
Now what? After the AI crash
Six weeks ago I predicted the AI crash, today as I write this the American stock markets slipped down an ever steeper slope into that same crash. The peak for Nasdaq, the major tech index, was just one month after predi…
以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。
Sora Thompson
Oct 30, 2026
Share
Six weeks ago I predicted the AI crash, today as I write this the American stock markets slipped down an ever steeper slope into that same crash. The peak for Nasdaq, the major tech index, was just one month after predicting the crash.
Well first the allure of infinite future profits companies waved in front of everyone in order to justify piles of debt so historic that the entire computer industry is sold out of everything for years in advance despite jacking up prices to unprecedented levels is going to hit slowly. If we remember back to the bubble diagrams, the first thing to is the funding from investors no longer putting money into the bubble. Now that such is gone companies will run on inertia until their unprofitability crashes them out. So the big debt bankruptcies are still ahead for anyone looking forward to that
But that’s the boring part. How long will the downslope last? According to economic research, about 11ish, very rough window, months of riding a recession.
And what will the future look like after? Well just go back to the previous post to see. IT departments are ready with plans and even hardware to supply their own AI instead of hiring it out if they’re in a big enough firm. Therein Chief Financial Officers that can finally stop listening to their delusional CEOs and start trimming budgets. And as long as they’re not part of the hyperscalers themselves, but still have enough scale they’ll find giving full time devs an expensive laptop or running their own GPUs less expensive.
Speaking of computer hardware, after the “Neoclouds” such as CoreWeave go out of business, and they’re on the way, giant piles of equipment will be available for cheap. Companies with actual bank accounts should be able to pick up their own AI computer hardware at a good discount relative to today’s peak of prices.
So to reiterate: going forward, what’s the future? Well that’s diversity. Your phone might run a model that tells you when to leave for that appointment you have because it checked the traffic for you. Your laptop might run a model that cleans up and masters the audio of the podcast you produce. Your neighbor might be a programmer running their companies codebase through a company owned part of a datacenter, because they don’t trust a commercial clouds not to use that codebase for training and so leak it.
Big, for hire datacenters will still exist, many have already been built after all. And there will still be reason for many to want them. Your neighbor that asks ChatGPT everyday question and pays $20 a month to do that will still be “in” to do that for some model or other. But the massive profits that have been deluded over these datacenters, and over these AI model makers, won’t materialize, they never would have.
For AI model makers, well we already covered that. Yes programmers “prefer” to use the latest and greatest models. But CFOs prefer not to pay an 80% profit margin to those model makers to do so, and it’s the CFOs that hold the checkbooks. A model that can do 95% as well, and costs 70% less (paid per “token” or just run on company hardware) is just what major corporate users are going to have to “suffer” through.
For datacenters: An open market for “compute” in the cloud is already under investigation, with analogies that compute is “oil” and thus a fungible commodity. This would already be worrying to data center companies in massive debt: if they don’t actually control the supply of this oil, being nothing more than middle men traders who can be swapped out in an instant for an even fractionally lower bid, then what sort of leverage and thus profit could they possibly have?
But compute is not “oil” by any stretch of the imagination. Computer chips can be used over and over again, sometimes for decades. Computer chips are always getting better and we’re always getting more of them. The only hold on supply today is a temporary one at best from an unprecedented spike in demand fueled by what one of the steepest spikes in debt in world history. Once the demand subsides the infrastructure being built to meet this demand will still exist, and memory makers like Micron will be right back into the boom and then oversupply bust cycle they swore up and down they wouldn’t be getting into this time.
Computer chips are made of silicon, literally the most common element on the earth’s crust. There’s enough supply to make the entire surface of the planet into computer chips. There’s no long or even medium term shortage of basic ingredients like there is with oil, it’s not even possible. There’s just a temporary manufacturing chain blip that will be forgotten. The chip blip is ending, the slide back towards a steady drumbeat begins. That’s, “now what”.
正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。