Unlike railroads and fibre, all the best compute in 2025 will be lacklustre in 2027. It won’t retain much value in the same way as the infrastructure of previous bubbles did?
> Unlike railroads and fibre, all the best compute in 2025 will be lacklustre in 2027.
I definitely don't think compute is anything like railroads and fibre, but I'm not so sure compute will continue it's efficiency gains of the past. Power consumption for these chips is climbing fast, lots of gains are from better hardware support for 8bit/4bit precision, I believe yields are getting harder to achieve as things get much smaller.
Betting against compute getting better/cheaper/faster is probably a bad idea, but fundamental improvements I think will be a lot slower over the next decade as shrinking gets a lot harder.
>> Unlike railroads and fibre, all the best compute in 2025 will be lacklustre in 2027.
> I definitely don't think compute is anything like railroads and fibre, but I'm not so sure compute will continue it's efficiency gains of the past. Power consumption for these chips is climbing fast, lots of gains are from better hardware support for 8bit/4bit precision, I believe yields are getting harder to achieve as things get much smaller.
I'm no expert, buy my understanding is that as feature sizes shrink, semiconductors become more prone to failure over time. Those GPUs probably aren't going to all fry themselves in two years, but even if GPUs stagnate, chip longevity may limit the medium/long term value of the (massive) investment.
“Within Parsons' lifetime, the generating capacity of a [steam turbine] unit was scaled up by about 10,000 times” [1].
For comparison, Moore’s law (at 2 years per doubling) scales 4 orders of magnitude in about 27 years. That’s roughly the lifetime of a modern steam turbine [2]. In actuality, Parsons lived 77 years [3], implying a 13% growth rate, so doubling every 6 versus 2 years. But within the same order of magnitude.
Unfortunately the chips themselves probably won’t physically last much longer than that under the workloads they are being put to. So, yes, they won’t be totally obsolete as technology in 2028, but they may still have to be replaced.
Yeah - I think that the extremely fast depreciation just due to wear and use on GPUs is pretty unappreciated right now. So you've spent 300 mil on a brand new data center - congrats - you'll need to pay off that loan and somehow raise another 100 mil to actually maintain that capacity for three years based on chip replacement alone.
There is an absolute glut of cheap compute available right now due to VC and other funds dumping into the industry (take advantage of it while it exists!) but I'm pretty sure Wall St. will balk when they realize the continued costs of maintaining that compute and look at the revenue that expenditure is generating. People think of chips as a piece of infrastructure - you buy a personal computer and it'll keep chugging for a decade without issue in most case - but GPUs are essentially consumables - they're an input to producing the compute a data center sells that needs constant restocking - rather than a one-time investment.
- Most big tech companies are investing in data centers using operating cash flow, not levering it
- The hyperscalers have in recent years been tweaking the depreciation schedules of regular cloud compute assets (extending them), so there's a push and a pull going on for CPU vs GPU depreciation
- I don't think anyone who knows how to do fundamental analysis expects any asset to "keep chugging for a decade without issue" unless it's explicitly rated to do so (like e.g. a solar panel). All assets have depreciation schedules, GPUs are just shorter than average, and I don't think this is a big mystery to big money on Wall St
Do we actually know how they're degrading? Are there still Pascals out there? If not, is it because they actual broke or because of poor performance? I understand it's tempting to say near 100% workload for multiple years = fast degradation, but what are the actual stats? Are you talking specifically about the actual compute chip or the whole compute system -- I know there's a big difference now with the systems Nvidia is selling. How long do typical Intel/AMD CPU server chips last? My impression is a long time.
If we're talking about the whole compute system like a gb200, is there a particular component that breaks first? How hard are they to refurbish, if that particular component breaks? I'm guessing they didn't have repairability in mind, but I also know these "chips" are much more than chips now so there's probably some modularity if it's not the chip itself failing.
I watch a GPU repair guy and its interesting to see the different failure modes...
* memory IC failure
* power delivery component failure
* dead core
* cracked BGA solder joints on core
* damaged PCB due to sag
These issues are compounded by
* huge power consumption and heat output of core and memory, compared to system CPU/memory
* physical size of core leads to more potential for solder joint fracture due to thermal expansion/contraction
* everything needs to fit in PCIe card form factor
* memory and core not socketed, if one fails (or supporting circuitry on the PCB fails) then either expensive repair or the card becomes scrap
* some vendors have cards with design flaws which lead to early failure
* sometimes poor application of thermal paste/pads at factory (eg, only half of core making contact
* and, in my experience in aquiring 4-5 year old GPUs to build gaming PCs with (to sell), almost without fail the thermal paste has dried up and the card is thermal throttling
These failures of consumer GPUs may be not applicable to datacenter GPUs, as the datacenter ones are used differently, in a controlled environment, have completely different PCBs, different cooling, different power delivery, and are designed for reliability under constant max load.
Yeah you're right. Definitely not applicable at all. Especially since nvidia often supplies them tied into the dgx units with cooling etc. Ie a controlled environment.
Consuker gpu you have no idea if they've shoved it into a hotbox of a case or not
Could AI providers follow the same strategy? Just throw any spare inference capacity at something to make sure the GPUs are running 24/7, whether that's model training, crypto mining, protein folding, a "spot market" for non-time-sensitive/async inference workloads, or something else entirely.
I have to imagine some of them try this. I know you can schedule non-urgent work loads with some providers that run when compute space is available. With enough work loads like that, assuming they have well-defined or relatively predictable load/length, it would be a hard but approximately solvable optimization problem.
I've seen things like that, but I haven't heard of any provider with a bidding mechanic for allocation of spare compute (like the EC2 spot market).
I could imagine scenarios where someone wants a relatively prompt response but is okay with waiting in exchange for a small discount and bids close to the standard rate, where someone wants an overnight response and bids even less, and where someone is okay with waiting much longer (e.g. a month) and bids whatever the minimum is (which could be $0, or some very small rate that matches the expected value from mining).
Yep, we are (unfortunately) still running on railroad infrastructure built a century ago. The amortization periods on that spending is ridiculously long.
Effectively every single H100 in existence now will be e-waste in 5 years or less. Not exactly railroad infrastructure here, or even dark fiber.
> Yep, we are (unfortunately) still running on railroad infrastructure built a century ago.
That which survived, at least. A whole lot of rail infrastructure was not viable and soon became waste of its own. There was, at one time, ten rail lines around my parts, operated by six different railway companies. Only one of them remains fully intact to this day. One other line retained a short section that is still standing, which is now being used for car storage, but was mostly dismantled. The rest are completely gone.
When we look back in 100 years, the total amortization cost for the "winner" won't look so bad. The “picks and axes” (i.e. H100s) that soon wore down, but were needed to build the grander vision won't even be a second thought in hindsight.
> That which survived, at least. A whole lot of rail infrastructure was not viable and soon became waste of its own. There was, at one time, ten rail lines around my parts, operated by six different railway companies. Only one of them remains fully intact to this day. One other line retained a short section that is still standing, which is now being used for car storage, but was mostly dismantled. The rest are completely gone.
How long did it take for 9 out of 10 of those rail lines to become nonviable? If they lasted (say) 50 years instead of 100, because that much rail capacity was (say) obsoleted by the advent of cars and trucks, that's still pretty good.
> How long did it take for 9 out of 10 of those rail lines to become nonviable?
Records from the time are few and far between, but, from what I can tell, it looks like they likely weren't ever actually viable.
The records do show that the railways were profitable for a short while, but it seems only because the government paid for the infrastructure. If they had to incur the capital expenditure themselves, the math doesn't look like it would math.
Imagine where the LLM businesses would be if the government paid for all the R&D and training costs!
Railroads were pretty profitable for a long time. The western long haul routes were capitalized by land transfers.
What killed them was the same thing that killed marine shipping — the government put the thumb on the scale for trucking and cars to drive postwar employment and growth of suburbs, accelerate housing development, and other purposes.
> the government put the thumb on the scale for trucking and cars to drive postwar employment and growth of suburbs, accelerate housing development, and other purposes.
The age of postwar suburb growth would be more commonly attributed to WWII, but the records show these railroads were already losing money hand over fist by the WWI era. The final death knell, if there ever was one, was almost certainly the Great Depression.
But profitable and viable are not one and the same, especially given the immense subsidies at play. You can make anything profitable when someone else is covering the cost.
There was alot of complexities. It's hard to really understand the true position of these businesses in modern terms. Operationally, they would often try to over-represent losses because the interstate commerce commission and other State-level entities mandated services, especially short-haul passenger service that become irrelevant.
National infrastructure is always subsidized and is never profitable on it's own. UPS is the largest trucking company, but their balance sheet doesn't reflect the costs of enabling their business. The area I grew up in had tarred gravel roads exclusively until the early 1980s -- they have asphalt today because the Federal government subsidizes the expense. The regulatory and fiscal scale tipped to automotive and to a lesser extent aircraft. It's arguable whether that was good or bad, but it is.
> The records do show that the railways were profitable for a short while, but it seems only because the government paid for the infrastructure. If they had to incur the capital expenditure themselves, the math doesn't look like it would math.
Actually, governments in the US rarely actually provided any capital to the railroads. (Some state governments did provide some of the initial capital for the earliest railroads). Most of federal largess to the railroads came in the form of land grants, but even the land grant system for the railroads was remarkably limited in scope. Only about 7-8% of the railroad mileage attracted land grants.
> Actually, governments in the US rarely actually provided any capital to the railroads.
Did I, uh, miss a big news announcement today or something? Yesterday "around my parts" wasn't located in the US. It most definitely wasn't located in the US when said rail lines were built. Which you even picked up on when you recognized that the story about those lines couldn't have reasonably been about somewhere in the US. You ended on a pretty fun story so I guess there is that, but the segue into it wins the strangest thing ever posted to HN award. Congrats?
If 1/10 investment lasts 100 years that seems pretty good to me. Plus I'd bet a lot of the 9/10 of that investment had a lot of the material cost re-coup'd when scrapping the steel. I don't think you're going to recoup a lot of money from the H100s.
Much like LLMs. There are approximately 10 reasonable players giving it a go, and, unless this whole AI thing goes away, never to be seen again, it is likely that one of them will still be around in 100 years.
H100s are effectively consumables used in the construction of the metaphorical rail. The actual rail lines had their own fare share of necessary tools that retained little to no residual value after use as well. This isn't anything unique.
H100s being thought of as consumables is keen - it much better to analogize the H100s to coal and chip manufacturer the mine owner - than to think of them as rails. They are impermanent and need constant upkeep and replacement - they are not one time costs that you build as infra and forget about.
> Effectively every single H100 in existence now will be e-waste in 5 years or less.
This remains to be seen. H100 is 3 years old now, and is still the workhorse of all the major AI shops. When there's something that is obviously better for training, these are still going to be used for inference.
If what you say is true, you could find a A100 for cheap/free right now. But check out the prices.
The A100 SXM4 has a TDP of 400 watts, let's say about 800 with cooling etc overhead.
Bulk pricing per KWH is about 8-9 cents industrial. We're over an order of magnitude off here.
At 20k per card all in price (MSRSP + datacenter costs) for the 80GB version, with a 4 year payoff schedule the card costs 57 cents per hour (20,000/24/365/4) assuming 100% utilization.
> Yep, we are (unfortunately) still running on railroad infrastructure built a century ago. The amortization periods on that spending is ridiculously long.
Are we? I was under the impression that the tracks degraded due to stresses like heat/rain/etc. and had to be replaced periodically.
The track bed, rails, and ties will have been replaced many times by now. But the really expensive work was clearing the right of way and the associated bridges, tunnels, etc.
I am really digging the railroad analogies in this discussion! There are some striking similarities if you do the right mappings and timeframe transformations.
I am an avid rail-to-trail cycler and more recently a student of the history of the rail industry. The result was my realization that the ultimate benefit to society and to me personally is the existence of these amazing outdoor recreation venues. Here in Western PA we have many hundreds of miles of rail-to-trail. My recent realization is that it would be totally impossible for our modern society to create these trails today. They were built with blood, sweat, tears and much dynamite - and not a single thought towards environmental impact studies. I estimate that only ten percent of the rail lines built around here are still used for rail. Another ten percent have become recreational trails. That percent continues to rise as more abandoned rail lines transition to recreational use. Here in Western PA we add a couple dozen miles every year.
After reading this very interesting discussion, I've come to believe that the AI arms race is mainly just transferring capital into the pockets of the tool vendors - just as was the case with the railroads. The NVidia chips will be amortized over 10 years and the models over perhaps 2 years. Neither has any lasting value. So the analogy to rail is things like dynamite and rolling stock. What in AI will maintain value? I think the data center physical plants, power plants and transmission networks will maintain their value longer. I think the exabytes of training data will maintain their value even longer.
What will become the equivalent of rail-to-trail? I doubt that any of the laborers or capitalists building rail lines had foreseen that their ultimate value to society would be that people like me could enjoy a bike ride. What are the now unforeseen long-term benefit to society of this AI investment boom?
Rail consolidated over 100 years into just a handful of firms in North America, and my understanding is that these firms are well-run and fairly profitable. I expect a much more rapid shakeout and consolidation to happen in AI. And I'm putting my money on the winners being Apple first and Google second.
Another analogy I just thought of - the question of will the AI models eventually run on big-iron or in ballpoint pens. It is similar to the dichotomy of large-scale vs miniaturized nuclear power sources in Asimov's Foundation series (a core and memorable theme of the book that I haven't seen in the TV series).
"...all the best compute in 2025 will be lacklustre in 2027": How does the compute (I assume you mean on PCs) of 2025 compare with the compute of 2023?
Oh wait, the computer I'm typing this on was manufactured in 2020...
Neato. How’s that 1999 era laptop?
Because 25 year old trains are still running and 25 year old train track is still almost new. It’s not the same and you know it.
Exactly: when was the last time you used ChatGPT-3.5? Its value deprecated to zero after, what, two-and-a-half years? (And the Nvidia chips used to train it have barely retained any value either)
The financials here are so ugly: you have to light truckloads of money on fire forever just to jog in place.
I would think that it's more like a general codebase - even if after 2.5 years, 95% percent of the lines were rewritten, and even if the whole thing was rewritten in a different language, there is no point in time at which its value diminished, as you arguably couldn't have built the new version without all the knowledge (and institutional knowledge) from the older version.
I rejoined an previous employer of mine, someone everyone here knows ... and I found that half their networking equipment is still being maintained by code I wrote in 2012-2014. It has not been rewritten. Hell, I rewrote a few parts that badly needed it despite joining another part of the company.
OpenAI is now valued at $500bn though. I doubt the investors are too wrecked yet.
It may be like looking at the early Google and saying they are spending loads on compute and haven't even figured how to monetize search, the investors are doomed.
Google was founded in 1998 and IPOed in 2004. If OpenAI was feeling confident they'd find ways to set up a company and IPO, 9 years after founding. It's all mostly fictional money at this point.
It's not about confidence. OpenAI would be huge on the public markets, but since they can raise plenty of money in the private market there is no reason to deal with that hassle - yet.
A really did few days ago gpt-3.5-fast is a great model for certain tasks and cost wise via the API. Lots of solutions being built on the today’s latest are for tomorrow’s legacy model — if it works just pin the version.
I don't see why these companies can't just stop training at some point. Unless you're saying the cost of inference is unsustainable?
I can envision a future where ChatGPT stops getting new SOTA models, and all future models are built for enterprise or people willing to pay a lot of money for high ROI use cases.
We don't need better models for the vast majority of chats taking place today E.g. kids using it for help with homework - are today's models really not good enough?
They aren't. They are obsequious. This is much worse than it seems at first glance, and you can tell it is a big deal because a lot of effort going into training the new models is to mitigate it.
>I don't see why these companies can't just stop training at some point.
Because training isn't just about making brand new models with better capabilities, it's also about updating old models to stay current with new information. Even the most sophisticated present-day model with a knowledge cutoff date of 2025 would be severely crippled by 2027 and utterly useless by 2030.
Unless there is some breakthrough that lets existing models cheaply incrementally update their weights to add new information, I don't see any way around this.
There is no evidence that RAG delivers equivalent performance to retraining on new data. Merely having information in the context window is very different from having it baked into the model weights. Relying solely on RAG to keep model results current would also degrade with time, as more and more information would have to be incorporated into the context window the longer it's been since the knowledge cutoff date.
I honestly do not think that we should be training models to regurgitate training data anyway.
Humans do this to a minimum degree, but the things that we can recount from memory are simpler than the contents of an entire paper, as an example.
There's a reason we invented writing stuff down. And I do wonder if future models should be trying to optimise for rag with their training; train for reasoning and stringing coherent sentences together, sure, but with a focus on using that to connect hard data found in the context.
And who says models won't have massive or unbounded contexts in the future? Or that predicting a single token (or even a sub-sequence of tokens) still remains a one shot/synchronous activity?
Not necessarily? That assumes that the first "good enough" model is a defensible moat - i.e., the first ones to get there becomes the sole purveyors of the Good AI.
In practice that hasn't borne out. You can download and run open weight models now that are spitting distance to state-of-the-art, and open weight models are at best a few months behind the proprietary stuff.
And even within the realm of proprietary models no player can maintain a lead. Any advances are rapidly matched by the other players.
More likely at some point the AI becomes "good enough"... and every single player will also get a "good enough" AI shortly thereafter. There doesn't seem like there's a scenario where any player can afford to stop setting cash on fire and start making money.
Perhaps the first thing the owners ask the first true AGI is “how do I dominate the world?” and the AGI outlines how to stop any competitor getting AGI..?
Except they behave less like shrewd investors and more like bandwagon jumpers looking to buy influence or get rich quick. Crypto, Twitter, ridesharing, office sharing and now AI. None of these have been the future of business.
Business looks a lot like what it has throughout history. Building physical transport infrastructure, trade links, improving agricultural and manufacturing productivity and investing in military advancements. In the latter respect, countries like Turkey and Iran are decades ahead of Saudi in terms of building internal security capacity with drone tech for example.
Agreed - I don’t think they are particularly brilliant as a category. Hereditary kleptocracy has limits.
But… I don’t think there’s an example in modern history of the this much capital moving around based on whim.
The “bet on red” mentality has produced some odd leaders with absolute authority in their domain. One of the most influential figures on the US government claims to believe that he is saving society from the antichrist. Another thinks he’s the protagonist in a sci-fi novel.
We have the madness of monarchy with modern weapons and power. Yikes.
Speculating but they pay to be integrated as the default ai integration in various places the same way google has paid to be the default search engine on things like the iPhone?
There is a YouTuber making a really remote cabin in I think northern Canada called Off Grid Engineering and he has a similar narrative style. Nice :). Recommended
There are lots of young whippersnappers and “old timers” in the “west” who could easily do the Low level make it quick on small hardware stuff, the US companies just aren’t asking us to?
> A human neuron is a thousand times bigger than a transistor.
Correct, it works on principles currently completely unapplied in ASIC design. We don't, for example, have many mechanisms that allow for new pathways to be formed in hardware. At least, not outside of highly controlled fashion. It's not clear that it would even be helpful if we did.
> There are directions hardware and algorithms have been going in - parallel processing - that are not limited by fabrication?
They are limited by the power budget. Yes we can increase the amount of parallel compute 100x but not without also increasing the power budget by 100x.
But further, not all problems can be made parallel. Data dependencies exist and those always slow things down. Further, coordination isn't free for parallel algorithms.
I'm not saying there's not some new way to do computation which hasn't been explored. I'm saying we've traveled down a multi-decade path to today's compute capabilities and we may be at the end of this road. Building a new model that's ultimately adopted will (likely) take more decades. I mean, consider how hard it's been to purge x86 from society. We are looking at a problem a million times more difficult than just getting rid of x86.
I imagine we know we have reached AGI when a technocrat stops sharing their AI. It goes from being something they can sell to something that they don’t want to share. Instead they ask it how they can dominate the world and be the first trillionaire and how they can stop anyone else acquiring an AGI etc.
This even works at a smaller not so general level: imagine that one of today’s popular code models improved to the point it is better (narrowly at programming) than a human. Suddenly the owner shouldn’t sell it to everyone: instead they should pivot and make software that outcompetes anything a human can make. So it doesn’t just replace humans making software but also replaces the programs that people made…
I was a the kind of person who was happy as a pig in mud to be paid to do my hobby of programming computers! Was ecstatic that people would pay money to a young kid to do that!
But most of the people I went to uni to study computer science with at the end of the nineties were there for the money. Even back then it was all about money for most programmers.
There is a generation of programmers that became interested in computers only because they felt that computers were cool. Mostly useless, except for playing games, but cool. Only later the knowledge also turned out to be a source of money.
And then there is a generation that grew up knowing that there was money in computers, so many of them learned to use them even if they didn't care about them per se. This generation also contains many hackers, but they are surrounded by at least 10x more people who only do it for money.
Twenty years ago, most programmers were nerds. These days, nerds are a minority among the programmers. Talking about programming during an IT department teambuilding event is now a serious faux pas.
Crikey, I”d hardly sell the UK as good with land access! The UK is pretty awful in comparison to the nearby Nordics. Sweden, for example, has a right to roam in nature which makes the constant antagonism between footpath walkers and landowners that are a mainstay of the English countryside seem so petty.
Depends of course upon which UK legal system we are discussing... The right to roam provided in Scotland under Scots law is often cited as one of the best examples of the concept. The UK is more than just England etc.
I'll tell you what, the Nordics themselves are pretty awful when compared with the Outback or American southwest. You can just roam wherever you want, no questions asked!
Our national park system in the US is larger than the entire country of Sweden and Denmark and Iceland (or close enough) - depending on how it's defined maybe throw Norway in there too. I don't need to walk through someone's private property to see all this great stuff.
Maybe the Nordics should set aside more public land and catch up to the United States?
I don't want people roaming through my yard and stepping on my plants and stuff or god forbid they bring their stinky dogs and their urine and feces so I have a fence put up to keep them out.
Sweden has the right to roam on privately owned nature property but it is trumped by exceptions for crops and the right to privacy at home. So it’s not ok - and not done - to walk in peoples yards etc. The rules are well taught at schools and well explained for tourists and it just works nicely.
In the United States at least we don't really have privately owned nature property like you might in Sweden. I live in Ohio for example, there's nothing to go see or look at. We have no need for the right to roam. What are you going to do, roam through a cornfield? A parking lot? The woods? The mall? Well you can already do that. We have state parks, local parks, national parks, etc. to get your nature fix and it works very well here, there are no complaints about this whatsoever.
Sometimes Europeans are so convinced that their way of life is better or their policies are the best they forget that sometimes their policies solve problems that don't exist in other countries. There's no need to have a right to roam in America. There's nowhere to roam to, and the places that you would roam to are already owned by the public where you have... the right to roam! Though we are much more strict about natural preservation in those parks which sometimes conflicts with the desires of some to go "off trail", but that's a separate issue.
The UK might be a little different, granted, but the no-true-scotsman approach to someone suggesting they enjoy the UK's right to roam but they can't because the Nordic countries are so much better in this regard is annoying, to say the least.
Yea but the discussion happening in this thread was about something else. If you don't want to participate that's fine but please don't derail ongoing discussions.
You come across very Italian, for sure. I guess? :)
reply