I’ve long admired Harris Kupperman, the founder of Praetorian Capital, for his ability to cut through noise and spot big-picture themes before they become consensus.
Harris just released his thoughts on markets, AI and his positions in his Q3 investor letter, which I’m happy to share with my subscribers. Harris warns that the long-awaited AI reckoning is finally here, and a broader market crash may be just beginning.
In his latest Q3 letter, Harris explores the increasingly unsustainable economics of the AI boom, the potential for a wave of bankruptcies to ripple through financial markets and the broader economy, and why he’s positioning his portfolio for a major downturn by raising cash and turning to U.S. government bonds.
He also reviews his current long positions and explains why he remains bullish on precious metals, sees significant upside in one little talked about sector as AI reshapes the labor market, and talks undervalued real estate assets and emerging markets in the future. (*Photos, charts and article links added for emphasis by QTR). *
I like to say that most things in finance are rather obvious; it’s the timing that’s difficult. In that vein, it’s always been rather obvious that the economics of AI are absolutely abysmal.
I think it’s highly unlikely for the industry to ever reach profitability as currently configured, as it would need a few percent of US GDP annually in revenue to break even on the capital already invested, and may need as much as ten percent of GDP to break even on the capital that many analysts expected to be invested by 2030. The numbers just cannot work—especially as tech tends to be deflationary.
How would anyone ever expect the US to be spending a few trillion dollars annually on AI, when software, which is already ubiquitous, and maturing, is currently around half a trillion of US spending?
Think back to the railroads; they became dominant modes of transport, but they periodically destabilized the US economy as they went bankrupt in repeated waves during the 1800s. Then, someone picked up the pieces, and railroad construction continued. I think we’re at a similar point in the AI journey. I definitely don’t think that AI is over, or that the buildout won’t continue. I just think a phase of rationalization is upon us.
It’s said that you go broke slowly, then all at once. I think this is a good mindset for how the AI sector will also fail. Slowly, then all at once.
The first signpost will be that the funding gets cut off, and I think we’re finally witnessing that. As funding grows more scarce and expensive, the AI sector will struggle to support unprofitable operations, and I expect to see many bankruptcies of neoclouds, followed by the large frontier AI labs. While they may have interesting assets, they also have hundreds of billions of contingent liabilities along with purchase commitments that I cannot imagine anyone wants to fund, especially if chip costs come crashing down with reduced demand.
Following bankruptcy and the rejection of these liabilities, I then assume that they get folded into the hyper-scalers.
However, these will be incredibly messy and complex bankruptcies. Just think back to how messy Enron was, with hundreds of SPVs and all sorts of future commitments over pipelines, utilities and whatnot. This will be worse, as many of the SPVs have guarantees from partners that will want to escape those guarantees.
I assume that all of this will cascade through the economy and the stock market, especially given how material semiconductors and AI have become as a percentage of US equity indexes. If you take away the performance from AI driven sectors, most stocks have done quite poorly over the past few years, and the stock market would be a good deal lower. I think stocks will revert there as AI spending slows.
AI capital expenditures are responsible for roughly three percent of US GDP currently and are predicted to grow further in 2027. While it’s harder to quantify the size of the Wealth Effect on consumer spending, it wouldn’t surprise me if it were over ten percent of US GDP.
Being conservative, I feel pretty confident in saying that the AI mania is responsible for well over ten percent of US GDP, and should anything happen to AI, even if it’s only a slow-down in spending, it would materially slow US GDP, and possibly even send it into decline, especially if the stock market collapsed at the same time.
I have gone through two US economic down-cycles during my investing career, along with a COVID panic. In all three cases, owning almost any part of the yield curve was the winning strategy—even if guys doing more exotic trades got all of the accolades. Having lost money shorting financials too early during the GFC, and then buying volatility too early during the COVID panic, I’ve decided that I’m going to focus on what’s worked in all three prior equity smashes, I’m going to buy US government paper. As of writing this to you, we already own some SOFR futures along with call spreads. I plan to add should the trade start working.
I know that the government is functionally insolvent, and that during “Project Zimbabwe” debt gets debased, but there are moments in time where fixed income can rally further than anyone can ever imagine. While bonds aren’t my favorite asset, they’re a blunt tool to express a view. As far as I’m concerned, outside of AI, the Wealth Effect driven by the AI mania, and health care for aging boomers, the rest of the economy is already effectively in a deep recession. I expect this recession to accelerate and broaden should AI spending slow, and equities sell off as AI businesses fail.
With a Congress that’s likely to swing to the left following November’s elections, I think that a fiscal response will be off the table until things get really bad. Therefore, early in this crisis, the only tool that the government will have to combat the market’s collapse is monetary intervention, and they’re likely to give it all they have. My thinking is we own bonds first, and then after the smash, we switch to debasement assets. (Though I am a bit skeptical on how bonds further out on the duration curve will perform, given current debt levels and the potential for inflation to remain elevated).
In preparation for this, I’ve taken exposure down and built cash. I’ve dramatically dialed down exposure to any businesses that are sensitive to GDP. What’s left on our sheets are businesses…that did surprisingly well coming out of the GFC, businesses tied to volatility…and businesses that do well with lower interest rates. We’ve been preparing for a deep and lasting recession for a while now.
It’s been postponed by the AI mania and the Wealth Effect that has built up around that mania. It’s been frustrating to have reduced exposure and a bearish outlook as equity markets rally. It’s been irritating to have missed out on an unsustainable mania in AI. Now, I think we’re finally at the point where this all tumbles down.
I’ve never thought that we could go fully into “Project Zimbabwe” without one last crash that gives government officials the mandate to blatantly debase. It’s hard to know exactly when this happens, but I have my roadmap, and I’m content to continue with reduced exposure. As of right now, I am far more concerned with the potential for substantial downside from here, the upside can come later.
Top core position weightings at quarter end from largest to smallest.
