> Why would Anthropic pay through the nose and make such desperate moves for compute?
That's pretty consistent with them having not been able to get the physical data centre capacity they have supposedly had built actually online, I think?
More to the point, Anthropic appears to mostly be capable of serving its customers, doesn't it? It's not so resource-starved that people aren't able to use it at all.
So what does it tell you that only 300 megawatts was enough to make a very meaningful difference to that?
If Anthropic's fortunes can be meaningfully changed by only 300MW which allowed them to uncap their customers, what is the case — even including OpenAI and Google -- for the demand for the 23GW of unbuilt AI data centre build commitments out there?
Satya Nadella himself said there will be some oversupply.
What if there is actually lots of oversupply?
My instinct is that the internet/software/IT industry sees its own demand for Anthropic and OpenAI and is misleading itself into thinking that this is representative of customer demand in the rest of the industry, let alone others, when what possible other human industry tasks are there that match the compute demands of writing software?
Another way to look at it: if the general demand for AI compute is so insatiable, how in this world of existing insatiable demand did xAI have spare capacity to lease? Why does Meta have spare capacity to lease?
The issue here is very evidently one of allocation. But allocation tends to get solved over time. That could leave quite a lot of very big data centres relatively empty.
If you get a little into grounded scifi the buildout seems more realistic. Robotics are making great strides while this buildout is happening. Imagine the implications of a human shaped robot? Now it isn't just knowledge work but literally every job on earth that is on the table. Now you can't get enough of that compute as you chew through obsoleting 8 billion humans from their current levels of participation in the economy.
I mean I do think there’s some likely extra demand, though I think it’s likely some demand will fall away from data centres to the desktop.
(And what you imply is also true, isn’t it - OpenRouter and equivalents will address this demand while also resolving some underlying underutilisation)
But the claims of awful shortages feel wrong when there’s little evidence of radical undersupply for the two big firms who are offering heavily subsidised plans.
It all seems pinned on the idea that some compute-guzzling application users clearly don’t have right now will emerge.
Due to the difficulty of building and the time lead, there will be a multiple of applications for every data center that actually comes to fruition.
But if you're talking about the capital investment that's being talked about as too high, I would agree that it's way too high, as all these big tech breakthroughs result in too much building at first. Which hurts the builder's profitability but helps consumers.
When our spot instances go away and don't come back, that's a sign that there's not enough capacity.
There is more demand for compute than there is supply, and that demand is going up.
Look at all of the things that have been and are being automated and realize that we're year 6 into a hundred year journey.
Look around you at all the things that aren't automated. They will be.
The world you grew up in won't be here for much longer. We're going through a transformation even more dramatic than the industrial revolution. There were no airplanes and then we landed on the moon.
Edit: downvotes as expected, but I'm shocked that a bunch of smart people at HN aren't taking the bet where this leads to a complete upending of the status quo. The holdout crowd needs to dream bigger.
The models are solving Millennium Prize problems, finding novel biological pathways our scientists can't see, putting every single software engineer out of the job of "coding by hand", and doing the same to designers, marketers, filmmakers, and 3D artists who used to have to operate pixel by pixel / photon by photon.
That's the usual bait and switch though. Very few people are saying AI the statistical software is going away. What people are dubious of is the commercialisation model of building a butt ton of data centres to serve massive general models by two heavily subsidised VC companies even as industry gets more mature and fickle as they get their heads around using it to solve problems and it goes onto Opex budgets. I think this is more like the early days of cloud which finished in "software built around clouds" rather than "everyone spins up their own fleet of servers for everything". Obviously the current investment levels need everyone to run their own fleet of Ralph looping agents on frontier models so that's what's marketed as the smart thing to do.
A lot of government appendages are still waiting for the dust to settle on this new fangled cloud computing thing in fairness. I don't think it's some deep masterplan
There's a huge gap between what's being described in that comment and an "extinction level event" is that's what ELE means.
What's being talked about is a massive extension of human capability, into the intellectual sphere rather than the physical environment. The atom bomb had a clear way to get to extinction, a sudden chain reaction that leads to collapse of the entire biome.
The only way AI gets there is something like control of the nuclear weapons on all sides and that control becoming suicidal. Every other major event leads to at most civilization collapse, after which I assume we will have some sort of Butlerian Jihad restrictions on getting into the same situation.
But, I would love to know what I'm missing here, if ELE has clearer routes!!
> The only way AI gets there is something like control of the nuclear weapons on all sides and that control becoming suicidal.
Not the way I was thinking off; I was thinking more along the lines of brain atrophy until humans are helpless (maybe 2 complete generations of humans).
And that's a short extinction; the KT extinction took place over thousands of years. If, 1000 years from now, we are finally at the place described in "The Machine Stops", only then will an observer be able to decide that we don't have the necessary critical mass of intelligent humans anymore.
Humans got to be the apex predator simply on intelligence alone. It's possible that without thinking ability we'd go extinct.
> Not the way I was thinking off; I was thinking more along the lines of brain atrophy until humans are helpless (maybe 2 complete generations of humans).
That's not how humans work. We developed math, art, literature, etc., because people love the act of doing it. We will continue to do that.
We're not going to stop thinking because we don't have to. We stop thinking because we subject ourselves to political propaganda, or religious belief systems, or other group control methods that turns off critical thought. But even then, it's only a subset of the population, it won't get us all.
> That's not how humans work. We developed math, art, literature, etc., because people love the act of doing it. We will continue to do that.
Sure, for art, literature, etc. For math, physics, biosciences, etc, we pay people to do it. There are too few people who would work for free to push progress in those areas. We are most certainly not going to do that, we're going to leave it to the AIs and hope for the best.
This sort of thing (human brains atrophying over generations) is like something out of a dystopian nightmare[1], but hey, that's natural selection for you: if intelligence is not a selection criteria (and it won't be if machines do all the thinking), it's unlikely that it will be present in future generations.
-------------------
[1]. "The Machine Stops", E. M. Forster, 1907. Or "Pump 6" (can't remember the author, Italian author). Just because it's fiction does not mean it's impossible.
I actually think the opposite… it takes all my skills as an engineer to steer these things effectively and I’ve learnt more than I imagined I could in a very short period of time. And I don’t even use AI as heavily these days due to other commitments. This also aligns with how heavy AI users report feeling drained after intense coding sessions.
My take is that after a period of turmoil, most humans will be doing higher level cognitive work than we’re doing now, because all the lower level work will be done by AI agents, and so progress will accelerate multifold.
On HN there are a lot of people who truly believe that this whole thing is hype and the bubble will burst and things will go back to how it was before.
The real answer to “how many data centers do we need” is hard. Anyone giving confident answers to this should be suspicious.
The easy type of answer you get from those kind of people will always be “we are building too much”. They’d have said that 3 years back and have no shame in admitting it.
The thing is, you can overestimate or underestimate. Both have cons. The people sincerely believing this is all hype will always underestimate because all their epistemic muscle haven’t been used.
To be charitable to your hypothetical HN naysayers, isn't the question more like "how many more data centers do we need right now?" as well as "why is the answer a shitton for all these guys?". It doesn't really feel like this is a debate about speculation or the quantitative "too much" of it, but more of the "why at all?"
It just seems like the AI guys could easily win this debate if they could just articulate the why of it beyond just the near-parodic "everything will be AI in the future!". Like why not be a positive articulator and defender of what the extreme influx of datacenters will bring to the people? Don't you think that would be at least a stronger argument to make here? Why police the ideology or scare people with your position when you could, I don't know, win their hearts and minds?
Isn't that why you (presumably) like AI? Because it brings good things to people?
The people who say it is a bubble say "Once it pops, we will get major recession and everyone will suffer. We are already im fascist -autoritarian phase and economic hardship will make it worst."
> They’d have said that 3 years back and have no shame in admitting it.
3 years back, we had strong push for absolutely unprepared AI into areas where it sux. All the while onslaught of lies about its capabilities and hype.
In the meantime we had "artists are all obsolete suckers hahaha" and "humans are obsolete" and "AI will raise your kids so that you dont have to interact with them".
And we had "we the AI company will destroy us" as if it was a good thing.
55mpg is nothing. I get 100+ mpg equivalent on my massive F-150 lightning and if you include cost in the comparison it’s probably a couple of multipliers more than that.
And that's just operating cost. PHEVs have a higher maintenance cost and worse reliability than pure EVs, typically. Certainly higher complexity and more parts.
On top of that, my Lightning was less expensive than the gas equivalent. I've been winning since day one. If I had to sell it today (that would make me so sad), the depreciation would be the lowest of any car I've ever owned.
You really do have to do the math using your local costs. I'm in a place with slightly cheaper gas but much cheaper electricity than California. Combined with cheap night time electricity I'm paying under $0.03 per mile for a large EV. I would need 150 mpg vehicle (assuming $4.50 gas) to beat that. However I'm cheating a bit - I'm on a plan which raises my day time electric costs a small amount in exchange for half-price at night, so they're getting some of that back, and I should amortize charger and electrician work ($1600 for me).
What's the math on that? Average kWh in California is around 18 cents. Assuming 0.35 kWh per mile, that's 6.3 cents per mile. You'd need to get 100mpg+ to beat an EV at current gas prices. A brand new Corolla hybrid my family recently bought gets around 40-60 mpg driving around a small city. And that's without factoring in maintenance differences costs per mile.
It wouldn't require perfect accuracy, just rough accuracy and a human to confirm, and it's already more than good enough for that. I do not understand this confusion surrounding modern math.
reply