I also believe it’s a bubble but how can it be a bubble if everyone thinks it’s a bubble? It seems like bubble is the general consensus but there hasn’t been something like a “pop” yet.
I don't see how it can be a bubble (or at least a risky one) without much debt. Places like Google and Meta profit $100B a year from operations and are funding most of this from cashflow. The risk is that they have to write down a bunch of overbuilt data center assets in a few years, but lenders won't be going under because there aren't any lenders involved. And without lenders being at risk, the system isn't at risk.
*There is some leverage. Coreweave has borrowed a bit for example. But none of this is really systemic and no one is levered to the eyeballs in the AI space.
There is a bit of nuance to the cashflow. That $100 bn cashflow was used for something else before AI: mostly share repurchases and M&A. Now it's being redirected to capex. That removes some of the support for the stock prices; there's no longer a multibillion dollar bid every year for GOOG / META / MSFT etc. stocks.
But you're right that this shouldn't affect lenders, unless we see a lot more borrowing (which is coming, BTW: ORCL and GOOG just issued $10+ bn debt each for AI data centers).
I guess the trillon dollar question is what might be the signs that we're at the top of this bubble.
Looking at a couple of prior bubbles ...
In the subprime housing crash of 2007/2008, lending practices got so bad that there were low-doc and no-doc loans (aka "liar loans") where you could get a mortgage without any evidence of being able to repay it.
Immediately prior to the dot com stock market crash of 2000, EVERYONE was talking about how much money they were making. At work it was insane with people monitoring their stock portfolios all day, reporting how much money they'd made that morning, etc. Tons of people were beyond "all in" - buying on margin.
So what about now with AI market caps, infrastructure builds, etc. Are we in the bat shit crazy "all in" stage yet, or can it conceivably go further?
As the article surmised at the end; AI is most definitely not free (and infact is quite costly) but major players are subsidising the price and cost of tokens to developers, such that right now everyone is pretty much getting AI tools and processing power for free (or very very low cost).
The signs are many products and services adding AI cause they can, but there being very little logic as to how much this addition makes sense. In a way this is the artificial demand they can point to for their (untrue) ongoing growth narrative - look companies are still adding more and more AI.
Let’s take the example of Figma, which is listed on the market after a successful IPO. At its core it’s a very nice design tool that bought the innovation of collaborative live editing (ah la Google docs) to an adobe illustrator like product. It’s still great for this.
But in recent times much development has gone into their AI and MCP features, these likely bring some value; but the market sees these are offered features built like any other feature; when in fact there is a pay-for-use baked in that is being massively subsidised (on the basis that the future price will be high). So as a result customers see these are fun additional features to try out (for free). BUT when the time comes Figma is going to have to massively crank up the price (either overall or more likely for use of the AI features) as the mechanism that has subsidised low cost AI usage tokens runs dry.
The real question is: how many customers would really be willing to say pay double for Figma to keep the AI tooling they getting now? Or how many will be like screw that; those tools were fun but they are in no-way worth paying for - leading to customers stepping away from newly expensive AI features and products.
Figma offers a more valuable fundamental service than most digital products; in a world where AI tokens go way up in cost, think of all the dedicated AI apps that folks will just be like, nope I’m not paying. There’s your big bust.
This is the dynamic and reality to consider; what folks will do when the tokens stop being subsidiesed.
It's hard to say how sustainable free tier pricing is - it depends on how much money they are making from every one else. It is a bit extreme though, with apparently only 5% of ChatGPT users paying for it.
I don't see this as the bell ringing at the top of the bubble though.
It’s not “free tier” it’s all the tiers being significantly discounted (underpriced). It’s ALL pricing.
Per the article is all underpriced… run the logic.. assume when the funds run out that prices across the board go up 5x, 10x, more. Assess where u think it’d all be at.
What do u think downward affect on the market will be?
> It’s not “free tier” it’s all the tiers being significantly discounted (underpriced). It’s ALL pricing.
That's possible, but how would we know what the cost to provide the service is from any of the companies, which will anyways vary from company to company since they are using different models and hardware (e.g. OpenAI/Microsoft NVIDIA, Google TPUs, Anthropic/Amazon Trainium).
We also don't know how much the average chat subscriber is using the service. A low usage subscriber may be profitable at $20/mo, but a high usage one unprofitable.
Chat subscription prices seem to have been steady at $20/mo for a while, but cost of providing the service is falling very fast. In recent Dwarkesh interview, Satya Nadella says costs are falling 5/10/40x year over year, or evern quarter over quarter.
Whatever this is (bubble/real investment) won’t crash right away. The reason is spelled out in the article: the data center buildout is still being financed by cash flow from operations. The buildout will start to cause problems when it will be a lot of it, mostly financed with debt and the debt close to maturity.
(BTW-there is one other item in the article that counters the collapse narrative: Meta being able to 100% control the AI buildout with only 20% payment - the other 80% is financed by a partner. If anything this type of deals gives big Tech even more run-way to get a business model working. The end is not near.)
it already has. those things have always been more important than bonds or gold. many books have been written about this topic, including the most famous ones.
Possibly companies that have performed poorly in the market recently due to being labeled an “AI loser”? I imagine that penalty would go away, though in a general market panic they would also be hit (just not as bad and would recover faster, probably).
I think of anything denominated in USD as a big risk right now. Trump is dead set on burning away the USD's privilege of being the world's reserve currency. If anything it seems like it's a deliberate policy to devalue the dollar, in order to improve US domestic competitiveness.
The benefit of crypto bubble was one could buy bitcoin 10 years ago, export it offline, and forget it until now. Is there any single benefit from AI bubble for an average loser who missed buying the US top 5 stocks?
I also believe it’s a bubble but how can it be a bubble if everyone thinks it’s a bubble? It seems like bubble is the general consensus but there hasn’t been something like a “pop” yet.
I don't see how it can be a bubble (or at least a risky one) without much debt. Places like Google and Meta profit $100B a year from operations and are funding most of this from cashflow. The risk is that they have to write down a bunch of overbuilt data center assets in a few years, but lenders won't be going under because there aren't any lenders involved. And without lenders being at risk, the system isn't at risk.
*There is some leverage. Coreweave has borrowed a bit for example. But none of this is really systemic and no one is levered to the eyeballs in the AI space.
There is a bit of nuance to the cashflow. That $100 bn cashflow was used for something else before AI: mostly share repurchases and M&A. Now it's being redirected to capex. That removes some of the support for the stock prices; there's no longer a multibillion dollar bid every year for GOOG / META / MSFT etc. stocks.
But you're right that this shouldn't affect lenders, unless we see a lot more borrowing (which is coming, BTW: ORCL and GOOG just issued $10+ bn debt each for AI data centers).
I think some of the valuation, or agreements made right now are based on future cash flow promises, which are assumed to be higher.
There are always signs of collapse before something collapses; we, somehow, choose to ignore those signs except in hindsight.
I guess the trillon dollar question is what might be the signs that we're at the top of this bubble.
Looking at a couple of prior bubbles ...
In the subprime housing crash of 2007/2008, lending practices got so bad that there were low-doc and no-doc loans (aka "liar loans") where you could get a mortgage without any evidence of being able to repay it.
Immediately prior to the dot com stock market crash of 2000, EVERYONE was talking about how much money they were making. At work it was insane with people monitoring their stock portfolios all day, reporting how much money they'd made that morning, etc. Tons of people were beyond "all in" - buying on margin.
So what about now with AI market caps, infrastructure builds, etc. Are we in the bat shit crazy "all in" stage yet, or can it conceivably go further?
As the article surmised at the end; AI is most definitely not free (and infact is quite costly) but major players are subsidising the price and cost of tokens to developers, such that right now everyone is pretty much getting AI tools and processing power for free (or very very low cost).
The signs are many products and services adding AI cause they can, but there being very little logic as to how much this addition makes sense. In a way this is the artificial demand they can point to for their (untrue) ongoing growth narrative - look companies are still adding more and more AI.
Let’s take the example of Figma, which is listed on the market after a successful IPO. At its core it’s a very nice design tool that bought the innovation of collaborative live editing (ah la Google docs) to an adobe illustrator like product. It’s still great for this.
But in recent times much development has gone into their AI and MCP features, these likely bring some value; but the market sees these are offered features built like any other feature; when in fact there is a pay-for-use baked in that is being massively subsidised (on the basis that the future price will be high). So as a result customers see these are fun additional features to try out (for free). BUT when the time comes Figma is going to have to massively crank up the price (either overall or more likely for use of the AI features) as the mechanism that has subsidised low cost AI usage tokens runs dry.
The real question is: how many customers would really be willing to say pay double for Figma to keep the AI tooling they getting now? Or how many will be like screw that; those tools were fun but they are in no-way worth paying for - leading to customers stepping away from newly expensive AI features and products.
Figma offers a more valuable fundamental service than most digital products; in a world where AI tokens go way up in cost, think of all the dedicated AI apps that folks will just be like, nope I’m not paying. There’s your big bust.
This is the dynamic and reality to consider; what folks will do when the tokens stop being subsidiesed.
It's hard to say how sustainable free tier pricing is - it depends on how much money they are making from every one else. It is a bit extreme though, with apparently only 5% of ChatGPT users paying for it.
I don't see this as the bell ringing at the top of the bubble though.
It’s not “free tier” it’s all the tiers being significantly discounted (underpriced). It’s ALL pricing.
Per the article is all underpriced… run the logic.. assume when the funds run out that prices across the board go up 5x, 10x, more. Assess where u think it’d all be at.
What do u think downward affect on the market will be?
> It’s not “free tier” it’s all the tiers being significantly discounted (underpriced). It’s ALL pricing.
That's possible, but how would we know what the cost to provide the service is from any of the companies, which will anyways vary from company to company since they are using different models and hardware (e.g. OpenAI/Microsoft NVIDIA, Google TPUs, Anthropic/Amazon Trainium).
We also don't know how much the average chat subscriber is using the service. A low usage subscriber may be profitable at $20/mo, but a high usage one unprofitable.
Chat subscription prices seem to have been steady at $20/mo for a while, but cost of providing the service is falling very fast. In recent Dwarkesh interview, Satya Nadella says costs are falling 5/10/40x year over year, or evern quarter over quarter.
Here, at 1:10:56
https://www.dwarkesh.com/p/satya-nadella-2
Whatever this is (bubble/real investment) won’t crash right away. The reason is spelled out in the article: the data center buildout is still being financed by cash flow from operations. The buildout will start to cause problems when it will be a lot of it, mostly financed with debt and the debt close to maturity.
(BTW-there is one other item in the article that counters the collapse narrative: Meta being able to 100% control the AI buildout with only 20% payment - the other 80% is financed by a partner. If anything this type of deals gives big Tech even more run-way to get a business model working. The end is not near.)
If the AI bubble broke, what’s a safe haven? US gov bond? Gold?
friends. community. family
Hopefully it doesn’t dip to that level.
it already has. those things have always been more important than bonds or gold. many books have been written about this topic, including the most famous ones.
Spoken like a true American.
Possibly companies that have performed poorly in the market recently due to being labeled an “AI loser”? I imagine that penalty would go away, though in a general market panic they would also be hit (just not as bad and would recover faster, probably).
I think of anything denominated in USD as a big risk right now. Trump is dead set on burning away the USD's privilege of being the world's reserve currency. If anything it seems like it's a deliberate policy to devalue the dollar, in order to improve US domestic competitiveness.
So, maybe rest-of-the world bonds?
Canned goods? Camping gear?
Except everyone who studies demographics knew Chinese real estate would collapse.
With AI you have to predict whether or not people will make useful tools from this technology or whether we have peaked and nothing will be new.
The benefit of crypto bubble was one could buy bitcoin 10 years ago, export it offline, and forget it until now. Is there any single benefit from AI bubble for an average loser who missed buying the US top 5 stocks?