2. Building factories requires capital to be invested now.
3. The return-on-capital will only be obtained in the next n years.
4. But if demand goes down, we'll have much more supply than demand, leading to a cutthroat price competition, which could prevent the factory costs from ever being recouped.
And to add to that 4th point, there is a very real chance that many the “AI” companies that are buying all this hardware could go bankrupt at anytime and in rapid succession, leaving drive/memory manufactures with a bunch of unpaid orders and a second hand market now saturated with liquidated equipment.
Much like the GPU/crypto boom/bust some years ago, but on an even larger scale.
> 4. But if demand goes down, we'll have much more supply than demand, leading to a cutthroat price competition, which could prevent the factory costs from ever being recouped.
Why not normalize that, if things don't work out, the government steps in, buys the business, and offers a better regulated version of the business?
Capitalism is generally touted as useful for innovation; but haven't we paid collectively for the initial investment for many businesses many times over now that there's little innovation done? By definition there's no innovation once the only thing happening in a business sector is consolidation.
So, I'd conclude that capitalism isn't useful to us at that point, and actually harmfully inefficient (too costly for consumers), so we'd be wise to do away with it by that point in favor of a more efficiently run operational/economic system... That would actually be innovative! What that system looks like, I'm not sure, but I imagine it's something like a cooperative.
Production was close to maximum anyway because of existing demand and how expensive new fabs are to bring up. The boom in AI use wasn't sufficiently planned for as it wasn't expected at the scale we are seeing, so to scale up means building more fabs - a long and expensive task. The situation is not helped by one of the AI companies successfully negotiating huge deals for a year worth of production with the two biggest providers at the same time and keeping it secret enough that neither bumped up prices in the deal as a result of knowing what the play was.
Most players are rational actors and this isn't their first time in this sort of market. Getting more production online if you do not have slack like downtime takes quite a while. And there is no guarantee that they can sell the new production even at previous rates if they end up in over supply.
Generally people just don't understand how long ramping up new production facilities can take and what is the realistic pay off period for them.
Adding to what others have said, several hdd, ssd and ram manufactures have said that they don't believe demand will be sustained in time for long enough to warrant extra investments in increasing production.
Building these factories goes way beyond "nontrivial". We may see a few come online in a couple of years though, but time will tell if the extra capacity alleviates the crunch.
Because the manufacturers of storage (and RAM) consider it to be an AI bubble, too. The surge in demand is a short burst, not a sustained one. Hence it makes no sense to throw a whole lot of ressources into scaling up production, when the demand won't be there anymore in 1-2 years when the factories will be ready.
Yes, but that isn't where the increase in demand is. Those things are affected by fabs that could be producing them producing something else, but there will be some stock floating in the system so some resistance to increased prices needed to justify new fab resource, and if things correct a bit in the coming year the maths for a new build might look more dubious. Those with the money to fund a new fab right now are more likely to fund something capable of producing the newer part types. I could be wrong, but the fact that significant new fabs like that are not in progress right now would suggest not.
Shouldn't countries wanting sovereign infrastructure create subsidies for creation of factories/job creation and also selling first/primarily within the region if it might cost on just a few million dollars (preferably a new competitor)
I think one flaw in my thinking could be that there might be a lack of experience within the people for something like this, do you consider it to be a factor and would it be difficult to hire people relevant to such fab?
Because it operates exactly as a drug dealer. It gives you first shot for free (reasonable opportunity to move up) and after you are hooked it makes sure that it extracts all your money (subscription and inability to own anything).
Because:
1. Factories take time to build.
2. Building factories requires capital to be invested now.
3. The return-on-capital will only be obtained in the next n years.
4. But if demand goes down, we'll have much more supply than demand, leading to a cutthroat price competition, which could prevent the factory costs from ever being recouped.
And to add to that 4th point, there is a very real chance that many the “AI” companies that are buying all this hardware could go bankrupt at anytime and in rapid succession, leaving drive/memory manufactures with a bunch of unpaid orders and a second hand market now saturated with liquidated equipment.
Much like the GPU/crypto boom/bust some years ago, but on an even larger scale.
No 4 is why nobody is willing to really move on new fascilities.
China is doing it but that was for merely creating tech independence.
> 4. But if demand goes down, we'll have much more supply than demand, leading to a cutthroat price competition, which could prevent the factory costs from ever being recouped.
Why not normalize that, if things don't work out, the government steps in, buys the business, and offers a better regulated version of the business?
Capitalism is generally touted as useful for innovation; but haven't we paid collectively for the initial investment for many businesses many times over now that there's little innovation done? By definition there's no innovation once the only thing happening in a business sector is consolidation.
So, I'd conclude that capitalism isn't useful to us at that point, and actually harmfully inefficient (too costly for consumers), so we'd be wise to do away with it by that point in favor of a more efficiently run operational/economic system... That would actually be innovative! What that system looks like, I'm not sure, but I imagine it's something like a cooperative.
Because when the AI customers explode N months down the line, you don't want to be on the hook for a new factory.
Building production facilities isn't like flipping a switch.
Production was close to maximum anyway because of existing demand and how expensive new fabs are to bring up. The boom in AI use wasn't sufficiently planned for as it wasn't expected at the scale we are seeing, so to scale up means building more fabs - a long and expensive task. The situation is not helped by one of the AI companies successfully negotiating huge deals for a year worth of production with the two biggest providers at the same time and keeping it secret enough that neither bumped up prices in the deal as a result of knowing what the play was.
Most players are rational actors and this isn't their first time in this sort of market. Getting more production online if you do not have slack like downtime takes quite a while. And there is no guarantee that they can sell the new production even at previous rates if they end up in over supply.
Generally people just don't understand how long ramping up new production facilities can take and what is the realistic pay off period for them.
Adding to what others have said, several hdd, ssd and ram manufactures have said that they don't believe demand will be sustained in time for long enough to warrant extra investments in increasing production.
Building these factories goes way beyond "nontrivial". We may see a few come online in a couple of years though, but time will tell if the extra capacity alleviates the crunch.
It is the law of inertia, it applies to many more situations than we think. Markets is one of the, especially large-scale ones.
Because the manufacturers of storage (and RAM) consider it to be an AI bubble, too. The surge in demand is a short burst, not a sustained one. Hence it makes no sense to throw a whole lot of ressources into scaling up production, when the demand won't be there anymore in 1-2 years when the factories will be ready.
A chip fab costs a billion dollars to build.
Modern cutting edge ones, yes. Not all fabs, by a long shot.
Who is going to pay for a non-cutting-edge fab that might be completely out of date relatively soon though?
TI started production at their SM1 fab back in December 2025, which focuses on 28 nm to 130 nm.
Automotive, space and defense happily use older processes at scale. Also embedded electronics in all sorts of electronics.
Yes, but that isn't where the increase in demand is. Those things are affected by fabs that could be producing them producing something else, but there will be some stock floating in the system so some resistance to increased prices needed to justify new fab resource, and if things correct a bit in the coming year the maths for a new build might look more dubious. Those with the money to fund a new fab right now are more likely to fund something capable of producing the newer part types. I could be wrong, but the fact that significant new fabs like that are not in progress right now would suggest not.
Shouldn't countries wanting sovereign infrastructure create subsidies for creation of factories/job creation and also selling first/primarily within the region if it might cost on just a few million dollars (preferably a new competitor)
I think one flaw in my thinking could be that there might be a lack of experience within the people for something like this, do you consider it to be a factor and would it be difficult to hire people relevant to such fab?
> just a few million dollars
TSMC Arizona projected investment is $165 billion. Not millions. And yes apparently hiring the right staff has been one of the issues.
People really underestimate the work of Maurice Chang.
USA tried this with the CHIPS act
Free markets, yes
Why isn't magic just doing the magic things the capitalists always tell us is magic?
Because it operates exactly as a drug dealer. It gives you first shot for free (reasonable opportunity to move up) and after you are hooked it makes sure that it extracts all your money (subscription and inability to own anything).