Ghost Story
- Karl Johansson
- för 4 dagar sedan
- 6 min läsning
Spooky stuff going on in American finance.
Let me tell you a story in three parts. I cannot guarantee that this story story is true, as history has demonstrated I’m neither Nostradamus nor Cassandra, but I can guarantee that it is interesting. It is a story about finance certainly, but more than that it is a story about the dangers of Goodhart’s law, the value of generalist thinking, and the necessity of thinking for yourself. It is a sort of out of season ghost story, so imagine me telling you this with a torch pressed to my chin, lit spookily from below.
This story begins in Silicon Valley, with the AI firms. I know I’ve written ad nauseum about the ways in which AI companies overhype their technology and consistently fail to produce a viable business model. Ed Zitron is the master of this line of thinking and is the person who convinced me of his thesis that AI is not worth what it costs to run. Where I disagree with Zitron is his insistence that the AI bubble popping will be less bad than previous bubbles like the Dot Com bubble. The sums malinvested in worthless data centres and training runs for models showing minor gains on unrepresentative benchmarks are astronomical. The valuations of companies which brand themselves as part of the AI ecosystem and who prattle on about how “agents will change everything” beggars belief.
As Zitron has shown, it seems most if not effectively all the demand for AI data centres comes ultimately from Anthropic and OpenAI. If you go on LinkedIn you see annoying posts about all manner of AI startups with strange and quirky names. Swedish LinkedIn especially is infested with posts from people working at Lovable claiming to be “building the last piece of software” – though why you would need the intermediary of Lovable between Claude Code and the people building the final software if AI truly could just build any software is beyond me. And yet, I’m writing this in Word because no new word processors with better features or a lower price seems to have been made. The same is true for all the other forms of software I use regularly. End consumers like me are not the ones driving the demand for AI, companies like Lovable are. They try to build something I will want when it is done, I’m not using their service to make it myself.
This is so intuitively obvious if you move offline. You regularly meet people who use AI for searching, and if they are pretentious they may even call it research. You also hear about programmers using it to develop software, but other than that it seems to have little use. In an almost post-modern way, the best use case for AI is to build a startup to get a high valuation, which is exactly what OpenAI and Anthropic are doing in the first place. People get so drawn into the metrics that they fail to see the forest for the trees. Pick any metric you want and AI is super impressive, users per month, user growth, tokens spent, venture capital money raised, benchmark scores, etc. But if we ignore the metrics and focus on what those metrics are trying to measure – if the bloody thing makes you work faster or enables you to do things you couldn’t do before – then AI suddenly doesn’t seem so impressive anymore.
The second part comes via Chase Taylor, who made this little explainer about how the data centre buildout is being financed. The reality is that like a lot of modern day international finance it is opaque – intentionally so – and there have been efforts to finance data centres ‘off balance sheet’ in the lingo. In plain English, this means debt structured in such a way as to not be shown in financial disclosures. Why avoid having this debt on the balance sheet? To present oneself as a better capitalised business and thus a more attractive investment. Sounds like a great little corporate finance hack, right? Except that debt still needs to be paid, and if you hide what you owe it becomes easier to borrow more than you can afford. Of course, most of the companies who are building out the AI data centres are very profitable in their old, pre-AI business lines, and can probably afford to pay their debts. Still, the principle of the thing. Why hide your bets on what you continue to claim to be the most revolutionary technology of our lifetimes? This is not the type of behaviour you expect from someone completely convinced of their claims.
The reason I bring this up is to underline the point made earlier: there are astronomical sums malinvested in AI which will never see a return. Add to that a lot of debt, some of which is opaque, and you get a volatile witch’s brew. Given how important AI and the hyperscalers are to American finance, it is possible that a revaluation about the earnings potential of Nvidia, or Oracle, or OpenAI triggers a wider liquidity crisis. Depending on who holds off balance sheet data centre debt, it could trigger a banking crisis if it turns out that data centres are a rotten business. There are also tons of funds, foreign and domestic, with large shares of their assets tied up in AI. To some that might sound like poor risk management, but to others an AI-focused fund is exciting. What happens if all the AI stocks drop at the same time?
The final part comes from Michael Green’s work on passive index investing. Green makes a compelling case that the index revolution in American finance has created essentially a new type of eco system which means that old evolutionary niches are being destroyed and new ones created. It used to be that betting against the markets could be consistently profitable when you were better informed than the average fund, but with more and more assets being invested in passive funds there it is increasingly risky to bet against the index. Add to that the fact that indexes tend to be market cap weighted, meaning that new money going into passive investments disproportionally allocates to the companies which have the highest market cap and you get an index with very high valuations.
Not coincidentally, the biggest companies are the ones caught up in the AI hype. Green focuses mainly on what the passive flows mean for the workings of the market, but implicit in his writing is the threat of a flow reversal. If the markets have gotten less efficient because a plurality just buys the index, and those consistent flows are enough to move markets generally, what happens if the passive crowd gets cold feet? What happens if they stop buying every month and put that money in a high interest account instead? What happens if the passive crowd all start selling at the same time?
As I said at the beginning, this is not a prediction or a prophecy but a worst case scenario. These three factors could interlock and form a three stage rocket propelling American investors towards a massive financial crisis, or it could just be a spooky story. The AI bubble could deflate without having ripple effects which reach passive investors or trigger cascades through off balance sheet debt. Nothing is certain and any one of these three risks could become a problem independently from the others, or turn out to be a manufactured issue blown way out of proportion by a paranoid blogger.
But you have to admit, if it were to all go off it would make one hell of a scary story. More Tom Clancy than Steven King, but still.
If you liked this post you can read a previous post about the war in Iran here or the rest of my writings here. I also have a section for longer reads I call essays here, I particularly recommend my series called The Bird & The Technoking exploring Elon Musk's takeover of Twitter, and its political and cultural implications. It'd mean a lot to me if you recommended the blog to a friend or coworker. Come back next Monday for a new post!
I've always been interested in politics, economics, and the interplay between. The blog is a place for me to explore different ideas and concepts relating to economics or politics, be that national or international. The goal for the blog is to make you think; to provide new perspectives.
Written by Karl Johansson
Sources:
Cover photo by juliane Monari from Pexels, edited by Karl Johansson
Kommentarer