> Because everyone and their mother are labeling regular data centers as AI data centers.
Companies (generally) build things with an expectation for a return on their investment: what "regular" data centre usage would necessitate these kind of build-outs?
To sell more Postgres or WordPress VMs/instances? Is that being used to justify the spending in shareholder conference calls and regulatory filings?
I just wish we forced every new data center to be built with renewables or something. The marginal cost over a conventional data center can’t be that big compared to the total cost, and these companies can afford it. Maybe it can help advance the next generation of small modular nuclear reactors or something.
No need. Renewables are already the cheapest way to power them. Companies like X using gas generators aren't doing that because it's cheaper but because it's faster. They need power now, not in five years. They will in the end replace those with something cleaner as a cost optimization. But they just don't want to wait for that.
Data centers and energy infrastructure investments go hand in hand. A lot of that money is actually being allocated for energy infrastructure. Because a data center is useless if you can't power it. That is also why companies like Amazon and MS are investing in nuclear. They need large amounts of cheap power to lower their energy bills long term and they need to secure access to energy supply. They see nuclear as a way of getting that. Coal and gas is just too expensive and undesirable. There's not enough of it and it's expensive to operate. Renewables & clean power (including nuclear here) is what they want really.
What would actually help is making the permitting around renewables and nuclear faster and more efficient. The energy demand is there. And that includes from data centers. And the capital is there. It's just that energy projects are bottle-necked on bureaucracy. Many countries have loads of viable energy projects stuck in their planning pipelines. For example, it takes years to get a grid connection for new wind or solar plants that can otherwise be built and delivered in less than a year. And likewise it takes years to get construction projects approved. This also affects investments in grid infrastructure. That's nuts. These countries have energy shortages that are slowing down their economic growth. And the capital to fix that. It's just that they are blocked on their own bureaucracy.
Unlike data centers, investments in energy infrastructure provide decades of return of investment. These are long term investments. Even if AI flops (which I think is hard to argue given how useful it is already), we get to keep the energy infrastructure. And we'll find a way to repurpose those data centers. I don't see that as a write off either.
I don't see how you can make the argument that a large portion of funds used for AI capex were diverted from other investments (and starving other industries), while simultaneously applying the economic multiplier to the whole sum when going from the investments to the GDP impact.
Surely you only get one of the two, because for diverted investments the multiplier applies equally on both sides of the equation.
I don't know... 1.2% of GDP just doesn't seem that extreme to me. Certainly nowhere near "eating the economy" level compared to other transformative technologies or programs like:
More than a decade long. The technology and industry here was broadly shared. They did things like highjacked bra manufacturers to make space suits.
> Railroads: 6% (mentioned by the author)
We're still using this investment today.
> Covid stimulus: 27%
The virus that was killing us the fizzled is probably not the best example... Only arguments will ensue if I even attempt to point thing out in this one.
> WW2 defense: 40%
I mean Russia made its last lend lease payment in 2006. It lead to America dominance of the globe. It looks like an investment that paid it self off.
How much of the hardware spend on AI is going to be usable in 5years?
There are some deep fundamental questions one should be asking if they pay attention to the hardware space. Who is going to solve the power density problem? Does their solution mean we're moving to exotic cooling (hint: yes)? Have we hit another Moores law style wall (IPC is flat and we dont have a lot of growth left in clock, back to that pesky power and cooling problem). If a lot of it is usable in 5 years thats great, but then the industry isnt going to get any help from the hardware side and thats a bad omen for "scaling".
Meanwhile capex does not include power, data, constables or people. It may include training, but we know that can't be amortized. (how long does a trained system last before you need another, or before you need a continuation/update).
Everyone is going after AI under the assumption that they can market capture, or build some sort of moat, or... The problem is that no one has come up with the killer app where the tech will pay for itself. And many in the industry are smart enough not to try to build their product on someone else's platform (cause rug pulls are a thing).
"AI" could go the way of 3d tv's, VR, metaverse, where the hype never meets up with hope. That doesn't mean we wont get a bunch of great tooling out of it. It is going to need less academics and more engineering (and for that hardware costs have to drop...)
Given the ratio of software to hardware I think it's pretty significant. Apollo put 12 people on the moon, once. AI is used by hundreds of millions every day.
I'm waiting for the shoe to drop when someone comes out with an FPGA optimized for reconfigurable computing and lowers the cost of llm compute by 90% or better.
The main argument builds on the assumption that the economy is a zero sum game when it clearly is not. Just because we invest these ressources in AI does not mean we could mobilize the same capital for other pursuits.
Precisely AI is being built out today because the value returned is expected to be massive. I would argue this value will be far bigger than railroads ever could be.
Overspending will happen, for sure, in certain geographies or for specialty hardware, maybe even capacity will outpace demand for a while, but I don’t think the author makes a good case that we are there yet.
> Just because we invest these ressources in AI does not mean we could mobilize the same capital for other pursuits.
No offense but you sound naive here. This is exactly how dry powder manifests in PE/VC and is even predictable under cash rich corporate regines via M&A.
Trends like this are ripe for acceleration under favorable environmental conditions like high interest rates. Not to mention, a lot of it develops out of peer pressure.
Sometimes when your job is to deploy capital, you just deploy it. Of course you try to put it in the best possible places. But when those are few, well yeah... This happens.
The premise of AI and certainly what a large subset of executives and investors believe is that AI will provide a significant productivity increase to a significant part of the work force.
Of 30% of the work is done 10% faster that leaves a 3% gain for other economic activities. If that is true the CapEx is justified.
The counterintuitive part of automation is that it removes parts of the economy rather than making the economy bigger. You end up with more goods but the value people assign to them goes down as they don't provide additional social advantage.
For example at one point nails were 0.5% of the economy and today owning a nail factory is a low margin business that has no social status.
Similarly the percentage of the economy and social status associated with frontend software dev will get automated and become a smaller percentage of the economy.
Since social status is a zero sum game people increase spending in other areas where social status can be helped.
That is why I stated transistor improvements what was previously known as Moore's law will continue for at least another 10 years. The Smartphone has carried us from 2008 - 2023. The money that is being used today are already invested into the next 2 - 3 years of Semi Conductor manufacturing. That is 2nm or A20 this year and A18 / 14 in two years time. There is enough momentum towards A10 and A8 by 2030, 2032. Even if things slows down by then it is enough to run till 2035 unless something catastrophic like WW3 or Market collapse happening.
That said even if we somehow reach A5 in 2035, we are only at about 12x density increase. If we include system packaging, chiplet, interconnect advancement pushing this to 30 to 40x. This is still a far cry from the 1000 to 10000x compute demands from a lot of AI companies. And that is assuming memory bandwidth could scale with it.
Railroads lead to a distribution of capital into the society and to a long term increase of wealth for many.
Ai leads to a capital concentration on the hands of those that already have money and might lead to a long term reduction of wealth for the middle class.
Less purchasing power in the population usually is not good for economic development, so I have my doubts with respect to a boom .
The opening quote from Xi seems to come from an exaggerated report from FT. The original Mandarin report [^1] has a milder tone. AI and EV were only mentioned in a passing comment, not the focus of the meeting or the report.
Feel free to skip the next section about the culture-aware interpretation of that quote. Xi's warning is more about the so-called "political tournament". Usually the central government sets some industrial policy goal (AI and EV now, but it could be other things like real estate, chips, or drones from before). The local government would collude with companies to start "projects". The goal is to get brownies points for the former and subsidy package for the latter. Of course, nothing usually comes out of these "projects". Most just stops at the factory building phase. Such practice has always been a real headache to the central government. Xi is warning about it showing up in the AI and EV industry now.
"How extreme is AI datacenter spending? It is to the point that even Chinese President Xi Jinping is warning about it. With more than 250 datacenters under construction in his country, this week he cautioned against the spending rush, threatening to intervene:"
This reminds me of the .com bubble when bandwidth, especially transatlantic bandwidth, was dire but then all of a sudden it wasn't and we had huge overcapacities for years.
40 comments
[ 4.1 ms ] story [ 74.5 ms ] threadCompanies (generally) build things with an expectation for a return on their investment: what "regular" data centre usage would necessitate these kind of build-outs?
To sell more Postgres or WordPress VMs/instances? Is that being used to justify the spending in shareholder conference calls and regulatory filings?
Data centers and energy infrastructure investments go hand in hand. A lot of that money is actually being allocated for energy infrastructure. Because a data center is useless if you can't power it. That is also why companies like Amazon and MS are investing in nuclear. They need large amounts of cheap power to lower their energy bills long term and they need to secure access to energy supply. They see nuclear as a way of getting that. Coal and gas is just too expensive and undesirable. There's not enough of it and it's expensive to operate. Renewables & clean power (including nuclear here) is what they want really.
What would actually help is making the permitting around renewables and nuclear faster and more efficient. The energy demand is there. And that includes from data centers. And the capital is there. It's just that energy projects are bottle-necked on bureaucracy. Many countries have loads of viable energy projects stuck in their planning pipelines. For example, it takes years to get a grid connection for new wind or solar plants that can otherwise be built and delivered in less than a year. And likewise it takes years to get construction projects approved. This also affects investments in grid infrastructure. That's nuts. These countries have energy shortages that are slowing down their economic growth. And the capital to fix that. It's just that they are blocked on their own bureaucracy.
Unlike data centers, investments in energy infrastructure provide decades of return of investment. These are long term investments. Even if AI flops (which I think is hard to argue given how useful it is already), we get to keep the energy infrastructure. And we'll find a way to repurpose those data centers. I don't see that as a write off either.
Using sus statistics to draw weird conclusions.
Surely you only get one of the two, because for diverted investments the multiplier applies equally on both sides of the equation.
But I don't hear anyone worried about the massive power consumption without a clear indication if this is a net positive for our society.
The 1880s 6% on railroads is an interesting number, I didn't know it was that much.
- Apollo program: 4%
- Railroads: 6% (mentioned by the author)
- Covid stimulus: 27%
- WW2 defense: 40%
More than a decade long. The technology and industry here was broadly shared. They did things like highjacked bra manufacturers to make space suits.
> Railroads: 6% (mentioned by the author)
We're still using this investment today.
> Covid stimulus: 27%
The virus that was killing us the fizzled is probably not the best example... Only arguments will ensue if I even attempt to point thing out in this one.
> WW2 defense: 40%
I mean Russia made its last lend lease payment in 2006. It lead to America dominance of the globe. It looks like an investment that paid it self off.
How much of the hardware spend on AI is going to be usable in 5years?
There are some deep fundamental questions one should be asking if they pay attention to the hardware space. Who is going to solve the power density problem? Does their solution mean we're moving to exotic cooling (hint: yes)? Have we hit another Moores law style wall (IPC is flat and we dont have a lot of growth left in clock, back to that pesky power and cooling problem). If a lot of it is usable in 5 years thats great, but then the industry isnt going to get any help from the hardware side and thats a bad omen for "scaling".
Meanwhile capex does not include power, data, constables or people. It may include training, but we know that can't be amortized. (how long does a trained system last before you need another, or before you need a continuation/update).
Everyone is going after AI under the assumption that they can market capture, or build some sort of moat, or... The problem is that no one has come up with the killer app where the tech will pay for itself. And many in the industry are smart enough not to try to build their product on someone else's platform (cause rug pulls are a thing).
"AI" could go the way of 3d tv's, VR, metaverse, where the hype never meets up with hope. That doesn't mean we wont get a bunch of great tooling out of it. It is going to need less academics and more engineering (and for that hardware costs have to drop...)
Do you think WW3 defense should be on the charts yet?
That’s an interesting perspective. It does feel a bit like we’re setting money on fire.
Precisely AI is being built out today because the value returned is expected to be massive. I would argue this value will be far bigger than railroads ever could be.
Overspending will happen, for sure, in certain geographies or for specialty hardware, maybe even capacity will outpace demand for a while, but I don’t think the author makes a good case that we are there yet.
No offense but you sound naive here. This is exactly how dry powder manifests in PE/VC and is even predictable under cash rich corporate regines via M&A.
Trends like this are ripe for acceleration under favorable environmental conditions like high interest rates. Not to mention, a lot of it develops out of peer pressure.
Sometimes when your job is to deploy capital, you just deploy it. Of course you try to put it in the best possible places. But when those are few, well yeah... This happens.
Of 30% of the work is done 10% faster that leaves a 3% gain for other economic activities. If that is true the CapEx is justified.
For example at one point nails were 0.5% of the economy and today owning a nail factory is a low margin business that has no social status.
Similarly the percentage of the economy and social status associated with frontend software dev will get automated and become a smaller percentage of the economy.
Since social status is a zero sum game people increase spending in other areas where social status can be helped.
That said even if we somehow reach A5 in 2035, we are only at about 12x density increase. If we include system packaging, chiplet, interconnect advancement pushing this to 30 to 40x. This is still a far cry from the 1000 to 10000x compute demands from a lot of AI companies. And that is assuming memory bandwidth could scale with it.
Ai leads to a capital concentration on the hands of those that already have money and might lead to a long term reduction of wealth for the middle class.
Less purchasing power in the population usually is not good for economic development, so I have my doubts with respect to a boom .
Feel free to skip the next section about the culture-aware interpretation of that quote. Xi's warning is more about the so-called "political tournament". Usually the central government sets some industrial policy goal (AI and EV now, but it could be other things like real estate, chips, or drones from before). The local government would collude with companies to start "projects". The goal is to get brownies points for the former and subsidy package for the latter. Of course, nothing usually comes out of these "projects". Most just stops at the factory building phase. Such practice has always been a real headache to the central government. Xi is warning about it showing up in the AI and EV industry now.
[^1]: https://paper.people.com.cn/rmrb/pc/content/202507/17/conten...
This reminds me of the .com bubble when bandwidth, especially transatlantic bandwidth, was dire but then all of a sudden it wasn't and we had huge overcapacities for years.
I wonder if this will happen witn compute?