Every single company in the tech space has negative cash flow. That is they're spending more money than they're making. A lot of it. So how they'll answer the margin call? If this is starting to sound a lot like the subprime mortgage crisis it's because it very much is, but with datacenters instead of houses. 7/
This is where we have to thank Ed Zitron for his excellent reporting on the finances of these companies and his remarkable work in connecting the dots whenever some CEO talked bullshit to investors while the rest of the press was looking the other way. This is also why the Gamers Nexus investigations about GPU smuggling where so important: the demand could not slow down, by all means necessary, so a little illegal smuggling on the side is OK, say a few tens of billions of it. 8/
And this is when we start talking about the tech itself and why it is unsustainable. Every single company in this space is losing money, every AI startup, every AI division in an existing company, everybody. They're losing money because the costs of running an LLM - inference - are large, and it's unclear if anybody has been profitable there. But even if they are nobody has been able to cover the escalating costs of training LLMs, which is of course a prerequisite to running them. 9/
Training LLMs requires a gigantic amount of hardware, brute force calculations, storage, memory and power. Enormous amounts of data whose scraping is not being paid by the companies involved, it's a cost they've spread over the entire internet. As models get larger training gets more expensive. But even as companies sink more money on training LLMs have already reached a plateau in term of functionality and in some cases are degrading in a process called model collapse. 10/
Let's talk statistics for a minute to understand model collapse. Say you have a bunch of measurements of something and you want to build a mathematical function that will tell you what values appear in the measurements and how often they appear, this is what is called a distribution. A distribution can be used to reproduce the material it was created from, but the output will not be the same because something was lost in the process of making it. The less frequent samples, the stuff that lives at the fringes of your data set, the details. 11/
Now let's say you use your distribution to generate a new bunch of data: it will look a lot like the inputs you used to make it, but it won't be the same. And if you try to make a new distribution out of that data it will be distorted. You'll lose more information and for every iteration the outputs will look different, they will be more uniform, they'll contain less details, like when you recompress a JPEG file over and over and over. 12/
And this is exactly what happens when you train an LLM on its outputs, because an LLM is at its core a distribution of human language. It's also the reason why LLM vendors started putting invisible watermarks in LLM outputs. They need to filter out the slop they make from their training set. But even then some slop will make its way in, and even the slop that gets filtered out is still an additional cost: you scraped it, you processed it to check for your watermark, and then you threw it away. 13/
As the internet gets sloppified the chances of keeping slop out diminish, and the training costs increase. And don't forget that all this training has been done without consent, attribution or compensation of the people making the training materials. This space will be regulated eventually, and when it is that will also make the training costs increase further. 14/
Which leads us to the hardware itself: this is the point where people will tell you that right now things are looking bad, but the prices will come down because the hardware will improve, and so you'll be able to make powerful local models and whatnot. But here's the thing: hardware improvements are over, in fact they've been for a while. Moore's Law, Dennard scaling, every single factor which lead to faster and cheaper hardware has ceased, years ago. 15/
There's no magic hardware improvement around the corner, quite the contrary. Every new silicon node comes with smaller and smaller increases in density and performance over the previous one, but at significant increase in costs. A fabrication facility to produce state-of-the-art hardware takes months to go from a silicon wafer to a working chip, using machines that cost billions. 16/
In other words the only way for a modern high-performance silicon process to be profitable is to produce a large amount of chips that will be sold at high price. See how we're coming back to the artificial scarcity where this all started? To keep the semiconductor market going we need more demand and AI represented the largest demand of silicon ever, but none of that demand was real. So when the prices crash the demand evaporates and the foundries will be left with a lot of idling capacity. 17/
This will cause a huge amount of capital that was poured into the tech sector to shift elsewhere and this is a good thing. Climate change is weighing down on us and on our future as a species, all the resources we have should be concentrated in fighting it, not chasing a technological fad inflated by a bunch of scammers and mega-rich buffoons having mid-life crises on commercial social media. 18/
I'll wrap this up with an appeal to the people who have been using LLMs extensively in their work, often becoming dependent on them: this is the time to come off this train before it's too late, though it might already be. Nobody is making any money with these things, but they made plenty of debts, so you'll be left holding the bag for the entire tech sector since you're the only ones paying for them. Over and out. 19/
@gabrielesvelto Thank you so much for this explaining the financial aspect in a way I actually understand. I read Ed Zitron but really struggle to follow all the jargon. Much appreciated!
@aurellence you're very welcome! Glad you found it useful
Thanks for the comprehensive summary.
Nov 14, 2026, 6:13 PM@gabrielesvelto Can you help us create places that financially support those of us who want to stop using them? Will write code for food….
@gabrielesvelto thanks for this thread, I follow this stuff a fair amount but the potential catastrophe of mark to market wasn't something I had considered
@gabrielesvelto Thread of the day - this goes to eleven
@JackMexa4 you're very welcome, I'm glad you found it informative 🙏
@gabrielesvelto Thank you, excellent thread.
@gabrielesvelto Good read, thanks for posting.
I can't help but think they'll socialise the loses the same way as 2008, except its a whole degree of scale larger.
Fun...
Nov 14, 2026, 9:10 PM@inpc @gabrielesvelto They're already positioning for bailouts back home.
@gabrielesvelto Thank you. Excellent thread.
I don't want to sound negative, but I honestly doubt they won't find something else to grift on.
All these capitalists just want to siphon money from the majority for the least effort. Building on renewables and stuff like that leads to consumer independence; it's like expecting a drug dealer to sell healthy things like veggies and bandaids.
The GPU grift went from gaming, to crypto, to "lockdown scarcity", and now to LLMs.
Nov 14, 2026, 9:46 PM@gabrielesvelto I do disagree with your point in post 19 — I think it is possible to make money with probabilistic pattern modelling or statistical sequence prediction in a variety of technical and other fields, perhaps not at the scale and pricing or with the specific products that the famous names are offering today. The invention is out of the bag, it will continue to be refined and improved.
@andyd machine learning can be profitable and will be again in many use-cases, but large language models specifically are not going to be part of it IMHO
Nov 14, 2026, 10:16 PMThis is why it’s so important to retain authorship over the things you produce. If you’re using LLM’s to vibecode or otherwise replace the actual work of understanding what you’re putting your name on. You’re reliant on a system that’s precarious at best.
I do wonder what the fallout of this will be in terms of existing / trained models. I’m sure these companies will want to milk them for as much value as they can. With slower progress, what will the cost of providing existing models be?
@gabrielesvelto
I, for one, would bet my bippy that the tech bros, hedge funders, data center entrepreneurs, and Ai evangelists are now spending as much time working on ways to unload their debt onto the people who WEREN'T responsible for the Ai bubble as they used to spend creating the Ai bubble. Look out Mr and Mrs taxpayer, the "smartest people in the room" are coming for your wallets and nest eggs.
#Ai #FinancialBubble #DataCenters
@gabrielesvelto I'm not saying this is fake because this follows my experiences and light economical knowledge, but do you happen to have some sources for the LLM bros that'll brush this off?
@gabrielesvelto assuming it doesn’t get poured into other tech shit or something else equally bad or at least not positive for the environment
Nov 14, 2026, 9:10 PM@PaperMartin @gabrielesvelto Count on it.
Resources? Being used rationally? To help ordinary people?!
Never. Not a snowball's chance in hell. A lot of poor people will lose everything, the richest rich people will become even richer, and then when everything stabilizes there will be another scam.
@gabrielesvelto in america, it's literally heads I win tails you lose.
Rich people have played both sides and that's why they end up on top
Some of us have lived through multiple catastrophic economic events that have not only domestic but international impacts, who find a way to survive a recovery period lasting *years*, then watch the process happen all over again, with, like you said, a new scam which will inevitably crumble.
I hate how indifferent, gullible, apathetic, and/or conditioned into accepting this is normal and reject any alternative.
@gabrielesvelto preach 🙌
Or cheap chips?
@gabrielesvelto Well, I would add an asterisk here: there's still hardware improvements, just not cost-effective ones. And we're almost done with CMOS, we're not expected to get improvements on lithography past high-NA EUV, and the last trick left is CFET gates. After that, technology stagnates, likely forever, no matter how much money one has. I don't see hardware design and synthesis improvements making up for it either, definitely not in a transparent way that doesn't require rewriting code.
@xerz yes, very true. And because it's become a margins game there's competition from the existing processes that have been already paid for. If power is not an excessive constraint it might be more cost effective to use more area on an existing node than switching to a new one for a smaller chip. And let's keep in mind that the semiconductor sector has also received significant public subsidies pretty much everywhere. Will that last? Because if it doesn't they'll also have to stand on their own legs.
@xerz
It's past time to say "this is good enough". We've leaned on faster silicon to cushion the impact of less efficient coding for decades. LLMs are by far the least efficient way to do *anything*.
Let's maybe go back a few steps, actually increase process size to cut leakage current, and start coding again as if we don't have access to infinite processing power.
Leaner, meaner, less power-hungry systems.
Nov 14, 2026, 7:18 PM
@gabrielesvelto Not only is the hardware not improving: AI has diminishing returns. To get incrementally better result, you need exponentially more processing power. So far we've seen huge progress because the biggest company in the world were able to purchase a huge chunk of the world processing power. But that doesn't scale. Even if the hardware improved it would need to improve exponentially, and that's not possible.
Nov 14, 2026, 7:01 PM@gabrielesvelto
This is a really comprehensive assessment of the AI tech sphere. Thanks! Re the model collapse based on regurgitating their own output, I think they are banking on expanded surveillance to collect even more details. Recent examples coming to my mind are where Open AI scooped mathematicians by collecting all the mathematicians careful work and announcing the proofs right before the mathematicians had completed their thorough documentation for publication of same proof.
@jon_harbert they have been deliberately moving into new areas precisely because of model collapse. All the hype around programming and math is about two fields that - until recently - were largely slop free. It was a treasure trove of well-labelled, well written training material. And now it's a mountain of slop too.
@gabrielesvelto I think this problem has been underestimated. I recently read an obviously AI-generated article online (ranked highly in search results) that had obvious errors. Now AI models will scan that POS and amplify the nonsense.
@gabrielesvelto not to mention the MASSIVE server costs for all those being actively scraped.
Sidenote: crap sites are incentivized to try and remove any watermark from generated content because it means that they will get preferential placement in any result set. This means that while the big engines are fine generating the content, smaller sites will use either mixers or a custom "fogging" engine that will either obscure or remove those watermarks from the generated content.
And thus the "Pioneer Models" learn the same lesson as the rest of us about "how you can't have nice things."
@jrconlin that is very true. Most applications of LLM that produce content available publicly and are profitable pretty much require one to launder the outputs as genuine. That's a financial incentive that goes directly against training costs.
@gabrielesvelto I said this would happen! That AI would eat itself, many years ago in fact.
So funny it's happening.
@gabrielesvelto My favourite example of generation loss is Alvin Lucier's I Am Sitting in a Room. https://www.youtube.com/watch?v=fAxHlLK3Oyk
I think it's useful as a metaphor for slop eating slop because as the signal fades, it's replaced by the artefacts of the system: the natural resonances of the particular room and equipment used. The way it smooths everything out until it matches itself is reminiscent to me of how LLMs "average" out text.
(I also think there's something to be said about disability, since Lucier is doing it in part to smooth out his stammer. I don't know if anyone's done this, and I don't believe I'm qualified.)
Oh man, for somebody who has background in the video business this resonates, hard. When you have digital video material that is subject to lossy compression, the video quality after a fed generations of editing can become quite atrocious
I just want all of them to be stuck into the matrix of their own making. Make them somebody important.
@gabrielesvelto A copy of A copy of A copy ...