September 29, 2026:
This article is all about the artificial intelligence boom. We’ll cover the enormous scale of AI spending, whether it might be a bubble, what makes the outcome so hard to predict, and how we’re investing. (For a primer on AI itself, this plain-English overview is a good place to start).
A historic investment boom
AI is a profoundly useful technology. It’s also really expensive, and companies are spending vast amounts to build out the capacity to deliver it.
Capital expenditure, or “capex,” measures how much companies are spending on long-lived assets such as buildings, computers, and machinery. Here’s a chart of capital expenditures by S&P 500 companies going back to 1995:
Source: Deutsche BankThe chart captures three past capex surges: the internet in the late 1990s, real estate in the mid-2000s, and shale in the mid-2010s. All of these are dwarfed by the scale of the current AI boom.
Much of the spending is driven by the “hyperscalers” — the companies that own and rent out most of the world’s cloud computing capacity. It’s the expanded cloud computing that enables artificial intelligence by providing the scalable computing power, storage, and infrastructure required to train and run complex AI models. This next chart shows that the major hyperscaler companies are spending several times more than they did just a few years ago:
Source: Vanderbilt UniversityThese companies have long been admired for how much profit they could earn compared to how much they had to invest. That is no longer the reality. The next chart shows hyperscaler earnings and capex in one chart. The orange line shows capex; the blue line shows earnings in the form of EBITDA (a rough measure of profits before taxes, interest, and capital spending):
Source: Deutsche BankFor most of the last decade, earnings ran well ahead of capex. This left the hyperscalers with plenty of cash, some of which they returned to shareholders in the form of share buybacks (green line).
Now, they are spending all of their cash on capex. Instead of buying back shares, they are issuing more, as shown by the green line going negative. They’ve also started borrowing a lot of money:
Source: BISBeyond what they’ve already spent, hyperscalers have signed contracts committing them to substantial future spending on AI infrastructure. The next chart shows the recent growth in these obligations, which have surged to multiples of current capex.
Source: Wall Street JournalIn the face of this immense spending, it’s reasonable to wonder if AI is a bubble. The next two sections lay out some arguments on each side.
Signs of a bubble
The spending boom is bigger than past bubbles
The chart below shows AI spending, as a share of GDP, along with major investment booms of the past. Three of these (canals, railroads, and telecom/fiber) turned out to be bubbles that led to major busts. The AI boom is on track to be larger than any of them.
Source: Wall Street JournalThe stock market is very expensive based on normalized earnings
The CAPE ratio of the US market, which compares prices to a decade of earnings rather than a single year, has only been higher at the dot-com peak, and then not by much. (This measure assumes today’s earnings are running above their sustainable trend; we’ll take up the other side in the next section).
We have seen bubble-like moves in some AI-related assets
The next chart shows a ratio of total returns for semiconductor manufacturers vs. the overall US market for the past 25 years. Chip makers performed roughly in line with the market for 15 years, then started to pull ahead, and then went vertical in the last couple of years. Before the correction visible at the right of the chart, chip stocks had beaten the US market by 121% in just 12 months. (Now they’ve outperformed by “only” 81% year-over-year).
Source: KoyfinMore recently, the entire Korean stock market has more than doubled over the past year, and at one point had nearly tripled. Granted, a couple big memory manufacturers account for nearly half of the index, but that’s still a huge move for an entire country.
The speculative behavior isn’t limited to public markets. Based on first-half 2026 data, venture capital funding to AI companies is running 3.5 times faster than in 2025, and 6 times faster than in 2024.
There’s a lot of circular investment in the AI industry
Chip makers own stakes in the companies that buy their chips, hyperscalers invest in the AI model companies who in turn buy their cloud services, and so on. These arrangements help the boom feed on itself, but their self-reinforcing nature could go into reverse if the boom turns to bust. (Here’s an interactive map showing all the interconnections in the industry).
Industry leaders suspect a bubble (but they’re spending anyway)
Here are some quotes from major tech leaders:
- Meta CEO Mark Zuckerberg: “If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously… but what I’d say is I actually think the risk is higher on the other side.”
- OpenAI CEO Sam Altman: “Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes.”
- Amazon founder Jeff Bezos: “This is a kind of industrial bubble.”
The quote that best sums up the apparent mood comes (second hand, admittedly) from Google founder Larry Page: “I’m willing to go bankrupt rather than lose this race.”
As for the slightly smaller fish… a KPMG survey of one-hundred large company executives found that 25% thought AI was a bubble, while 80% were spending on AI.
A good technology doesn’t equate to a good investment
Many of history’s bubbles have centered on new technologies that were indeed “the real deal.” Canals and railroads, for example, both transformed the economy, while also ushering in brutal busts.
More recently, the internet has radically changed the world since its creation, possibly even more than its early boosters predicted. But investors in the late 1990s got way ahead of themselves. In the 3 years leading to the peak, the tech-heavy NASDAQ 100 index more than quintupled in price. Within 3 years after the peak, it had given all those gains back.
Source: KoyfinAI is almost surely a hugely transformative technology that’s here to stay. But that doesn’t necessarily make it a good investment at any price.
Signs of something else
Not every capex boom is a bubble
Investment booms are a feature of the economy, and they don’t necessarily lead to busts. A good recent example is the cloud computing buildout of the 2010s, which paid off handsomely for the companies that undertook it.
The market is reasonably priced based on today’s earnings
Measured against the past year’s earnings, the market is just a little more expensive than its average valuation over the past decade. If we use next year’s expected earnings, it’s right in line with the average. (This assumes today’s boom-time earnings can be sustained).
Companies are making real money
AI adoption has been lightning fast compared to prior new technologies. This usage is translating to dramatic revenue growth not just for the “picks and shovels” suppliers, but for AI model builders themselves:
AI is transformative enough to be worth all the spending
AI’s most enthusiastic backers envision dramatic economic growth, a rapid acceleration of scientific progress, and superintelligent machines that can outperform humans at every task. An outcome like that could justify immense investment, and the bigger risk would be falling behind.
This is the most compelling argument that the AI boom is fundamentally sound. If it turns out to be right, all the spending will look entirely justified in hindsight. Unfortunately, it rests on a claim that is very difficult to prove or disprove in real time.
Nobody really knows
Part of the problem is that there’s a lot of uncertainty about AI’s potential impact. Predictions span an extraordinary range. At one end is a utopia of abundance and scientific miracles; at the other is mass unemployment, or worse. Here’s a chart from the Dallas Fed showing economic growth under different AI scenarios:
Source: Dallas FedThanks for clearing that up! In seriousness, the chart’s authors said, “Today there is little empirical evidence that would prompt us to put much weight on either of these extreme scenarios.” That’s the right way to think about it, in our view, at least when it comes to investing. It’s always easy to imagine all the drastic outcomes that could take place, but reality doesn’t usually turn out that way. People and societies usually adapt to big changes — enough, at least, to head off the extremes.
Even putting aside utopia and doom, there’s wide disagreement on how, and how much, AI will impact the economy. It’s also difficult to know who will be the winners and losers in an AI-enabled world:
- The big AI model producers are in stiff competition with each other, with much cheaper Chinese models that aren’t far behind. It’s not clear that any of them will have a sustainable edge.
- Transformative technologies have often ended up benefiting their users more than their producers, and AI may follow the same pattern. Companies outside the AI buildout could reap the efficiency gains without taking on its massive cost. (As discussed below, this dynamic would be helpful to our portfolios, but we wouldn’t be depending on it for investment success).
And even if AI generates a lot of profits for its producers, much of that success is already reflected in their stock prices. Many AI-related stocks will require tremendous growth to justify their current prices, and even more if they are to deliver further gains. By contrast, plenty of non-AI stocks are priced for fairly ordinary earnings growth.
Maybe a bubble, definitely a gamble
Some historical bubbles have been fairly straightforward to identify as they were happening. This isn’t one of those cases. There certainly are a lot of similarities to bubbles past — enough that they shouldn’t be dismissed. But there is enough novelty and uncertainty here to leave the door open that this is something different.
So if forced to answer whether this is a bubble we’d say… probably? But we wouldn’t feel very confident about it.
We’re a lot more confident saying that it is a big gamble. The bet is not that AI will “merely” change the world, but that it will do so in a way that is profitable enough to offset its massive costs. This bet stands a chance of paying off, but it is very far from certain.
How we’re investing
We aren’t comfortable making a big gamble on a single narrow outcome. Fortunately, we don’t have to. The all-or-nothing AI bet is priced into much of the stock market (most notably in the US), but not all of it.
Many areas of the US and international stock markets are priced to deliver satisfactory returns whether AI lives up to expectations or not. We are heavily tilting our stock exposure toward those areas, and seeking returns outside of the stock market as well.
This makes our long-term returns less dependent on an either/or AI outcome. It won’t spare us from a broad downturn should the AI boom go bust, because most stocks go down in a bear market. (That’s a risk we feel can’t be avoided without giving up the benefits of stocks). But reasonably priced assets tend to recover soon enough. Bubble-priced assets do not.
Conclusion
The AI investment boom may end up being the biggest infrastructure buildout in modern history. The sheer size of the spending is a bubble red flag. There are others: dramatic asset price moves, the extrapolation of boom-time earnings, signs of speculative fervor in public and private markets, circular investments, and industry leaders’ own bubble worries. AI wouldn’t be the first genuinely transformative technology to spark an investment mania.
But there’s a case to be made that this is not a bubble. Not all investment booms go bust, and the market could prove to be reasonably priced if current earnings can keep up. AI is a revolutionary technology that’s already generating rapid revenue growth for its producers. It could prove so valuable, and so profitable, that today’s spending will look entirely justified in hindsight.
That last claim is important, but hard to evaluate because nobody really knows if it’s true. Even dismissing the more extreme scenarios as unlikely, there’s a lot of disagreement about AI’s potential to reshape the economy. It’s also difficult to know who the winners will be, as the industry is very competitive, and transformative technologies often benefit their users more than their producers.
Our conclusion is that while AI does look suspiciously bubbly, we can’t make a confident call either way. But we are confident in characterizing it as a gamble that AI will meet the extremely high economic expectations that have been priced into much of the stock market.
Fortunately, we don’t feel that we have to take either side of that bet. Many investments, both inside and outside the US, are priced to do well in the long run whether AI lives up to expectations or not. We’re leaning heavily into those areas, with the goal of doing well no matter how the great AI boom plays out.