Your post does a good job of outlining the contradictory forces underlying the current macro-economy, but I suspect it underestimates the magnitude of the deflationary shock that would follow the collapse of the AI bubble. The economy could probably sustain the fall in share prices of the companies investing in AI without too much damage, but once those share prices fall sharply the companies are unlikely to be able or willing to continue their massive investment program in chips and data centers that is propping up aggregate demand, leading to a sharp fall in output measured like GDP and a sharp rise in unemployment. Even those events might not be the whole story, since it is likely that those effects would destabilize the credit markets leading to a general financial meltdown similar to the financial crisis of 2007-8, and we know what kind of lasting damage that inflicted. The question for the bubble is not so much whether AI will start to generate some profit revenues, which it probably will, but whether those revenues and their growth rate will remotely justify the huge investments that have been made to pursue AI dominance.
Thanks Duncan. I think those questions about the magnitude of AI's effect are motivating my next post, which will be an AI industry analysis. I think the term AI bubble is thrown around a bit too loosely, and to really think about the impacts on the economy, rather than just investors, you need to think about the different companies, their different positions, and which ones you really see as impacted.
I think it's a useful observation that some companies and investors are doubling down on their most optimistic views, and since optimism like that is rarely rewarded in full, there some change coming. But this isn't universal and the different forms it takes could be a lot as important as if it happens or when. To give one simple example, it's hard to look at NVidia and suggest they are investing unwisely. For the most part they are making profits immediately. Whether their market cap stays at $5 billion is a different question, but that valuation is not a requirement for them to continue the same operations.
OpenAI though, is tied into commitments that are unresolvable if they don't make the revenue growth in their forecasts, so in that case their something has to be unwound if those aren't met.
Those are very different effects, and one could happen without the other. Effective competition could challenge NVidia, and bring down their profit rate dramatically, without meaningfully changing the amount of compute provisioned or AI services sold, though someone in that chain, or the ultimate consumers would benefit by lower prices vs. NVidia shareholders.
One of the key dynamics I'm thinking of here, is what if you look at infrastructure costs as if they were a cost of doing business for compute providers, rather than new business? Would they have still happened if they some shock had introduced new requirements with those costs to continue the same business? I think you have a hard time saying no. You can entertain the question, but it's almost impossible to imagine how computing would have been scaled back in that scenario, and it wouldn't have simply made sense to pay the cost. That might seem to break supply and demand dynamics, but we suspect that information technology has created a social surplus, and we can imagine that being pulled back to keep the providers operating, without reducing demand. It would be a loss to the social surplus and thus society, but it's also something that could have gone unnoticed if it was spread broadly enough without significant obstacles to it's distribution.
It's interesting how you just nailed the 'inconveniently inaccesible' timing. So tru!
Your post does a good job of outlining the contradictory forces underlying the current macro-economy, but I suspect it underestimates the magnitude of the deflationary shock that would follow the collapse of the AI bubble. The economy could probably sustain the fall in share prices of the companies investing in AI without too much damage, but once those share prices fall sharply the companies are unlikely to be able or willing to continue their massive investment program in chips and data centers that is propping up aggregate demand, leading to a sharp fall in output measured like GDP and a sharp rise in unemployment. Even those events might not be the whole story, since it is likely that those effects would destabilize the credit markets leading to a general financial meltdown similar to the financial crisis of 2007-8, and we know what kind of lasting damage that inflicted. The question for the bubble is not so much whether AI will start to generate some profit revenues, which it probably will, but whether those revenues and their growth rate will remotely justify the huge investments that have been made to pursue AI dominance.
Thanks Duncan. I think those questions about the magnitude of AI's effect are motivating my next post, which will be an AI industry analysis. I think the term AI bubble is thrown around a bit too loosely, and to really think about the impacts on the economy, rather than just investors, you need to think about the different companies, their different positions, and which ones you really see as impacted.
I think it's a useful observation that some companies and investors are doubling down on their most optimistic views, and since optimism like that is rarely rewarded in full, there some change coming. But this isn't universal and the different forms it takes could be a lot as important as if it happens or when. To give one simple example, it's hard to look at NVidia and suggest they are investing unwisely. For the most part they are making profits immediately. Whether their market cap stays at $5 billion is a different question, but that valuation is not a requirement for them to continue the same operations.
OpenAI though, is tied into commitments that are unresolvable if they don't make the revenue growth in their forecasts, so in that case their something has to be unwound if those aren't met.
Those are very different effects, and one could happen without the other. Effective competition could challenge NVidia, and bring down their profit rate dramatically, without meaningfully changing the amount of compute provisioned or AI services sold, though someone in that chain, or the ultimate consumers would benefit by lower prices vs. NVidia shareholders.
One of the key dynamics I'm thinking of here, is what if you look at infrastructure costs as if they were a cost of doing business for compute providers, rather than new business? Would they have still happened if they some shock had introduced new requirements with those costs to continue the same business? I think you have a hard time saying no. You can entertain the question, but it's almost impossible to imagine how computing would have been scaled back in that scenario, and it wouldn't have simply made sense to pay the cost. That might seem to break supply and demand dynamics, but we suspect that information technology has created a social surplus, and we can imagine that being pulled back to keep the providers operating, without reducing demand. It would be a loss to the social surplus and thus society, but it's also something that could have gone unnoticed if it was spread broadly enough without significant obstacles to it's distribution.