A reversal of foreign capital flows is the most critical early-warning signal for Japan’s AI and semiconductor market

Overview

The report reviews academic studies related to the Internet bubble and distills bubble formation and collapse mechanisms from four angles: investor capital flows, arbitrage constraints, investor sophistication, and stock supply. It then applies this framework to the current AI and semiconductor rally in Japan. The authors conclude that there is no clear evidence that the Japanese market has yet reached the kind of broad overheating driven by retail investors and margin trading seen in the late Internet bubble; the rally remains supported mainly by foreign spot activity and momentum.

Core views

The core view is that bubble inflection points are typically not triggered by a single bad headline but by multiple factors that cause institutions to sell in a synchronized manner. The most important variable to monitor now is whether foreign capital flows reverse; if foreign investors turn net sellers while retail investors, investment trusts, and margin trading absorb the supply, market structure becomes closer to the fragile state seen around the collapse of the Internet bubble. Earnings revision trends, interest rates, financing supply in the AI ecosystem, and the strength of negative-news reactions are key additional variables for assessing whether the rally is weakening.

Analysis framework

The report uses a historical analog and literature-review approach, first summarizing the behavior of institutions, hedge funds, retail investors, and mutual funds during the Internet bubble, then mapping these patterns to observable indicators in the Japanese equity market, including foreign buy/sell flow, investment trust flows, margin purchase balances, IPO and refinancing activity, the rate environment, earnings revisions, and news reactions. It then discusses factor-level behavior of price momentum, earnings revision factors, and Earnings Window Momentum under conditions of continuation or reversal of the rally.

Methodology notes

• Historical analogyInternet bubble literature review

  Uses capital-flow dynamics and supply shocks from the 1997–2002 technology bubble to explain potential turning points in the current AI rally.

  The report cites research by Griffin et al., Brunnermeier and Nagel, Abreu and Brunnermeier, Greenwood and Nagel, Ofek and Richardson, highlighting that institutions do not necessarily correct bubbles, and may participate in bubbles and amplify the downside when exiting.

• Capital flow monitoringForeign capital flow reversal signal

  A shift from net foreign buying to net foreign selling is the primary indicator for judging whether the Japanese AI and semiconductor rally is entering a fragile phase.

  In the Japanese market, foreign investors can proxy institutional behavior; if foreign selling is absorbed by retail investors, investment trusts, and margin trading, the market may be transitioning from institutional-driven participation to a more fragile marginal-demand structure.

• Behavioral financeSynchronization risk

  Even rational investors may continue to hold during a bubble because of the opportunity cost of exiting too early and the risk of synchronized liquidation.

  The immediate mechanism of a bubble rupture is not the bad news itself, but that news or price decline causes arbitrage timing to align, eventually exceeding behavioral investors’ capacity to absorb supply.

• Fundamental factorsEarnings revision factor

  Compared with traditional price momentum, revision-backed momentum may be more resilient when the AI rally reverses.

  The report notes that in the 12 months after the Internet bubble burst, the price momentum factor experienced sharp drawdowns, whereas the revision factor was relatively robust; currently, earnings revision trends still support high-AI-beta stocks, but deceleration in revisions could also trigger valuation pressure.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Japan AI and semiconductor-related stocks

  Core research focus

  Strengths

  Profits and cash flow are stronger than many loss-making technology firms in the Internet bubble era; the earnings revision trend remains relatively positive, with foreign buying and momentum still supporting the rally.

  Weaknesses

  Valuation already embeds high positive expectations, making the market more sensitive to slower earnings revisions, rising rates, and amplified negative-news reactions.

  Comparison

  Similarities to the Internet bubble include a long-run growth technology narrative, concentrated high-valuation growth stocks, and momentum dominance; differences include that current levels have not shown equivalent overheating in retail, mutual funds, and margin trading.

  Risks

  Foreign investors turning net sellers, retail investors and investment trusts stepping in as absorbers, increased AI-related IPOs and refinancing, downward revisions to earnings expectations, and amplified reactions to negative news.

• Traditional price momentum factor

  Beneficial in rally continuation, pressured in rally reversal

  Strengths

  May continue to generate returns by following market trends during the AI and semiconductor upcycle.

  Weaknesses

  If the thematic rally reverses, it is prone to significant mean reversion and drawdowns.

  Comparison

  The 12-month price momentum factor showed a pronounced correction after the Internet bubble burst.

  Risks

  Thematic crowding, synchronized institutional selling, and rapid deleveraging triggered by negative news.

• Earnings revision factor and Earnings Window Momentum

  Candidate factor with relatively higher resilience

  Strengths

  Supported by improving fundamental expectations, and historically has smaller pullbacks than traditional price momentum, potentially participating in upside during continued AI rallies.

  Weaknesses

  If the earnings revision trend shifts from rising to slowing or downward revisions, factor performance would also come under pressure.

  Comparison

  The report argues that the revision factor was more robust than price momentum after the Internet bubble, and a similar relative performance profile may appear currently.

  Risks

  Overly high earnings expectations, deceleration of revision momentum, and overreaction by the market to even minor downward revisions.

Key data

• Report date2026-07-06The report was published on July 6, 2026 at 11:10 AM GMT.
• Institutional buying share during the uptrend of the Internet bubble63.6%Griffin et al. show that from January 1997 to March 2000, institutional investors accounted for 63.6% of active buying in technology stocks.
• Direct retail buying share after the Internet bubble peak49.0%From April 2000 to March 2001, retail investors became the major marginal buyer, with 49.0% bought directly and another 14.6% purchased through mutual funds.
• Hedge fund technology stock weight29% vs 17%In September 1999, hedge fund weight in technology stocks was about 29%, above the 17% market-combination benchmark, but relative over-allocation had begun to narrow about six months before the NASDAQ peak.
• U.S. Fed rate hikes during the Internet bubble period4.75% to 6.50%From June 1999 to May 2000, the federal funds target rose from 4.75% to 6.50%, putting pressure on high-valuation growth stocks.
• Current rate outlookNo further rate hikes expected in 2026, with rate cuts expected to resume in March 2027The report cites Morgan Stanley U.S. economists, noting that the current rate environment differs markedly from the tightening cycle around 2000.
• Margin purchase balanceAt an absolute high since August 2002, but not clearly overheated relative to market capThe report believes margin activity overall has not yet shown broad overheating comparable to the Internet bubble, though AI, semiconductor, and data-center themes still require monitoring on a segmented basis.
• Earnings revision for high-AI-beta stocksTrending higher since mid-last yearStrong earnings and positive revisions have helped investors maintain confidence in AI and semiconductor themes, but slowing revision momentum could create price pressure.

Impact & implications

For Japanese equities, as long as the AI and semiconductor rally persists, market preferences for growth, high momentum, large-cap and high volatility stocks may continue. Once the rally reverses, these factors may face mean-reversion pressure. From an investment perspective, foreign capital flow, the direction of earnings revisions, AI ecosystem financing supply, rate changes, and reactions to negative news should be treated as portfolio risk-monitoring metrics, with emphasis on fundamental-support-driven momentum factors rather than simply chasing price momentum.

Risks

• Foreign capital flow shifts from net buying to net selling.
• Retail investors, investment trusts, and margin trading become the main counterparties absorbing supply.
• Increased IPOs, refinancing, or convertible bond issuance by companies related to AI models, cloud computing, data centers, and storage create additional supply pressure.
• Rates rise significantly, lowering discounted future cash flows and worsening external financing conditions.
• The upward trend of earnings revisions in high-AI-beta stocks slows or turns negative.
• The market begins to respond more aggressively to negative news on cash burn, regulation, capex, pass-through of costs, and lock-up expiries.

What to watch

• Whether foreign spot capital flows in Japanese equities continue to weaken.
• Whether investment trusts, retail investors, and margin purchase balances rise in sync and absorb foreign selling.
• Supply from IPOs, secondary offerings, convertibles, and lock-up expiries in the AI and semiconductor value chain.
• Rate paths in the United States and Japan, especially the upside risk to real rates and long-end yields.
• The direction and pace of earnings revisions for high-AI-beta stocks.
• Whether market reactions to news on regulation, hyperscaler capex, end-product price pass-through, and cash-flow quality shift from being positive-biased to negative-biased.