/s
(edit: fix name)
We will know once it all unfolds, on the other hand, at least there's people who are getting paid to build these.
Worse case, we have acccess to (hopefully) really well-built data centers that can be used for anything hosting related in the future.
After "AI": Anticipating a post-LLM science & technology revolution
https://www.evalapply.org/posts/after-ai/index.html
I, for one, welcome the coming age of the post-LLM-datacenter-overinvestment-bust-fueled backyard GPU supercomputer revolution.
(edit: in which case, I'd say more people would get paid to build more of these, except not as obnoxious township-destroying behemoths ... the community library, as opposed to the giant book seller in the cloud)
Older GPUs can still be used to run model inference, albeit more slowly than newer GPUs. We already see frontier labs turning this insight into an opportunity to charge a premium on token delivery speed, when models run on newer hardware. As AI becomes more widely spread throughout the economy, not all workloads will require fast inference, and there will be many use cases for previous generation hardware. The only exception is if the latest batch of specialised chips become ubiquitous very quickly, but then price of compute AND intelligence per watt will both go down dramatically, which would reduce the need for sustained high capex and opex
- Demand drops like a stone: raw TFLOPS / per model. This will happen if someone figures out insanely fast and power-efficient math (specifically, multiplication) for existing chip architectures, that makes it dirt-cheap to keep re-computing ever-better SOTA LLMs, as well as to on-line compute inference with ever-bigger models (scaling laws rule). These are different workloads, AFAIK.
- Moore's law re-starts, because someone figures out entirely new transistor designs that can deliver at least up to 100x per-core frequency gains over Si transistors (which would be nothing considering that Si transistor tech delivered ~1Mx gains since invention) and which can be manufactured at similar bin sizes, at similar scale (i.e. equivalent per-die yield), price-competitively (no more 10x the equivalent in SOTA Silicon compute).
I'd say that a math breakthrough might be more accessible than the transistor-manufacturing breakthrough.
That assumes that there is no hard, physical limit on Moore's Law, which is obviously false. There is a limit, and as long as we depend on mass flow of charge through conductors, it's nowhere near quantum sizes.
Where that limit is has not yet been determined, but at this point the support structure to write transistors is the size of a house trailer, using a staggering amount of toxic chemicals (just to run the exotic lasers). So, advancements today may scale very nonlinearly in cost.
The very phrase "Moore's law restarts" acknowledges that we are no longer on-track.
No, theoretically Moore's Law has a very hard limit, regardless of time.
Defraying of capex also depends on how the hyperscalers scrap used hardware.
AFAIK, all hyperscaler hardware is effectively a consumable. Disks, memory, boards, chips; all the high-value pieces do not get refurbished into new computers for sale into secondary markets. The commercial and legal liability is too high---leaking their own proprietary IP as well as leaking customer IP, not to mention secrets and hardening configs and so forth.
I can't recall where I read it, but if a datacenter is modeled as a bunch of black-box hot-swappable shipping containers, then for pre-LLM workloads, per-container unit economics works only as long as a container delivers over 30% capacity. After that you got to swap it out, and send the black box into a furnace to recover them rare earth metals for the next generation of chips.
To a first approximation, high-throughput, gpu-maxxing math workloads (LLM inference) are surely harder on chips and memory, than are bread-and-butter SaaS / data lake / ETL / RDBMS workloads. Speculating further, I'd guess that the ~40% operating margin of datacenters is due to the relative compute-friendliness of non-math workloads for lack of better words.
Could it be that they _have_ to run them like that because of the shortage of compute volume and once there are sufficient data centers, they can be run relatively cooler and thereby extending their lifespans? That doesn't bode well for NVIDIA/AMD.
> AFAIK, all hyperscaler hardware is effectively a consumable. Disks, memory, boards, chips; all the high-value pieces do not get refurbished into new computers for sale into secondary markets. The commercial and legal liability is too high---leaking their own proprietary IP as well as leaking customer IP, not to mention secrets and hardening configs and so forth.
Maybe that could be a catalyst for going for more "commodity" approach to hardware instead of highly spcialized (atleast for some sections of the hardware) so that the IP is not the hardware but the model/software and they can re-coup the costs by selling the hardware to "tier-2" providers that might emerge based on using cheaper second-hand hardware?
[Type A] Constant use a constant power level: Quantum world rules would dominate.
When run hot, electron tunneling will kill off individual transistors, and AFAIK that is a cascading failure mode once it takes root. A big enough failed spot, even if localised, could effectively kill a core (if not the whole chip---too many errors). A local hotspot will directly affect neighbours, and indirectly the chip as a whole because the same work is being pushed on fewer cores.
[Type B] Thermal cycled: Mechanical and electrical factors would dominate...
Microfractures / mechanical damage in weak areas of whole chips, and motherboards. In a hyperscaler server rack, if a board dies that part of the rack is dead---micro-maintenance is probably too expensive.
And speculating further, it stands to reason that Type B failures would likely cause Type A failures, and vice-versa.
server parts are tuned for power efficiency and longevity
B200 has an operating range of 10-35C. https://resources.nvidia.com/en-us-dgx-systems/dgx-b200-data...
RTX 5090 (and most consumer CPU/GPUs) have thermal limits at 90-100C.
No, it does not, its just revenue minus compute. Per their S1, Anthropic has never turned an operational profit.
Do they? Is H200 $-competitive with an ASIC for crypto? Bearing in mind they're 4 figures.
Neural networks in general are here to stay. Whether that is the current day frontier-model style LLM or not doesn't really matter. We've already had multiple different Deep Learning/Machine Learning architectures that have been incredibly applicable to multiple different domains.
Especially right now in the RAM-pocalpyse where it is difficult to get any machine, yes, these older GPUs are still very very useful. A guy at my company was having trouble getting compute allocated, budgeted, and approved - so eventually he went and dug up a bunch of old Volta-based GPUs, setup a local compute farm, and is getting work done. They are absolutely slower than current state-of-the-art, but they are still better than no compute at all.
I think one of the silver-linings of the current hardware high prices and shortages, is that people are going to (A) focus on software efficiency (B) not be so obsessed with having the latest and greatest hardware. Even on the consumer side, how many games ACTUALLY can utilize a RTX 5090?
It _could_ be the cost of running inference to respond to requests which bring in revenue, however the cost is described as "compute and infrastructure expenses", while revenue is described as "in-period revenue". I don't see these numbers matching anything else in the filing, however I do not claim to go over the entire document.
So to be perfectly honest I'd just ignore this metric completely :)
A data center is a warehouse for sure but the building is only a few percent of the cost. The GPUs are probably 80% of the total cost or more. And they probably only have a 3-5 year primary lifecycle.
Pay the building off over 20 years? Sure but 80% of your capex problems remain. 2 trillion already spent? Just those GPUs need to generate 500 billion a year over the next 5 years.
But as a fraction of inference revenue, there is good reason to estimate that the direct electricity costs are perhaps 1-5% of inference revenue.
That’s a lot, but hypothetically if electricity prices double, it’s more likely to lead to angry residential consumers than a doubling of inference token prices.
Lawrence Berkeley National Lab, November 2026: https://eta-publications.lbl.gov/sites/default/files/2026-11...
There are substantial economies of scale on both power generation (grid-scale) and inference power consumption (batching and utilization).
But that’s a feedback loop that could lead to more supply being built. Or it could lead to regulations that put enormous surcharges on data center power consumers, which effectively shield consumers directly from the lack of supply.
This would be the largest biomass plant in the US. 5x larger than the next largest plant.
A 500MW plant would require almost 50x the amount of available waste material from logging and sawmill operations in the entire state of TN (where the plant is being built).
The developer of the project said that the plant was designed to create demand for low grade pine timber following pulp mill closures in the area and that the plant will burn between 350-500 truckloads of logs per day.
IF prices of GPU/RAM come back to pre bubble costs, we might see the 5 year cycle of replacement that's relatively common in normal datacenters pre-ai bubble.
At current prices ? those GPUs that are there now will be running there in 10 years still.
If they don't burn out after two or three. Annualized failure rates are about 9% according to Meta.
Less than half of the original hardware will be running after ten years even if you assume that rate does not increase over time. Do H100s come with replacement guarantees?
This paper puts the failure rate of A100s closer to 0.3%/yr, H100s under 1%.
Mind you, most of the H100s that have ever been sold are still in their packaging, I suspect. Or at least not plugged in.
> [...] because quite frankly, the biggest issue we are now having is not a compute glut, but it’s power — it’s sort of the ability to get the builds done fast enough close to power. So, if you can’t do that, you may actually have a bunch of chips sitting in inventory that I can’t plug in. In fact, that is my problem today. It’s not a supply issue of chips; it’s actually the fact that I don’t have warm shells to plug into.
https://www.tomshardware.com/tech-industry/artificial-intell...
I do agree that 5 years isn't the hard cutoff I was suggesting but I'm not convinced that 10 years is a reasonable timeframe to consider paying the debt down when Moore's Law is involved. I know it's no longer the 12-18mo doubling it used to be but even if it's 2ish years then over 6 years you get 3 doublings or 8x performance gains. At 10 years it's another 2 doublings or 32x performance gains. It's hard for linear payback curves to deal with exponential performance curves past 2-3 doublings.
I guess this is the big gamble, the idea that if you just spend enough money as fast as you can you can become the gods of AI forever and can just charge whatever you want for your magic beans.
But what is the moat? If in reality these things become commodities and just compete with each other to spit out cheap tokens then there's no way the margins are there.
'Winner take all' doesn't mean convincing others to buy your product. If others have something to buy your product with, you've not taken all, fundamentally.
'Winner take all' means having the most powerful army. The assumption is that once AI crosses a certain threshold then there won't be anything the rest of us can do to stop the machine of whomever gets there first, so they will truly be able to take all. This is why the AI labs sometimes talk about the risk of human destruction.
But I am not sure you are right that it is strong enough to secure all. I search daily, but cannot remember the last time I used Google. And I'm not alone. Google's stock price recently took a nosedive because investors have grown quite worried about the competition. Without an army stopping them, there will always be someone else ready to try to outdo you.
Google shares dipped because AI shares dipped, not because their core business at risk. Google is in a _very_ strong position to capitalize on AI. Better than almost any other Big Tech company.
With AI, it looks to me like competition is much tighter, with no company having fundamentally superior technology or monetization plan. So a winner-take-all scenario is much less likely.
Maybe you mean that you could see the employees of those companies staging a coup to seize the army for themselves? That's certainly possible. They'd be in a good position to do so. Predicting the future is impossible, but if you want to be on the "right side" of that future army, owning its foundation today is still a reasonable bet.
Are you saying that winner-take-all markets are incompatible with capitalism? I'm confused
In reality the people would likely revolt by no longer recognizing private ownership of capital if that were to occur, but technically they don't have to. If they truly believe in capitalism they can continue to uphold the right for that private entity to own all the capital.
For the sake of discussion, if we entertain your thought, that means capitalism also doesn't exist beyond a thought experiment, which questions why it was brought into this discussion? Whether or not a winner-take-all-future is compatible with that thought experiment with no actual grounding really doesn't matter, does it?
Still, if you've ever actually dealt with a monopoly, you must surely understand the difference between that and an actual competitive market right?
Yeah, I know. We see you do that below.
> You should really try working in cloud.
I do. Yet I don't need to negotiate with a police force to protect it, so clearly there isn't a free market. But even if you want to pervert what a free market is by cherrypicking some things you like from it while ignoring everything else, I still don't see how you would ever consider the cloud to be free. It is marred by things like copyright and patents that make free exchange unrealistic.
> Our customers churn all the time because they have meaningful choice, and our margins reflect that
Is this humorous story intended to be an arbitrary tangent or something related to the rest of the discussion? The best interpretation I can come up with is that you are trying to use this story to explain that your cloud activities exist in a market, which is no doubt true, but you haven't given anything to suggest it is within a free market. Even non-free markets need to have choice. You cannot have a market at all, neither free nor non-free, if there is no choice. But there is more to markets than actors being able to choose. People having choice was present in societies even before the idea of a market was invented.
I don't think anyone serious believes this any more. The tech is thoroughly commoditized. There are a lot of plays to be made, like OpenAI cornering memory, or spending heavily on unique data for RL, if you can stop distilling, but there is no proprietary core IP at all.
There is no way to monopolize a technology this big. It would be like monopolizing literacy.
More specifically: when everything you could previously have earned money with is subsumed by AI, the "frontier" of profitable knowledge exists solely with the highest concentration of capital. Because that's where the highest amount of compute and consequently valuable data is.
Already now the idea of the tech being "commoditized" is true only if you exclusively look at the bubble of rich countries. AI-worthy compute is prohibitively expensive otherwise for individuals.
It's almost as if arithmetic were the least important part of mathematics. :-)
Here's the thing: we could bring it back. It had a good 50 year run, it could have another. Lina Khan was working on it, you probably saw the flak. This was on the ballot in 2024 and it will be again in 2028.
The latter doesn't imply the former. In other words it doesn't matter what the reality is, it matters what the excuses for sinking money down the hole are.
> There is no way to monopolize a technology this big. It would be like monopolizing literacy.
Again, I don't know if literacy is monopolized or not, that's a separate and quite long conversation, but whatever the case is, it doesn't mean anything wrt AI because GPUs and RAM are already monopolized and successful market scalping requires monopolization.
The bubble is simply an extraction mechanism, the conditions for it are present now, as are the will and power available to those who can do it and benefit from it. How long it's going to continue is a question with many unknowns but the conditions for it can be maintained for at least another 3 to 7 years.
If, and it is a big "if", someone were to become "the gods of AI", they get to perform a "pivotal act", i.e. take over the world. And if they do it wrong, end the world.
From the public statements, I get the impression that Zuckerberg is saying words solely to give positive vibes and does't actually believe any of that; that Musk might believe all of that but has only given as much cursory thought to it as one might get from reading a Wikipedia plot summary of Colossus: The Forbin Project, all The Culture novels, and an elevator pitch (~30 seconds or so) with Yudkowsky in 2013; others have certainly painted Altman in similar light; which leaves Amodei as a solid "C-" in the sense that while Anthropic still isn't taking this even close to seriously enough, he hasn't overtly said or done things that implicitly or explicitly show he thinks it's just fairy tales to get nerds interested in working for him.
People not only ignore second order effects, they also don't think about what would actually happen when a "winner" really "took all".
While AI is a tremendously useful technology, business types wildly misunderstand it and its consequences. Greed eats brain.
I think in 3-5 years close to 100% of current GPUs/RAM will still be in service. New hardware matters only if gains in efficiency are large enough to put you out of business because the competition gets them. Extreme case of that are bitcoin miners. With AI inference there are way too many factors that influence the profitability. It's not like you can't run new model without the new hardware. It can just run on more of the old one. 3-5 years for GPU are estimates when the main application was gaming with new games with new requirements for running on a single GPU being constantly released.
I wouldn't be surprised if most hyperscalers are betting on cheaper GPUs on their next upgrade through either pricing pressure on NVidia, competition from AMD/Broadcom or in-house chips.
https://epoch.ai/data-insights/ai-datacenter-cost-breakdown
https://www.csis.org/analysis/impact-tariffs-ai-data-center-...
The first one says that networking infrastructure costs 13% and the other says 1.5%. Not a great start.
Let's imagine that a traditional full rack can host 80ish H100s and that each one costs $25k. That's 2 million right there. There's probably only 10x cases in that rack so that's 10x dual core motherboards, RAM, network, etc. Call them 20k each. Maybe you need a 32 port 100gig fiber switch, that's 8k. Let's say it's 2.3 million per rack and 2 million of it is just the GPUs. We're at 85% right there.
I suspect that it would take more than just another 100k per rack to finish out a data center so my "80% of the cost is in the GPUs" is definitely wrong. But I didn't see anything in these articles that indicated to me that 50% of the cost of a datacenter is anything really durable like buildings or electrical infrastructure or cooling. There's still an enormous amount of cost in stuff whose performance follows more of an exponential performance curve, not linear. It's just an absolute nightmare to try and amortize exponential performance curves with linear repayment.
For GB300 series, pricing along the lines of $500k/server, 9 servers/rack ($4.5M/rack) before looking at water cooling, 800Gb RDMA network, and your 100Gb IP network
There’s a risk that likely will happen that once demand is met capacity will overshoot significantly. However, when will that occur is very up in the air. Everyone is training every larger models and there’s more AI usage than ever
If Semi-Analyses is correct [1], even H100s would lose money at current prices - I doubt that is actually true of H100s in practice because from what I can tell it is based on a 3-year depreciation from the new price, but the point is that there is a point at which old hardware's operating inefficiencies dominate.
https://inferencex.semianalysis.com/infx/profit-estimator/de...
why? I keep seeing this over and over again but I've not seen an explanation. I would really appreciate some insight into this line of thinking, since it's not obvious to me. What is the benefit of holding hundreds of billions of computer hardware when it's not proven that it's even needed.
When AI replaces human workers and researchers, the "frontier", with which you can earn money and gain an edge over your competition, resides in those data centers with the highest concentration of computational power.
If you own those data centers, you control the frontier of human knowledge, the edge where money and power is being carved out.
Just simple napkin math.
Now, lets say there are 5 companies competing to be the winner. Only one can be the winner, but all 5 are trying to be the winner. How much should each company individually spend?
Well... starting-bid is $2 Trillion, or ~$10 Trillion in the aggregate. Its basically the waste of competition. Because everyone is vying for monopoly power, the only fair for them to all invest over $2 Trillion in the hopes of being the dominant one at the end. Even if $8 Trillion is wasted, that's "other companies", your competitors, who lose out.
But wait: if one company spends $2.1 Trillion, that one company likely becomes the winner and the other 4 companies lose out. So then another company spends $2.2 Trillion, to get ahead of that war. Etc. etc.
Yes.
> I have so many questions.
I'm not saying this is correct. I'm just saying a lot of other people believe this. If they are wrong then they will lose a lot of money. If they are correct, most of them will lose lots of money but one of them will make money.
I agree with you that the chance of "everyone is a loser" is higher than the current Zeitgeist believes.
The real gamble: betting there is a large enough pool of c-suite that doesn't fundamentally grasp the concept of vertical integration that they're willing to pay to train their competition. Because if you think AI companies aren't training models how to do everything your company does with your data you are deeply delusional.
The real real gamble: praying sovereign governments stay anesthitized to the national security implications of a handful of extra-national organizations operating the conputer security equivalent of WMD with no oversight.
or is the "winner takes all" mentality based on the assumption of being the first to achieve the technological singularity?
Alternatively, it may be based on becoming the favored provider of the defense or intelligence industries.
If you think that, then all this greed and stupidity makes sense.
If not, well, it's just greed and stupidity init.
There's plenty of data indicating that inference at typical API prices is profitable after accounting for the cost of GPUs. However that doesn't account for the arms race of training new models while performing R&D.
Personally, still haven't found cause to alter my opinions, stated here:
- After "AI": Anticipating a post-LLM science & technology revolution
- Riff: LLMs are Software Diamonds
- Halting AI (as in, the Halting Problem, not stopping it dead in its tracks --- a foolish idea).
- Riff: Classifying Tools for Thought (what, relatively speaking, is the intelligence level of LLM tech, in my book... It is intelligence. But what kind?)
Cf: https://www.evalapply.org/tags/ai/
---
[0] Kind of stupid to bet oneself against ten trillion dollars backed by absolutely massive corporations as well as world superpowers, but it takes all kinds to make the world :shrug:
My bad! Sorry for the annoyance.
[1] https://news.ycombinator.com/item?id=50422043
> The hackernews crowd just can't emotionally handle the coming economic disruption.
The “hackernews crowd” has a lot of highly-paid SC wage workers, even venture capitalists (whatever that is about). Personally though I just have a normal job and don’t want to be disrupted out of how I earn my living. Now you can go on and gloat. Gloat about how you think that I will become economically destitute. This is an open invitation.
https://openai.com/index/building-ai-infrastructure-with-the...
Yeah, Open AI and Anthropic may be buying the bulk of their compute from other providers but they are now moving into building their own infrastructure directly. There is a scale beyond which it makes more sense to build vs. rent
>As of and subsequent to September 30, 2026, we have entered into purchase commitments and lease agreements, including short-term commitments, as well as leases not yet commenced, with an aggregate committed spend of approximately $895.4 billion. These commitments include approximately $447.5 billion expected to be incurred before December 31, 2030, and substantially all of the remainder is expected to be incurred between 2031 and 2045. Approximately $69.8 billion of these commitments are due in the 12 months following September 30, 2026. A significant portion of this total committed spend relates to Anthropic Direct, and we expect it to represent a growing share of our compute investments over time. The flexibility from Anthropic Direct allows us to backfill the short-term compute we procure during periods of high demand with more cost-effective, long-term compute as our dedicated infrastructure comes online.
So they decide to change the way they procure compute (rent vs buy) for which they will have to raise a little shy of $1T in the next 20 years, with $70B due in the next 12 months (and $450B in the next 4 years). This adds significant risk because they have not only to execute the build-up (which they haven't done before) but also raise the funds to pay for it, all while continue to operate and grow the core business.
> A typical one-gigawatt AI data center requires $38 billion in up-front capital expenditure (CapEx) and $0.9 billion in annual operating expenses (OpEx). If CapEx is annualized over each asset’s lifespan, the total cost of ownership equates to $8.5 billion per year. Servers dominate this cost at $5 billion per year, or 60% of the total. Operating costs are small by comparison: even energy, the largest OpEx category, costs only $0.6 billion per year.
They get bulk discounts because their load is 24x7, so the fixed costs covered by a per kWh charge is spread over many more kWh than a residential customer
For DCs doing training, their load is constant as long as training is happening. But as soon as they end a run, their load can disappear quickly. This is actually a problem, as the grid and generators are not designed to see massive amounts of load go away instantly
Several system operators are working on new guidelines about these large loads, and they're working on ways and regulations such that these loads have to ride out grid instability and not just disappear without warning
- 20% Capex has ~20-year lifetime: The shell (land, building, cooling gear, rack mounts, power lines etc.).
- 80% Capex has ~3-year lifetime: The computers, including bulk commitments to power and maintain them.
And interest rates are why this split is a ticking time bomb. The sheer quantum of investment to build one is almost entirely debt funded (~90% of total outlay). I would guess the endgame would be bad even if the split for these entirely different asset classes is 50/50, which would be very generous, based on what I've understood.
Datacenter funding is extremely leveraged. For comparison, only some banks are allowed that much leverage for that size of capital, that too only if they have federal reserve backing. Even in such a case, the retail account holder (pseudo-investor, because our deposits are loans to someone else) has practically no downside protection. If the bank collapses you will lose everything.
In fact, to focus on the short lifetime assets alone, this more recent link suggests servers are 50% of CapEx, and GPUs are about 25 percentage points.
https://cib.bnpparibas/ai-infrastructure-investment-2026/
So already about 50% of CapEx spend is one-off construction / installation with a very long lifetime.
That leaves servers and GPUs. There is a ton of evidence out there that even chips from multiple generations ago, ie way more than 3 years old, are commanding pretty high rates. So even if they are replaced by newer chips every year, the older chips still hold a lot of value and can’t simply be written off to 0.
I think this is the most comprehensive look at the AI economy so far, probably a better source than the rest: https://www.exponentialview.co/p/the-state-of-the-ai-economy — it suggests everything is walking a tightrope but still sustainable for now.
Like I said, it's a heuristic, and the error bar on that opinion is... "50/50 split is also bad", for debt financing reasons. I'm skewing my guesstimate toward a pareto-split bear case, because 50/50 is the rosy-eyed best-ever case being proffered by parties who need the story to remain believable. I doubt anyone has a true grasp of the opaque financin.
In any case, I'm trying to reason from first principles.
A) The sheer quantum of debt financing skews the whole US economy, and that is a hard thing to skew. This means debt will keep getting repriced, higher interest rates, because risk exposure to financiers increases non-linearly when said risk becomes industry-wide.
B) Compute and power become inelastic, at the scale and rate of expansion of LLM datacenter. The become highly contested practically fixed resources. Prices will go up, and they are likely to go up faster with every marginal increase in demand.
C) Far better informed parties have been warning about severe under-quoting of liabilities for a while now. e.g. Some of the resources informing my opinion include publishing by Asad Ramzanali / Vanderbilt Policy Accelerator and Michael Burry / Cassandra Unchained. Worth reading their stuff:
After the AI Crash, by Asad Ramzanali, Vanderbilt Policy Accelerator (March 2026): https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6476159
Michael Burry Warns: AI Infrastructure Depreciation Is Masking Billions in Costs https://nova.kapualabs.com/goog-sys_onboarding/michael-burry...
(To be clear, just because Burry was spectacularly on the money that one time about CDOs, doesn't automatically mean he is right about this, but he does have a habit of calling out bullshit as he sees it, particularly dodgy financial engineering of the kind these people are doing... Corroborated by others like Asad Ramzanali.)
I skimmed through the "After the AI Crash" paper, and it is really not an analysis but a policy brief that mostly reiterates, and uncritically accepts, a lot of the large numbers thrown about in other media sources. There is no analysis of the demand side and cash flows and timelines involved, and spends a lot of time on the circular investment issue with any attempt to unravel it.
The exponential view analysis is much more methodical and finer-grained in its approach, and it addresses a lot of the aspects you mentioned, including prices and demand.
As for whatever Michael Burry has said, I've looked into it a bit (as an outsider, no inside knowledge) and I don't quite buy his thesis. There are indications from too many different sources that GPUs are economically valuable for much, much longer than the timeline he assumes.
A good thing to Google AI Overview would be "How many market crashes has Michael Burry predicted?", and wonder why all those other predictions didn't get the airtime this one is. I have a theory: https://news.ycombinator.com/item?id=50421245 ;-)
Apropos schadenfreude, I don't have a dog in the fight---I don't even actively use LLMs, because they don't work for how I'm wired.
I'd like to have a future-dog in the fight though. I want the LLM-hyperscaler bear case to happen because I want access to dirt-cheap compute, which I hope will be the aftermath of a financial failure of the hyperscaler junta. What will be left is industrial capacity to make tons of chips and computers, which I hope translates in to dirt-cheap retail compute. eCoin-world is unlikely to absorb all that because mining unit economics and coin volatility are both significant enough to make people consider the bet.
And like, LLM tech is quite cool and useful and all (ignoring for the moment, the planet-scale piracy by the hyperscalers), but the other "normal" stuff is so much more so.
As for the odds of them making a God, I'm taking an inverse Pascal's wager, as I wrote here: https://news.ycombinator.com/item?id=50427160
I would like a world of cheap compute too really, but I fear at this point it could only happen if all AI demand suddenly disappeared and never came back. Given how fast it is growing and how far it still has to grow, compounded by the limited rate at which new capacity is being built, it seems unlikely we'll get to an overbuilt state for an extremely long time... unless, like you said, there are dramatic breakthroughs in computation efficiency.
Hacker News increasingly feels like being a horse in 1916, reading in the Horse Bulletin that Ford is a scam and a bubble because the mass-market automobile will soon be commoditized.
There's an object-level claim to consider, but behind it the more baffling implication that this would be good for horses.
Also, inverse Pascal's Wager, because horse v/s motorcar is a false dichotomy... they are bound by the same laws of physics (faster horse, less shit, more noise and more smoke), merely different economic realities.
Hyperscaler bet is "Superintelligence" and/or "AGI", or we will make a God. That is a technological singularity. There will be a "before God" and "after God" world, and we will very likely never be able to understand what the AGI/Superintelligence's world looks like. (Because to understand what it is like to be a bat, you have to become the bat, but at that point your meaning of "understanding" itself changes entirely).
So I believe ten trillion dollars and a bunch of pirated books and source repos, and well pirated everything, mostly of very dubious quality and provenance, will not result in a God. I've been acting accordingly.
Right, the implication is that the global economy will continue to spend $30T/year on human cognitive labor, but spending, what $1T a year for superintelligent AI will be too much to swallow.
Cars will be so low-margin and commoditized that car companies go bust, and everyone will stick with horses.
Indeed, the entire car business first boomed, and then has continued to exist only because oil is massively subsidized. If they suddenly started pricing in the fully-loaded cost of oil into retail prices, the cheapest way to own an all-electric (forget IC engine) would be to drive it home from the shop, park it in the driveway, and set it on fire.
Come to think of it, Happenings in Hormuz is doing some of that to everyone now.
Detroit is not the city of dreams it used to be. Hasn't been for a long time.
Datacenter subsidies will burn out faster. What will remain is industrial capacity to manufacture really _really_ fast computers, at a huge enough volume, to become "commodity android devices for the common person" --- I'm betting on that.
Fatal misunderstanding.
Datacenters are not warehouses. Amazon.com being borderline-profitable owing to massive ongoing capex on warehouse expansion is not at all like AWS becoming borderline-profitable owing to massive ongoing capex on LLM datacenter build-out and SOTA model training (it's CAPEX, but carries OPEX affect because of GPU depreciation schedules and competitive hardware upgrade cycles), and opex of delivering LLM inference services, and being able to price it all competitively.