AI spending will cause a recession
US GDP growth in 2026, estimated at ~2%, is projected to be surpassed by hyperscaler capex spending (~700 billion).
What is this money being used for? At first, they said it was to compete for the top spot. Just exactly what privileges that “top spot” comes with has never been made clear, nor what exactly that looks like, but vague notions of technical revolution and the supposed eventual freedom to fire accountants and lawyers has carried the vibe of significance sufficiently as far as the market is concerned.
So besides representing a race to the bottom of the corporate cash pile, what other consequences are there for this spend? The answer is: competition not just with one another, but with you.
Competition for memory, hardware. Competition, now, for more of your capital through public stock offerings and record debt issuance. Competition for water and electricity. AI capex spending has not only been dominating economic growth, it has been a central driver of inflation.
Enter the federal reserve. What could be their response to this kind of inflation? If they raise rates modestly, it has the intended impact of reducing the velocity of money .. except for the fact that these hyperscalers are perhaps the least interest rate sensitive entities in the picture. For every pound of distress higher rates cause economically, maybe an ounce is absorbed by Google. And the “intended effect” is deflationary for everything .. except the very objects of capex spending: your utility bill, your new phone purchase, anything involving memory or chips. Food can be affected.
This techbology is nascent, speculative, and reliably has no top line contribution to the economy. There is no serious suggestion otherwise. At best, and optimistically at that, the upside for the value proposition of AI is our income.
And I say optimistically because I use AI. I use it as somewhat of an enhanced google search. A majority of its responses contain either completely fabricated or incomplete information. Most of the time it gives me responses retrofit to my biases - and why wouldn’t it? That’s a model that works with any other social media. It figures that it works just as well for anti-social media. But besides general purposes as a sophisticated search engine and calculator that understands English, it offers no real utility.
What does it do really well? Well for one, it’s certainly quite validating. So it’s quite understandable for someone with a narcissistic bent to feel like some”one” finally *gets* what they have to say.
This is the technology we’re talking about. Unreliable utility, probably incapable of handling essential tasks unsupervised for decades (if at all), that socially has the net effect that we saw from social feed manipulation but on steroids.
Anyone who has spent any length of time trying to coach an AI to remember rules to be a proper brainstorm or workshopping partner will have no disagreements on these points.
This is the technology that has captivated the egos of the companies that represent half of the US stock market. The effect is that they now compete with you on your regular purchases for a technology that produces no value commensurate with its expense - depending on how greatly you value personal access to a digitized pseudo schizo with no accountability and an emphatic and flattering vocabulary.
Meanwhile, semiconductors and memory are attracting all of the capital in the market place. Technologies that have grown exponentially to accomodate the resource intensive models currently being produced by American companies - while China develops models that are more efficient by a couple orders of magnitude. Slight differences in accuracy aren’t really an argument for a serious person to make here, as the deficit between frequently wrong and sometimes right is a rounding error to begin with. But it represents a path to innovation in this technology that is emblematic of what allowed these hyperscalers to accrue such large sums to waste to begin with - one that is asset light.
From multiple angles, we are in the process of what might be the greatest destruction of capital in history. In absolute terms without question. On a relative basis, it remains to be seen.
Best case, one company CEO after another will stop role playing as Tony Stark and will connsumately be rewarded by the market for doing so. That’s the peaceful sunset. Middle case, hyperscalers lose their unofficial “credit rating” from the equities market and go the way of defunct enterprises in history. Worst case, capex spending continues to warp the economy and monetary policy around the dubious prospect of a digital revolution driven by hardware volume - until it hits a wall at 80mph as earnings growth crests, and the technology remains insufficient to do anything useful for other sectors besides lay off support staff.