Is AI a Bubble? Lessons from Howard Marks' Latest Memo
Is this time different?
It's the four most expensive words in investing. And right now, they're being whispered about artificial intelligence.
NVIDIA has become one of the most valuable companies on Earth. OpenAI is valued at over $150 billion despite never turning a profit. Billions of dollars are flowing into data centers, chips, and AI infrastructure. The excitement is palpable.
But is it rational?
Howard Marks, co-founder of Oaktree Capital, legendary investor, and author of some of the most respected investment memos in finance, recently tackled this question head-on in his memo titled "Is It a Bubble?". His answer is nuanced, historically grounded, and worth understanding whether you're invested in AI stocks or simply trying to make sense of the moment.
Here's what every investor should take away.
The Two Types of Bubbles
Marks begins with a crucial distinction that most commentary misses.
Not all bubbles are created equal. He identifies two fundamentally different types:
Inflection bubbles fund the infrastructure that changes the world, even as they bankrupt the investors who paid for it. The fiber optic cables laid during the dot-com boom? They powered the internet economy for decades. The investors who funded them? Many lost everything.
Mean-reversion bubbles are pure speculation. Prices rise because prices are rising. When sentiment shifts, nothing remains.
The question for AI isn't whether enthusiasm exists. It clearly does. The question is: which type of bubble are we in?
Why Marks Can't Give You a Simple Answer
Here's where Marks' intellectual honesty shines. Despite 50+ years of investing experience, he admits he cannot determine whether AI valuations are justified. The answer depends on information that doesn't exist yet.
The critical unknowns:
1. Who will win? History shows that early leaders often lose to later entrants. The dominant buggy manufacturers didn't become car companies. Today's AI leaders may not be tomorrow's.
2. Will anyone make money? Competition tends to commoditize technology. Will AI services become profitable businesses, or will margins compress to zero as everyone races to offer similar capabilities?
3. How fast will adoption happen? Data center construction is booming. But demand forecasts for AI computing are essentially guesses. Overbuilding infrastructure for demand that never materializes is a classic bubble pattern.
4. Will the assets last long enough? AI chips and data centers are being financed with long-term debt, sometimes 30-year bonds. But will today's chips be useful in 10 years? 5 years? The obsolescence timeline is genuinely unknown.
"The speed and scale of improvement mean it's incredibly hard to forecast demand."
Howard Marks
The Debt Problem: Why This Time Might Be Riskier
Marks identifies something genuinely concerning about the current AI boom that differs from previous technology cycles: the extensive use of debt financing.
Technology investments have traditionally been equity-funded. This makes sense. Most tech bets fail, but the winners win big. Equity investors can lose everything on their losers but capture unlimited upside on their winners. The math works.
Debt doesn't work the same way.
With debt, you absorb full losses when borrowers fail but only collect interest payments when they succeed. There's no upside to offset the inevitable failures.
Marks flags several warning signs:
- •Hyperscalers' capex is outpacing revenue growth: companies are spending faster than they're earning
- •SPV structures are obscuring balance sheet obligations: complex financial engineering to hide leverage
- •Circular deals: money "round-tripping" between related parties, reminiscent of Enron-era accounting
- •Unprofitable startups borrowing to build infrastructure for other unprofitable startups: a fragile chain
OpenAI has announced over one trillion dollars in investment commitments. It has never been profitable. This is being financed, in part, by debt instruments that assume decades of stable cash flows from a technology that didn't exist three years ago.
Historical Context: How Does AI Compare?
Marks provides valuable perspective by comparing current valuations to previous bubbles:
Notably, today's AI leaders trade at lower multiples than the internet stocks of 1999. This suggests less irrational exuberance by pure valuation metrics, though Marks cautions that this alone doesn't mean prices are reasonable.
The internet bubble taught us that transformative technology and terrible investments can coexist. The technology can be real, the revolution can be coming, and you can still lose most of your money.
Marks' Bottom Line: Exuberance Without Certainty
After weighing the evidence, Marks reaches a characteristically measured conclusion:
"Investors are exhibiting clear exuberance about AI, but whether it's 'irrational' cannot be determined until years later."
This reflects intellectual honesty about the limits of prediction.
His practical advice for investors:
Don't go all-in. The risk of ruin is real if the AI thesis disappoints.
Don't stay all-out. The risk of missing a generational opportunity is also real.
Use prudent judgment, not FOMO. Moderate, selective positions based on careful analysis, not fear of being left behind.
The Deeper Worry: What Happens to Jobs?
Marks closes with a concern that extends beyond portfolios: What happens to employment?
Studies suggest AI could automate 43 percent of work tasks across jobs. Unlike previous technological revolutions, it's unclear what new jobs will replace the displaced ones. Previous automation eliminated physical labor and created cognitive jobs. AI may eliminate cognitive work without a clear replacement.
Marks worries about:
- •Widespread joblessness
- •Social division
- •Political instability and populist reactions
- •Whether "universal basic income" can replace the psychological benefits of meaningful work
He openly appeals to optimists: "Will you please explain why I'm wrong?"
Beyond investment analysis, this raises a genuine question about the kind of society we're building.
What This Means for Your Portfolio
If you own AI stocks or are considering buying them, Marks' framework suggests several principles:
1. Acknowledge uncertainty. Anyone claiming to know exactly how AI will play out is overconfident. Size your positions accordingly.
2. Prefer equity to debt exposure. If you want AI exposure, stocks at least offer upside. Bonds offer all the downside with capped returns.
3. Diversify across the value chain. Winners and losers will emerge. Spreading bets reduces the risk of picking the wrong horse.
4. Watch the debt markets. If AI-related debt starts showing stress, it may be an early warning signal before equity markets react.
5. Think in decades, not quarters. The internet took fifteen-plus years to generate the returns that justified 1990s valuations. AI may follow a similar timeline.
Read the Full Memo
This summary captures the key insights, but Marks' original writing is worth reading in full. His memos are free, accessible, and represent some of the clearest thinking in finance.
Read "Is It a Bubble?" on Oaktree Capital
Howard Marks has been writing investment memos since 1990. His book "The Most Important Thing" is considered essential reading for serious investors. This summary is provided for educational purposes and does not constitute investment advice.
The Signal
Curated perspectives from respected voices across the investment landscape. Ideas that challenge, clarify, and deepen understanding.