copilot is poor compared to it's rivals. There's a reason they are down 20%
A significant part of the capex is just energy, so even if there is some sort of AI black swan event and the data centers become obsolete overnight (unlikely), energy is literally the root of all bounty so it is good that something is incentivizing increased resource allocation in that area.
AI plans are not going to stay at $20/mo.
People will go to alternative models, but it likely will be as popular as Linux.
Yeah, this is something I am thinking a lot about. Companies won't be able to sustain this level of spending forever, and one of two things will need to happen:
1. Models become commodities and immensely cheaper to operate for inference as a result of some future innovation. This would presumably be very bad for the handful of companies who have invested that $1T and want to recoup that, but great for those of us who love cheap inference.
2. #1 doesn't happen and the model providers start begin to feel empowered to pass the true cost of training + inference down to the model consumer. We start paying thousands of dollars per month for model usage and the price gate blocks out most people from reaping the benefits of bleeding-edge AI, instead being locked into cheaper models that are just there to extract cash by selling them things.
Personally I'm leaning toward #1. Future models near as good as the absolute best will get far cheaper to train, and new techniques and specialized inference chips will make them much cheaper to use. It isn't hard for me to imagine another Deepseek moment in the not-so-distant future. Perhaps Anthropic is thinking the same thing given the rumors that they are rumored to be pushing toward an IPO as early as this year.
Back of the envelope calculations point to $60-$80/mo plans for 5-10y payback period.
This also fits with OpenAIs announced advertising cost, and is something most consumers can stomach.
That why they need widen the moat; it appears not giving us access to hardware might be that moat.
They desperately need LLMs to stay rentier and hardware advances are a direct attack on their model.
Yeah, they'll be free - on device and "good enough".
If you want the best, then pay.
Economics will be decisive force. Paying 1000USD a month for AI or buying server for 10kUSD, loading there Chinese AI model which can do 90% of what SOTA models can? Looks like a no brainer.
Man if China can catch up on the hardware front we could be seeing the 'TikTok' story repeat there. (They provide a better product>US govt panics> bans the US from the good stuff)
Possibly, but that assumes continuity. New math and algorithmic breakthroughs could make much of today’s AI stack legacy, reshuffling both costs and winners.
I don't know about you, but I benefit so much from using Claude at work that I would gladly pay $1,500-$2,000 per month to keep using it.
That is more than one month rent for most of the world. Most people are simply not going to pay this.
My rent is less than that. But if you add up salary, payroll taxes, benefits, social security etc my employer still spends around four times that amount on employing me. More if you include misc overheads associated with having one more employee. Personally I could never afford 1500-2000€/month for dev tooling, but my employer should rationally be willing to spend that for anything that makes me more than 25% more effective.
I'm not sure today's Claude Code could ask for that. But I don't think it would be a crazy goal for them to work towards
There have been many many productivity improvements over the last 50 years that provided more than a 25% boost. I’ve yet to see an employer pay that much per employee for any of them.
Also a 25% boost per individual doesn’t necessarily equal a 25% boost to the final output if there are other bottlenecks.
Well then I'm sorry but unfortunately they are going to be left behind.
People who are cut out to be software developers can afford the means of production.
Things that can only be used by an exclusive elite don't tend to survive, unless we're talking super-yachts.
AI is only going to work if enough people can actually meaningfully use it.
Therefore, the monetisation model will have to adapt in ways that make it sustainable. OpenAI is experimenting with ads. Other companies will just subsidise the living daylights out of their solutions...and a few people will indeed run this stuff locally.
Look at how slow the adoption of VR has been. And how badly Meta's gamble on the metaverse went. It's still too expensive for most people. Yes, a small elite can afford the necessary equipment, but that's not a petri dish on which one can grow a paradigm-shift.
If only a few thousand people could afford [insert any invention here], that invention wouldn't be common-place nowadays.
Now, the pyramid has sort of been turned on its head, in the sense that things nowadays don't start expensive and then become cheaper, but instead start cheap and then become...something else, be that more expensive or riddled with ads. But there are limits to this.
> People who are cut out to be software developers
You mean the people AI is going to replace? What's the definition of 'cut out to be' here?
The people who own "the means of production" isn't you.
A $2k/month model, should it ever arise, won't need you.
I haven't looked at a cost analysis recently, but it's possible that we basically already have $2k/month models, if they were priced to be even slightly profitable.
Sure they can also code without the help of a model, probably not that much slower.
Your identity as real software developer relies on the community's broad, inclusive definition of what it means to be one. Something you're failing to extend to others.
To be sitting that far out on a limb of software development while sawing at the branches of others is quite an interesting choice.
Pretty edgy response. I'd say trying to scale in price rather than in quantity is a bad business strategy for tech period, specially if you hope to become Google-sized like OpenAI and company want.
Are you OpenAI? If not, you don't afford the means of production. You're the sharecropper.
Big yikes bro.
At that cost I'd just buy some GPUs and run a local model though. Maybe a couple RTX 6000s.
That's about as much as my Framework Desktop cost (thankful that I bought it before all the supply craziness we're seeing across the industry). In the relatively small amount of time I've spent tinkering with it, I've used a local LLM to do some real tasks. It's not as powerful as Claude, but given the immaturity in the local LLM space—on both the hardware and software side—I think it has real potential.
Cloud services have a head-start for quite a few reasons, but I really think we could see local LLMs coming into their own over the next 3-5 years.
Same but I imagine once prices start rising the prices of GPUs that can run any decent local models will soar (again) as well. You and I wouldn’t be the only person with this idea right?
I mean, will it? I would expect that all those GPUs and servers will ends up somewhere. Look on old Xeon servers, it all ended up in China. Nobody sane will buy 1U serve home, but Chinese has recycled these servers by making X99 motherboards which takes RAMs and Xeon CPUs from these noise servers and turning into PCs.
I would expect that they could sell something like AI computers with lot of GPU power created from similar recycled GPU clusters ussed today.
Those won’t be sufficient to run SOTA/trillion parameter models
Distilled models are good enough.
And most tasks don't demand that.
I use my brain, it's free
Fitting response for an account called "clownpenis_fart".
The future is here and it's time to stop ignoring it.
Your analog 1x productivity is worthless in comparison to my AI backed 10x productivity.
10x productivity means you should have had time to build an your own programming language/OS/integrated dev environment or something equally impressive. Can you link to it?
Yuck.
https://news.ycombinator.com/newsguidelines.html
> Be kind. Don't be snarky. Converse curiously; don't cross-examine. Edit out swipes.
> Comments should get more thoughtful and substantive, not less, as a topic gets more divisive.
> When disagreeing, please reply to the argument instead of calling names. "That is idiotic; 1 + 1 is 2, not 3" can be shortened to "1 + 1 is 2, not 3."
Honestly, if you just made your profile a day ago to yell overly confident and meaningless statements into the void, like a Mandrill in the jungle trying to shout over all the others, go back to LinkedIn, they like that kind of stuff there.
I even agree that AI has a place in our world and can greatly increase productivity. But we should talk about the how and why, instead of attacking others ad hominem and just stopping any discourse with absolutist nonsense.
And this gamble is paid for by American taxpayers, increased cost of utilities, and multibillion dollar corporations receiving tax breaks/subsidies from the cities/counties they build in.
This country is so awful. Great if you are rich. Awful if you are not in this top 0.01-1%.
A massive $79T has been transferred from bottom 90% to top 1% since the 1970s. [1]
I love how they say with a straight face that when AI takes over they will finally share all the fruits of capital with us.
Y'all gotta stop looking at politics this way.
You know why they don't share the fruits of capital with us now? Because Americans hate getting taxed to pay for welfare, and so they've been voting against taxes for 50 years. This whole political landscape changes when people lose their jobs to AI, a thing that everyone thinks should be taxed. In fact, the entire ideological underpinning behind extreme wealth accumulation is gone when AI runs everything.
Saw a video summarising this on gamersnexus today and it is nauseating - especially Jensen.
The French had a tool for this problem.
And look how it has worked out for them.
Not sure what you're implying, but I'd say their society is doing fairly well.
Only because the U.S. and U.K. conspired against them. The French did everything they could to keep the fire burning, by hosting people from various countries to teach them about revolution. Organizing globally against the rich parasites was hard and expensive back then. Now the only hold back, is that the rich parasites own most of the internet.
But WE BUILT IT, and can take back the internet when we finally realize it's not dems vs reps, but rich vs poor. It's always been a class war, they just are much better at keeping us distracted.
I think we need reforms and I’m very much against the accumulation of power that we’ve allowed the billionaire class.
But the French Revolution is nothing to emulate. If you’ve read the history of the French Revolution you know that it quickly moved on from rich parasites to murdering and imprisoning people over minor philosophical differences and real or lack of perceived lack of enthusiasm for continued murder. And it eventually led to global war and attempted global conquest.
Yes, guns, clubs, fire, and steel weapons. And afterward they had the Reign of Terror, and the rise of the French Emperor Napoleon. It seems like it mostly worked out in the long run, though subsequent World Wars left the French Empire as a weakened shell of itself. In the short term, up until Napoleon was finally taken down by the combined British and Prussian forces at Waterloo, it seemed to have led to all sorts of calamities. How many died? How many did Robespierre manage to get sentenced to death before he met the same fate? Would Napoleon have risen and caused the death of so many?
One thing would-be revolutionaries don't appreciate is that, well, similar to Mr. Putin's experience today, revolutions (and wars) are much easier to start than to control. One day you're chopping off the leader's head, the next day you are pressed into military service and your Constitution is gone. I personally would rather be patient and work on reforming institutions, even if it takes a much longer time. Often times when we get rid of them, it's not that something better fills the void, as anarchists (communists or libertarians alike) like to claim, but instead it's nothing and that capability is gone until some calamity restores the need.
Exactly this. Violent revolutions are very rarely successful in increasing the average welfare and freedom of the populace.
Good luck using guillotines on an army of militarized drones outnumbering you 10 to 1.
To be intellectually honest about it, you have to answer a bunch of questions:
1. Awful compared to what? 2. Was there an equivalent transfer outside America? 3. What is the cause? What ratio rent-seeking/shady activity vs a consequence of natural forces (e.g. technological change)
If it’s any consolation, I’m rich yet the country is still shit. (Comparing to Europe as a previous immigrant of Western Europe.
Other than a few parasitic industries it's pretty great. If we can just get some common sense reforms in insurance, healthcare, advertising, and reverse some regulatory capture it would be comparably utopic.
Not to mention all the land being gobbled up to build these data centers.
Of all the externalities under discussion, I think land use is a very minor one.
Most of this land was low-utility anyway. You should realise it is good for the land owners to convert it to high yield output, which in turn the government can tax and return some of the gains to the people.
What's the alternative ?
> A massive $79T has been transferred from bottom 90% to top 1% since the 1970s
This assertion is based on comparing reality with a counterfactual where income distributions remained static from 1975 to the present. Real median personal income roughly doubled over this time period.
The use of the word "transferred" seems a little intellectually dishonest here. The use of the counterfactual seems to suggest that income distribution has no relationship with growth in total income, and total income would have been exactly the same regardless of income distribution. I see no reason to assume that to be the case.
Well you have a data point of one, so I guess we live in the best of all possible outcomes?
I don't understand what you mean by "data point of one"
Do you think I'm talking about my own, personal income?
I'm talking about mean personal income in the United States, because the figures I found for household income only go back to 1985