The AI Bubble

“Some development, seemingly new and desirable…captures the financial mind…the price of the object of speculation goes up…this increase and the prospect attract new buyers…the speculation building on itself provides its own momentum…There are those who are persuaded that some new price-enhancing circumstance is in control …a new world of greatly, even infinitely increasing returns…then there are those…who perceive the speculative mood of the moment. They are in to ride the upward wave; their particular genius…will allow them to get out before the speculation runs its course.”
John Kenneth Galbraith “A Short History of Financial Euphoria”

Calling a bubble is next to impossible. Timing one is even harder.

As the title of this letter suggests, we are, perhaps foolishly, planting our flag: we have officially entered the AI bubble. 

We’ve questioned the sustainability of AI spending for a while, but our skepticism crystalized two weeks ago when Oracle announced its second quarter earnings. The results were modestly below expectations, but the stock jumped 40%. Why? Oracle’s future performance obligations soared to $455bn, a 359% jump from last quarter, driven largely by a single $300bn contract with OpenAI. OpenAI may be growing rapidly, but it remains deeply unprofitable, which raises the obvious question: where will the money come from?

Does Money Grow on Trees?

Not long ago, the obvious answer was big tech. Today, however, those companies have already committed billions to new investments that consume, and even exceed, their cash flow. Oracle takes the cake, burdened by the group’s weakest balance sheet.

The only way to fund these new investments is to issue stock, get other people to pay for them (reducing profit), borrow money or get creative. History shows that such creativity works until it doesn’t.

We’re already seeing it. Nvidia guarantees CoreWeave’s unsold cloud capacity — selling chips but also backstopping the risk if customers don’t appear. Just as we were writing this letter, Nvidia has announced plans to invest $100bn into OpenAI, all but ensuring itself a massive chip customer. Microsoft’s arrangement with OpenAI is more subtle but no less circular: its billions in funding effectively pay for the very cloud capacity OpenAI records as revenue. None of this is illegal, and Nvidia can certainly afford it, but it’s growing increasingly difficult to follow the money.

Cloud Competition

Whether or not Oracle is successful, it’s now abundantly clear that there are no guaranteed AI winners. The cloud, once dominated by Microsoft & Amazon, is now crowded.

Oracle is staging a comeback, Google is doubling down, and dozens of NeoClouds like CoreWeave are racing to build. Collectively, they’ve spent billions, with billions more committed, yet no single player holds a decisive edge. For much of big tech, this marks a new paradigm, where growth is increasingly fueled not by cash flow, but by debt.

Big Numbers

Just how much equity is at risk if minor bumps turn into major potholes? Nvidia is worth $4.5 trillion. If we add in Apple, Amazon, Google, Microsoft, Tesla, Broadcom, Meta, and Oracle, the total climbs to $23 trillion. That figure excludes the data centers, component suppliers, HVAC manufacturers, power generators, and non-public entities like xAI, OpenAI, and Anthropic. All in, it’s about the size of the entire U.S economy at $29 trillion.

Sheer size is not inherently a problem; these companies are some of the greatest businesses of all time. But when large valuations rest on equally large expectations, the setup becomes fragile. Nowhere is this more apparent than in the chipmakers where Nvidia and Broadcom command over $6 trillion in value. Broadcom, a company Compass clients purchased a decade (or more) ago trades for 65x free cash flow. We paid 12-18x.

What is a bubble

Last week, Meta CEO Mark Zuckerberg said he’d risk, “misspending a couple of hundred billion dollars” rather than miss out on AI. Coming from one of the world’s richest men—whose company already burned $100bn on virtual reality—the hubris is striking.

In A Short History of Financial Euphoria, John Kenneth Galbraith identified recurring themes that appear in every major speculative boom:

1) Short financial memory
2) The link between money and intelligence
3) Something new in the world
4) Financial innovation & debt

By Galbraith’s framework, today’s moment checks all the boxes of a bubble. Yet calling one presents two challenges. The first is that you are wrong. The frenzy was justified, and the new thing strengthens the incumbents, justifying higher valuations.

The second is that you are right, but too early. As Howard Marks said, “One of the biggest mistakes you can make is to think that overpriced and going down tomorrow are synonymous. Markets that are overpriced often keep going.” 

What to do in a bubble

This newsletter is not an attempt to call the market top, or a forecast that everything will unravel tomorrow. Will it be a recession, a greater focus on returns, runaway power prices, Nvidia’s decision to invest directly in power plants, or something entirely different? We won’t know until after the fact whether Oracle’s moment was the equivalent of 1997 or 2000, when the .com bubble burst.

The very act of writing this letter probably suggests the party isn’t over, but we aren’t going to wait for the music to stop. Instead, we’re using the run-up in AI stocks to crystallize gains and rotate into more compelling opportunities. In undertaking this rotation, the key question to ask is who will capture the value created by AI.

To be blunt, if your business is “AI” – we are skeptical. On the other hand, if you have a business with existing moats that might be able to use these tools when they are refined and developed to deepen your moat, the potential to create lasting value is far greater.

Venture capitalist, Jerry Neumann makes the point in his essay AI Will Not Make You Rich. He compares AI infrastructure to the invention of shipping containers in the 1950s. Manufacturers and shipping companies profited at first, but over time it was the customers who captured real wealth. IKEA, then a small Scandinavian retailer, became a global powerhouse. Today, the combined market value of retailers like IKEA, Walmart, Costco, and Amazon far outweighs that of the global shipping industry.

By analogy, today’s containers and shippers are the chipmakers, data centers, and power producers. They are minting money, but valuations assume it will last forever. As Neumann warns, “A lot of the money pouring into AI is therefore being invested in the wrong places, and aside from a couple of lucky early investors, those who make money will be the ones with the foresight to get out early.”

How to invest in AI

For starters, you don’t have to chase AI to make money. Some of the best opportunities are in areas largely untouched by the frenzy. For example, integrated energy companies that generate sustainable cash flow, even at $60 oil, and return it to shareholders through dividends and buybacks. Many consumer businesses also offer reliable cash returns without the massive capital spending required to compete in AI.

One promising but more uncertain area is biotech and healthcare. Both sectors are under immense pressure, yet if AI fulfills even part of its potential, it could transform diagnostics, drug design, and treatment development. The timeline is unclear, but in the meantime, investors can buy into a sector with healthy cash flows at depressed valuations.

In cyclical sectors hit by a slowing economy, AI will likely accelerate the divide between winners and losers. The middle will continue to be hollowed out, making it even more important to identify the best operators with the discipline to invest in sustainable growth.

AI will reshape how all companies, not just tech, use their data. Businesses that organize, secure, and protect data will grow in importance, while those able to leverage data to expand their markets will steadily outrun competitors.

The hardest question to answer is whether a shift in sentiment will also drag down the broader market — and if so, how wide the pain could spread. With so much market value concentrated in just a handful of stocks, the risk is real. History shows that when a bubble bursts, even companies with no ties to the mania can get caught in the downdraft.

That’s why, rather than trying to predict the exact turn, we’ll focus on what we can control: investing in durable cash flows, resilient operators, and businesses that can harness AI as a tool to deepen their moats rather than depend on it for survival. This discipline won’t insulate portfolios from every bump in the road, but it gives us the best chance of compounding value long after the current frenzy has passed.

_{Compass Wealth Management LLC is a SEC registered investment advisor, clearing transactions primarily through Pershing Advisor Solutions and Pershing LLC subsidiaries of Bank of New York Mellon Corp. This letter is written by Compass for the benefit of its clients and does not necessarily represent the opinions of its affiliated organizations. It is based on information believed to be reliable, but which is not guaranteed to be correct. Nothing herein shall be construed to be a solicitation to buy or sell securities, indicate that past performance is predictive of future returns, or recommend individual investments.}