"So are we in an A.I. bubble? It sure looks like it to me. That doesn’t mean we won’t get large economic advances (and disruptions) out of A.I. "
This is the most plausible looking path forward: LLMs + conventional ML + conventional software inverts how our economy operates over the next few decades, but over the next few years a lot of people are going to lose a lot of money when the singularity is actually a sigmoid curve.
I suspect there is a substantial first mover disadvantage right now. The extreme investments will not be profitable, leading to the bubble bursting at some point in the not so distant future. This will lead to short term price increases for inference and slower innovation in a period, the tech will emerge more mature and stable etc.
As one who clearly see the huge potential of this tech this is an interesting outlook; make sure to make your products resilient to changing vendors and price hikes and it will probably be fine.
Side note: Google seems to be playing the long game..
I think any comment about an AI bubble needs to start by defining who the players are and how it affects them differently.
AI foundries, Nvidia, the hyperscalers, enterprise buyers of AI, consumers, the US, China, the rest of the world, startups, investors, FOSS, students, teachers, coders, lawyers, publishers, artists... each stand to win or lose in profoundly different ways.
> Even my moderate view, though, is premised on the assumption that A.I. will continue to advance up the steep part of the sigmoid curve for a while before hitting one or another physical constraint that creates a new inflection point and slows advances further.
I feel like we've pretty much already hit a fundamental barrier in compute that is unlikely to be overcome in the near future barring a profound, novel algorithmic approach or an entirely novel computing model.
I do feel that AI has been overhyped a bit for now, but what happens when we scale our electricity and GPU production 10x, 100x, go nuclear etc and can 100x AI models - lets see. Its too early to tell really.
Effectively everyone, including Sam Altman [0], is saying we're in a bubble.
The only questions left, the only ones that matter, are:
- When is it going to pop? Tomorrow? Next year? 2030?
- How hard is the crash going to be? Only a bunch of AI startups and one or two of the big "AI" companies (OpenAI, Anthropic) go down, or a global financial crash that wipes out hundreds of companies and hundreds of thousands of jobs.
- What's left of "AI" tech in the wreckage. Once the hype is over, what real use-cases exist?
IMO there seems to be more consensus among even the biggest hypemen that this is a bubble. But to protect their bag (and their LP money) they have to find some way that the silly investments they made weren't totally irresponsible.
So here's the line of thinking we'll see more of:
"Yes it's a bubble, but so were the railroads, and yet plenty of people made out big time! The railroads themselves were left over and hugely valuable!"
Then the slightly more astute observer would say, well hold on, that's not quite analogous because the depreciation on AI buildout is way faster than in railroads.
Then the even more astute observer would say, even that downplays the stupidity: the actual value creation from railroads was in the land itself.
There is no analogous dynamic in AI-land, and so will probably be far more broadly catastrophic to the bubble blowers.
The definition I use for a financial bubble in case of AI is the following: financial debt obligations/investments that can not be repaid/serviced w/ future growth. Unlike pure software, the growth in AI is coupled w/ huge investments in real infrastructure (data centers, power plants, network interconnect, etc). This means that regardless of what happens the infrastructure is not going away & given how software can be so easily reprogrammed these days there is no way that all of it will suddenly go up in a puff of smoke. OpenAI & others will figure out how to redeploy the infrastructure for different use cases & keep the money flowing. This is unlike office buildings & general real estate b/c it's not like an office building or a family home can be quickly repurposed for something else to keep the money flowing to whoever owns the underlying infrastructure.
I don't think it's a bubble. The numbers seem large but that's b/c the underlying infrastructure costs a lot of money & unlike other forms of infrastructure computers can be used for all sorts of different things by simply redeploying different software to it that consumers will find compelling (even if it's no longer 6 second clips of cats doing backflips from diving boards).
I have a simple rule. It's too cute by half but its track record has been good since I've deployed it.
Markets can not be forecasted, therefore if everyone is predicting a downturn, the downturn will not come. There needs to be some ambiguity, a kind of FergusArgyll uncertainty principle
I think in the future we'll see the productivity boosts negated by more burnouts. Sometimes meaningless work is OK to recharge the brain.
That is if the work produced is actually useful, and more and more do we see that unless we're hyper-specific, we don't get what we want. We have to painfully iterate with non-deterministic stuff hoping that it gets it right.
My concern is that the "Magnificent 7" stocks are heavily into AI (except Apple) and because of their huge market caps, the most popular etfs have a large component of Mag 7. Even the "global" Vanguard fund is about 20% Mag 7 (1), for example (and the next one on their list is TSMC). The Vanguard S&P 500 etf is 36% Mag 7 (2).
If there's an AI pop, a most people and most huge funds will lose money, driving liquidity out of the global financial system.
I've started investing into a mix of Europe etf and global etf, instead of just the global etf, because the global ETF is so exposed to US computer companies.
15 comments
[ 3.3 ms ] story [ 35.3 ms ] threadThis is the most plausible looking path forward: LLMs + conventional ML + conventional software inverts how our economy operates over the next few decades, but over the next few years a lot of people are going to lose a lot of money when the singularity is actually a sigmoid curve.
As one who clearly see the huge potential of this tech this is an interesting outlook; make sure to make your products resilient to changing vendors and price hikes and it will probably be fine.
Side note: Google seems to be playing the long game..
AI foundries, Nvidia, the hyperscalers, enterprise buyers of AI, consumers, the US, China, the rest of the world, startups, investors, FOSS, students, teachers, coders, lawyers, publishers, artists... each stand to win or lose in profoundly different ways.
Otherwise we all end up talking past each other.
I feel like we've pretty much already hit a fundamental barrier in compute that is unlikely to be overcome in the near future barring a profound, novel algorithmic approach or an entirely novel computing model.
And if they aren't, then they will be soon enough.
The only questions left, the only ones that matter, are:
- When is it going to pop? Tomorrow? Next year? 2030?
- How hard is the crash going to be? Only a bunch of AI startups and one or two of the big "AI" companies (OpenAI, Anthropic) go down, or a global financial crash that wipes out hundreds of companies and hundreds of thousands of jobs.
- What's left of "AI" tech in the wreckage. Once the hype is over, what real use-cases exist?
[0] https://www.cnbc.com/2025/08/18/openai-sam-altman-warns-ai-m...
So here's the line of thinking we'll see more of:
"Yes it's a bubble, but so were the railroads, and yet plenty of people made out big time! The railroads themselves were left over and hugely valuable!"
Then the slightly more astute observer would say, well hold on, that's not quite analogous because the depreciation on AI buildout is way faster than in railroads.
Then the even more astute observer would say, even that downplays the stupidity: the actual value creation from railroads was in the land itself.
There is no analogous dynamic in AI-land, and so will probably be far more broadly catastrophic to the bubble blowers.
https://share.google/aimode/49jtBuQy9X3wSKy5R
I don't think it's a bubble. The numbers seem large but that's b/c the underlying infrastructure costs a lot of money & unlike other forms of infrastructure computers can be used for all sorts of different things by simply redeploying different software to it that consumers will find compelling (even if it's no longer 6 second clips of cats doing backflips from diving boards).
Markets can not be forecasted, therefore if everyone is predicting a downturn, the downturn will not come. There needs to be some ambiguity, a kind of FergusArgyll uncertainty principle
That is if the work produced is actually useful, and more and more do we see that unless we're hyper-specific, we don't get what we want. We have to painfully iterate with non-deterministic stuff hoping that it gets it right.
If there's an AI pop, a most people and most huge funds will lose money, driving liquidity out of the global financial system.
I've started investing into a mix of Europe etf and global etf, instead of just the global etf, because the global ETF is so exposed to US computer companies.
(1) https://investor.vanguard.com/investment-products/etfs/profi...
(2) https://investor.vanguard.com/investment-products/etfs/profi...