[adampunk]

I think we are *for sure* in a bubble. There is a kind of venture-capital driven craze for compute because...well a whole host of silly and sensible reasons. It's hard to talk rationally about the money being thrown around if you have suspicions that billions of it are being thrown around because some dudes think they'll create god in the machine and own it, or that they have some theory of consciousness tied up in the matter. Animal spirits, in a sense.

Some of the money thrown around is by companies that want to lock in some kind of interdependency because the only downside they can conceive is Oracle or Dell or whoever invents AGI and they get left behind. So huge circular deals are getting made which increase the correlation coefficient for any collapse to 1 lmao. These deals are getting made for reasons that feel like 2004-2005 american real-estate, where the downside risk of a national profile of mortgages was actually (not joking) taught in textbooks to be 0. So naturally if you're maximizing revenue by making things interdependent, you really only consider the upside.

All these forward looking energy contracts and local generation of energy are signs that the market is under strain more than we expect increasingly exponential future use. Giant companies are thinking they're locking counterparties into the right risk structure (here with an energy company being somehow willing to foresake this infinite future energy price they could charge by just waiting), but really that energy company is perfectly happy to accept some money to start a project which will generate revenue long before electricity. That energy company has an idea of its own risk and revenue profile and they can extract money from a bubble, too.

I give us...months? Maybe 18 months--probably less--before things things get really nasty and messy for the firms who thought they were buying a golden ticket. It then gets messy for users (if not sooner) who are right now being subsidized nicely--not the ludicrous 5k being thrown around recently but compute is at a subsidy right now, so long as you want to rent it or can run a model like Qwen locally. Lots of other people are paying for that subsidy while extraordinary amounts of money flow from one part of computing to another. That's already having weird consequences as firms who spent money on what is essentially rental compute pushing their employees to use more of it in order to keep the person who made the contract safe. More companies will do the Microsoft route (no I'm not talking about Copilot!) and try to push tasks into their internal pipelines like MSFT does with Azure--where with e.g. github, what github needed to do took a back seat (literally haha) to integration with the compute pipeline. That's good-bad-whatever, depending on how you want to think about it. But it's certainly disruptive and I think right now a lot of genAI is doing that kind of disrupting, where people and orgs are being forced through money shaped holes.

I don't know what happens when the music stops. I just know that it is playing.

[bronlund]

I don't think that many will disagree with you on this. If we compare it to the dot-com bubble in 2000, just as speculators then tried to get in early in this new thing called Internet, today they are trying to capture early market shares in this new thing called AI. But just like then, when this bubble bursts, the technology that everyone was chasing, will still be relevant. Today the internet is regarded as critical infrastructure.

And I do think you are underestimating just how much money these guys can print, if some event is disrupting the machine. If this thing really goes down, it will be by design and because they got the Thing 2.0 ready to go.

[adampunk]

The problem I have with this analogy is it cuts the other way. His tempting to look at the dotcom bust and think of pets.com or some other venture where somebody lost their shirt. It gets a little harder when you expand it to companies that today own an enormous slice of the world, because they bet on the Internet. Most people lost in the dotcom boom; Amazon, for instance, didn’t. The whole premise was that if you moved soon enough, you could buy in on the ground floor—-that was emphatically the case if you bought Amazon. Now they own a huge almost inextractable chunk of the Internet. It’s quite hard to do business or pleasure on the Internet without somehow involving Amazon.

What I am saying and what I think a lot of folks who are trying to get this point across are saying is that this will be critical infrastructure sooner than 20 years. The right frame of mind is to look at this like we look at the Internet circa of the 1980s or the 1970s. This is big messy and experimental right now. We are in the middle of rebuilding computing with foundation models. That is happening at an enormous subsidy for the time being.