AI Is a Bubble? Fundstrat Says That's Exactly Why It Isn't

Let's set the scene. Broadcom CEO Hock Tan compared AI investing to gambling. Ray Dalio dropped the B-word (bubble) and threw in a 2008 comparison for good measure. The Philadelphia Semiconductor Index fell 2.7%, $SK Hynix and Samsung each cratered roughly 9%, and some Taiwan year-to-date leaders gave back 8 to 9% in a single session. Chaos, right?

Not so fast. Fundstrat's latest YouTube take cuts against the doom narrative with an argument that's genuinely worth sitting with: the very fact that everyone is calling this an AI bubble is evidence it probably isn't one. Real bubbles, the argument goes, form in silence, when complacency is so thick nobody is asking the uncomfortable questions. When half of financial YouTube is screaming bubble, that's not euphoria. That's anxiety. And anxious markets don't tend to blow up the same way as complacent ones.

Bloomberg's Big Take offers the scarier counterpoint, and it's well-sourced. Global imbalances are near 150-year highs, with AI chip dollars flowing to Taiwan and South Korea, whose private investors are recycling that capital straight back into US AI hyperscalers. Unlike 2001, when cautious central banks parked surplus savings in safe Treasuries, today's capital is being routed by nimble private investors into high-risk AI plays. If sentiment flips, the feedback loop unwinds fast. The Bloomberg team explicitly calls this dynamic reminiscent of pre-2008 capital flows.

The Traveling Trader and Joseph Carlson are both in the buy-the-dip camp, and they have numbers to back it. S&P 500 Q3 2026 earnings growth is estimated at 29.6%, which would be the third straight quarter above 25%. Carlson flags $ASML specifically, projecting 50 to 55% revenue growth and roughly 89% EPS growth, calling it a monopoly in semiconductor lithography with nowhere near enough fear priced in. Meanwhile, KCM Trade's Tim Waterer sees 5.5% on the 10-year Treasury as the line in the sand: stay below it and equities hold, cross it and bonds start eating lunch.

Where do the creators actually agree? Three things: earnings matter more than vibes right now, hyperscaler capex guidance during this earnings season is the single most important data point to watch, and the riskiest AI trades, the blue-sky cyclical names that got promoted to structural status on pure sentiment, deserved the haircut they got. $META releasing its Llama 4 frontier model and gaining 3.4% while chipmakers bled is the market telling you exactly that: picks-and-shovels speculation is getting punished, but the actual AI platforms are still getting paid.

Turns out the most dangerous trade isn't believing in AI. It's believing in AI and not knowing which part of it you actually own.