How Could the AI Bubble Pop?
AI is of course having a profound impact on the economy and work itself, but in times of exuberance, investors often decouple their valuations of companies from reality and underlying fundamentals. This creates a situation where a new technology (e.g., railroads, internet, cars) will become integral to the economy, but at the same time it creates a financial bubble which unfortunately always pops. The big question though is when and how the AI bubble actually pops and I wanted to propose a few possible scenarios.
Before I do that though, I wanted to clear up a very common misconception about the AI market which is that despite hundreds of billions of dollars being spent on AI, the market for AI is not hundreds of billions of dollars today. The actual market today for AI is about $20-30B. Let me explain with an analogy: during the railroad bubble, the market for railroads was measured initially in trains, miles of track and bridges and was thought to be huge much like today where we are spending hundreds of billions on chips, data centers and general infrastructure. For observers of history, we know from the railroad bubble that one of the reasons it popped is that the actual market was passengers and tons of freight - i.e., the people who pay money for the underlying service. In the case of AI, end users are currently only paying about $20-30B for using AI where OpenAI comprises about 75% of the market share. Unfortunately in the case of AI, the infrastructure will be useless in 4-5 years where we still use much of the railroad bubble infrastructure today.
So how does the bubble pop?
There are a couple different ways that the AI bubble can pop which I will detail below:
Earnings Pressure: With the exception of Nvidia (AI product creator) and Apple who is investing 40% of their free cash flow on AI, the Mag7 are investing 100%+ of free cash flow ($350B+) into AI. If they continue to do this, in 4-5 years they will show minimal earnings in their financial statements as depreciation for GPU's which comprises 90% of these costs is depreciated over 5 years and not decades like traditional capital expenditures. Within 2026, we will start to see from the Mag7 substantial declines in earnings with this rate of investment even if they continue their >14% annual revenue growth. Can investors stomach rapidly declining earnings for the Mag7?
Electricity: A fundamental limit to AI is electricity itself where today AI data centers in the US consume about 11 GW (about 4% of US data consumption) of energy. There is simply not an unlimited supply of additional electricity within the US and reasonably the current energy production capability could support 2-3X the current AI demand without causing blackouts across the US. The current AI narrative though is that compute energy usage will grow 50-100% per year and for example OpenAI has expressed that through their Nvidia partnership that they will add 3.5 GW of usage themselves in 2025. This means that within the next 2-3 years, all of the available energy flux within the US will be consumed by AI and at the same time, energy prices will skyrocket as they are extremely inelastic. Consequently, this means that the volume of new chips and AI infrastructure within the US will flatline within the next 2-3 years as energy production is very slow to come online. Does the AI narrative continue if the rate of growth decreases over the next 2-3 years to match the rate of growth of the electrical production system itself (2-3%)? Does the AI narrative continue if communities stop allowing for AI data centers to be built as their costs swell?
Revenue Generation: Imagine you started a business and spent $350 per year and only generated $30 in revenue. After you paid all of your bills and you add everything up, you lost $320+. Even if you thought that this investment would pay off big one day, how long could you continue to do this before investors pulled the plug? The answer is probably not that long unless you had a really really good narrative! When we look at AI, we see that the underlying demand for AI is currently $20-30B and growing at about 100% per year which though impressive, means it would take five years of continuing this growth just to make enough revenue to match your initial first year investment….not profits.
Many AI boom apologists will say that the demand justifies this and that the industry cannot keep up with it. However, this growth and demand is predicated on continually spending $350B+ per year in capital expenditures and effectively subsidize the price of AI where without this investment, the price would skyrocket and demand would plummet. Can investors tolerate declining earnings with very little increase in growth derived from AI for many years?
With all of these pins to pop the bubble being described, one must also consider the probability that the bubble never pops and either slowly deflates or maintains its size while the underlying fundamentals catch up. For example, Tesla has been able to continue its 200+ P/E ratio compared to General Motors (7) and Ford (15.6) on the promise of big revolutions (e.g., household robots, robotaxis, self-driving cars) for over a decade. Can AI keep the narrative alive with extreme earnings declines, electricity capping growth and little revenue being generated from AI over the next 4-5 years? Even if it doesn’t, AI will certainly have a meaningful impact on the economy and let's hope if the financial bubble pops, the resulting damage will not be too systemic.
As always, let me know what you think!
