Ray Dalio Says AI Is the Biggest Bubble in History. The Data — and Bridgewater's Own Portfolio — Say Otherwise

Ray Dalio has a simple story: artificial intelligence has created the greatest financial bubble in history — larger than 1929, larger than the dot-coms. Investors are buying revolutionary technology at any price, often with borrowed money. Companies are trapped in a CapEx arms race, spending hundreds of billions without a guaranteed return.

Fernando Sánchez's response is equally simple: check the data — and check what Bridgewater's money is doing.

The Prophet's Track Record

Before accepting a bubble call from a famous name, Fernando revisits Dalio's forecasting history — summarized, not unfairly, as predicting "50 of the last two crises."

In 1982 he foresaw a depression; what followed was one of the great bull markets of the century. Between 2015 and 2019 he repeatedly warned of a recession that did not arrive. In January 2020 he said cash was trash — immediately before the COVID crash, when cash was precisely what investors needed.

Incentives matter too. Dalio no longer runs Bridgewater day to day. He sells books and conference stages. Apocalyptic headlines travel farther than measured ones.

None of that proves he is wrong this time. It does prove that a Dalio warning is not, by itself, a reason to liquidate a portfolio.

The Numbers Are Not 2000

Against the rhetoric, Fernando stacks valuation and earnings reality:

Forward multiple. The S&P 500 trades at about 19.6x estimated earnings — somewhat above its long-term average of 17.2x, and nowhere near the 25x reached before the dot-com bust.

What drove the rise. From 1995 to 1999, much of the advance was multiple expansion: paying more for the same earnings. Over the past five years, roughly 78% of the technology sector's rise came from real earnings growth.

Margins justify a higher multiple. S&P 500 companies now post a record average net margin of 16.7%, versus 10.9% at the end of 2020. Higher structural profitability supports higher valuations than thirty years ago.

Concentration is not the 2000 kind. The top 10 companies are about 39.1% of the index — and generate about 41.8% of its earnings. In 2000 the top names were roughly 25% of the weight and only 15% of profits. Today's concentration is concentrated earnings, not concentrated hope.

A bubble, in Fernando's closing definition, is not prices going up. It is prices going up without earnings behind them.

Where the AI Money Actually Sits

Fernando splits the AI ecosystem into buckets so investors stop treating "AI" as one trade:

Hyperscalers / cloud. Growing around 55% a year at roughly 21.5x earnings, with recurring contracts. This is where he sees attractive value relative to growth.

Hardware and software. Software grows slower and looks more expensive. Hardware carries richer multiples (around 26.9x).

Semiconductors. They can look "cheap" at about 18.9x while growing over 100% — and still be a value trap near a cycle peak. Cyclical businesses at maximum earnings often print their lowest multiples just before the cycle turns. Low PE is not the same as a bargain.

That last point lands directly on the July memory wipeout and on any urge to average aggressively into HBM names solely because the multiple "looks" low.

The Bridgewater Contradiction

Here is the detail that turns a valuation debate into a credibility check.

While Dalio warns of an AI bubble, the firm he founded has been buying AI-related equities. In the latest quarterly report Fernando cites:

• Nvidia raised from about 2.6% to 3.7% of the portfolio
• Broadcom, Micron, and Amazon increased — Amazon's weight roughly doubled to 4%
• A large new position in Taiwan Semiconductor
• Sales of software names such as Salesforce

Talk bubble. Buy the stack. That does not make Dalio a hypocrite in every moral sense — founders and operating CIOs are not the same person — but it is a brutal filter for retail investors: when the guru announces the apocalypse, look at what the money does, not only at what the books say.

Three Rules — and What They Mean Now

Fernando closes with three rules that cut through both panic and euphoria:

1. A bubble is not rising prices. It is rising prices without earnings.
2. Beware low multiples in cyclical businesses at cycle highs — they are often traps.
3. When a guru sells the end of the world, read the 13F before you sell your shares.

For investors who just lived through July's CapEx panic, Korea's leveraged ETF disaster, and Aschenbrenner's forced sale, the message is consistent with Mutuactivos and with Cava's longer 2026–2027 growth case: the AI story did not die; the era of buying anything labeled "AI" without asking for returns did. Hyperscalers with contracts and margins are not the same trade as leveraged cyclical semiconductors at the wrong point in the cycle. And a television warning from Ray Dalio is not a sell ticket — especially when Bridgewater's own book is still pointing at Nvidia, Amazon, and TSMC.

---

This analysis is based on Fernando Sánchez's commentary on Ray Dalio's AI bubble claims, August 2026. For informational purposes only — not financial advice.