Hacker News Reader

Comment by frithsun | Hacker News Reader

I believe this is a "good" bubble in the sense that the 19th century railroad bubble and original dot com bubble both ended up invested in infrastructure that created immense value.

That said, all of these LLMs are interchangeable, there are no moats, and the profit will almost entirely be in the "last mile," in local subject matter experts applying this technology to their bespoke business processes.

View on HN

America's railroad boom isn't a great example, it got us the worst rail infrastructure in the world, built by private monopolies solely for maximum short-term profit, i.e. moving freight and not passengers, and now American industry is largely gone and we're stuck with rail infrastructure that is useless to almost everyone and it costs far more to maintain it than it's even worth.

America's internet infrastructure, like the railroads, was also left in the hands of private monopolies and it is also a piece of shit compared to other countries. It's slow and everyone pays far too much for it and many are still excluded from it because it's not profitable enough to run fiber to their area.

The AI bubble won't leave behind any new infrastructure when it bursts. Just millions of burned out GPUs that get sent to an e-waste processing plant where they are ground up into sand, trillions of dollars wasted, many terawatt hours of energy wasted, many billions of liters of freshwater wasted, and the internet being buried under an avalanche of pseudorandomly-generated garbage.

> "good" bubble in the sense

how can massively buying hardware that will have to be thrown away in a few years be a "good" bubble in the sense of being a lasting infrastructure investment?

I am pretty optimistic that as long as hardware capacity exists, people will find ways of using it. Whether it will be profitable or not is another story of course.

Rivers overflowing with legacy hardware and villages incinerating boards for their metals, and the caustic effects on people & their environment that causes, are already happening. The hardware capacity exists only as long as it is operational and within a few generations. Perhaps we should be careful before building Manhattan-sized data centers.

Up to a point it is better than having additional compute sitting idle at the edge, economies of scale and all that, but after some point it becomes excess and wasteful, even if people figure out ways to entertain themselves with it.

And if people don't want to pay what it costs to improve and maintain these city-sized electronic brains? Then it all becomes waste, or the majority transformed into office or warehouse space or something else.

Proceeding with combined 1% (US GDP)-sized budgets despite this risk being an elephant in the room is what makes it a bubble.

I completely agree with a lot of your points; the whole thing is quite stupid. My only objection is that the infrastructure will probably not go unused. And if we are lucky, those uses will be better than helping teenagers cheat themselves out of a good education.

Aren't AI GPUs a drop in the bucket compared to consumer electronics?

Nvidia sold ~3M blackwells in 2025: https://wccftech.com/nvidia-has-sold-over-three-million-blac...

Compare that to laptops which sell in tens of millions per manufacturer: https://en.wikipedia.org/wiki/List_of_laptop_brands_and_manu...

Plus, it's way easier to collect boards for recycling from a centralized data center.

Nvidia had $35.6 billion in data center revenue vs $2.5 billion "Gaming and AI PC" revenue in the 4th quarter of 2024 so data center stuff stuff is like 93% of it

Completely agree. I would ask also what “infrastructure” the dotcom bubble created?

Data centers and fiber optic connections across the world.

They’re referring to this: https://en.wikipedia.org/wiki/Dark_fibre

But the large majority of this was created after the bubble bursted, no?

I wouldn't call that wasted yet. It's just latent. Someone has to invent the startup that uses dark fibre maps to figure out exactly how far away one can build a house away from civilization and still have a 1000 MB/s connection.

The models themselves, and methods and knowledge used to build and use them, are part of the "infrastructure" being built.

You're redefining infrastructure. A supply and demand model is not infrastructure. A Taylor expansion method is not infrastructure.

that's the point - learning how to train the models and run them at scale is the "infrastructure" funded by the bubble that will be useful after it pops.

Why would that hardware "have to be thrown away"? I've seen quite old GPUs still in use; given the current demand, I expect the vast majority of hardware used in these data centers to see a lot more extended use than most other types of electronics around the world (e.g. phones).

GPUs in data centers have short lifespans.

https://www.tomshardware.com/pc-components/gpus/datacenter-g...

Oh, interesting - it's about failure rate following degradation after prolonged use. I didn't think of that - but my take is that if companies like Google are actually using those components until they 100% exhaust what the components are capable of, then we can argue about whether this use of resources is better than an alternative use of them, but it's by definition not a waste of resources.

The prices are falling down. I do a lot of Machine Learning and sometimes work with large datasets. The ability to (1) put all data in VRAM and (2) have the results in hours/days instead of weeks is amazing - and in the past it wouldn't be easy for a normal researcher like me. Now I can have access to these beefy machines, do my research and publish the results without taking a loan from my bank.

One large but forgotten effect of the dotcom bubble was an excess fiber capacity, that allowed smooth growth of internet in the following 25 years---average internet speed in the US is 200 Mbps, and a significant number of households is on a gigabit uplink. I take your point that GPU hardware amortizes away faster than fiber, but that's true of all computing hardware: the average lifecycle of a server is around five years.

All over rural New England you'll find abandoned rail lines. Many of these were used for passenger service between walkable towns. Now, Boston area commuter rail sucks big time. Towns now have commercial stips "served" by cars-only "stroads."

How well used do you think those AI data centers are going to be?

Ok, but what's the infrastructure that will remain after the AI bubble that can be retooled like railroads or dot com?

Maybe we'll all be able to afford secondhand H100s and finally get to see raytraced cyberpunk at 4k120

I keep saying to people - "if you have a good idea that can make use of large amounts of really really cheap GPUs to do something genuinely useful - get ready for a massive glut of spare capacity". I still haven't thought of anything, unfortunately...

These are sort of compute-focused GPUs, right? I bet a lot of university labs would like them.

I wonder if ubiquitous, user-friendly finite elements analysis tools could become a boon for 3D printers.

Hopefully they can be repurposed for something like cheap drug discovery rather than shitcoin mining.

If that ends up being the case then we could all genuinely agree that good has eventually emerged from the compute/inference infrastructure that LLMs paid for. I hope that comes to pass.

Yeah but we all know it's definitely gonna be shitcoin mining

I would love to build a render farm out of cheap decommissioned GPUs...

I hope it will bring us reliably good memory bandwidth in consumer devices, an area where many hardware vendors are skimping.

Reply on news.ycombinator.com