This may be the most important signal that the AI boom has peaked—a lesson borrowed from Japan during the Internet bubble era.

This may be the most important signal that the AI boom has peaked—a lesson borrowed from Japan during the Internet bubble era.

AI trends continue to deepen, but the question of a bubble always lingers in the market.

Morgan Stanley’s Japan quantitative strategy team released a report on July 6, stating: To judge whether Japan’s current AI trend is approaching a turning point, the key is not the valuation, but the marginal change in the structure of capital flows.

The team reviewed the history of the 2000 US stock internet bubble and found that the Nasdaq tech rally actually peaked, not simply because of high valuations, but because US-based professional institutions collectively reduced their holdings and retreated at the same time. In the Japanese market, overseas institutions are the core forces influencing overall market trends.

For investors, the most important warning signal to monitor in the Japanese stock market is overseas capital flows: Once foreign investors shift from continuous net buying to net selling, domestic retail investors, local investment trusts, and other domestic funds become the passive buyers of large volumes of sell-off, the capital structure of the Japanese market will mirror the high-risk pattern seen before the US internet bubble burst.

However, Morgan Stanley also notes that the Japanese market currently does not display the typical widespread overheating of retail investors seen in the late stage of the US internet bubble. Although the current Japanese AI and semiconductor rally has shown the common characteristics of a bubble phase, such as funds clustering and strong performance from high-growth stocks, the scale of retail capital, domestic fund subscriptions, and margin trading volume have not reached the overheated levels of the 2000 US bubble peak, meaning the current AI rally still lacks key triggers for a substantive turning point.

When the internet bubble really ended, institutions exited first, and retail investors took over

The market commonly believes that the US internet bubble was created by frenzied retail buying. But Morgan Stanley, after reviewing classic academic studies, found that this wasn’t the case.

During the rapid expansion period of the bubble from 1997 to 2000, institutional investors were the biggest buyers of tech stocks, accounting for 63.6% of active buying, far surpassing retail investors and mutual funds. Hedge funds took on the most aggressive tech stock allocations, but notably, they didn’t short early due to high valuations; instead, they continued to participate in the rally and only gradually reduced positions before the market peaked.

The real turning point happened after the market peaked in March 2000. As institutional funds began to withdraw, their share of new tech stock buying dropped sharply to 36.4%; meanwhile, retail’s direct buying rose to 49%, with another 14.6% coming from mutual fund flows. The market structure gradually evolved into retail investors buying what institutions sold, and tech stocks entered a continuous decline.

Therefore, Morgan Stanley concludes, The bubble burst was not because “smart money” anticipated the high valuation bubble, but because institutions chose to collectively exit at the same time.

Four changes that might prompt institutions to retreat in unison

The report states that changes in stock supply, interest rates, profit expectations, and market sentiment can all act as catalysts for institutions to adjust their positions.

Stock supply: Financing pressure is accumulating

During the internet bubble, numerous IPOs, lock-up expirations, and insider selling were key catalysts for the market top. Currently, AI industry chain IPOs, share issuances and convertible bond financings have increased significantly. While it is still too early to judge whether this will evolve into systemic risk, the pressure from stock supply is worth continued attention.

Interest rates: Growth stocks are more sensitive to financing conditions

The internet bubble peak coincided with the Fed’s rate hike cycle. By contrast, the market now expects no further rate hikes this year, and expects rate cuts to resume by March 2027. But Morgan Stanley points out, If interest rates move significantly higher than expected, high-valuation AI growth stocks may still face valuation pressure.

Profit: Watch for slowing upward revisions in earnings

Unlike the internet bubble era, the current round of AI leaders have a more solid profit base, and earnings forecasts have been consistently revised upward, which is an important support for the rally. Still, Morgan Stanley reminds us, More important than the profit level itself is whether upward earnings revisions start to slow, as this often influences stock prices first.

Market sentiment: Are bad news stories becoming amplified?

In the later stages of the internet bubble, the market became increasingly sensitive to negative news about regulation, financials, and lock-up expiry. Similar signs are appearing in the current AI industry chain, as relevant stocks are showing greater volatility in response to IPOs, capital expenditures, and regulatory news. Changes in sentiment are worth monitoring.

The key to watch is whether overseas funds begin to retreat

Compared to the four potential catalysts above, Morgan Stanley believes the real determinant of whether the rally is at an inflection point is still capital flows themselves. Before the internet bubble burst, the Japanese market saw continuous net selling by overseas investors, while investment trusts continued to buy, with market supply gradually taken up by domestic funds.

By contrast, the current Japanese market presents a completely different picture. Overseas investors are still maintaining strong buying, and investment trusts have not seen significant continuous inflow; retail investors overall are still net sellers, and margin buying balance has reached its highest level since 2002, but as a percentage of total market capitalization, is not significantly higher than historic levels, nor has levered overheating as seen in the late-stage internet bubble appeared.

In other words, while the current AI rally shares many similarities with the late internet bubble, including growth stocks leading and capital concentrating in a few leaders, retail-driven overheating has not yet appeared, and the market structure remains meaningfully different from before the 2000 bubble burst.

Risk Warning and DisclaimerThe market has risks, investments require caution. This article does not constitute personal investment advice, nor does it take into account individual users' special investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article are appropriate to their specific circumstances. Investing based on this is at your own risk.