Key Takeaways:
The current AI stock rout has followed the same three-part pattern that preceded every major tech correction during the dot-com era, according to CICC.
The Philadelphia Semiconductor Index has fallen more than 20% from its June record high, entering bear market territory, as the global AI trade faces its third major correction since ChatGPT launched in late 2022. The Nasdaq 100 has shed 9.7% of its forward price-to-earnings multiple, which now sits at 22.4.
"The triggers — overheated valuations, macro headwinds, and industry-specific setbacks — are nearly identical to the four corrections the Nasdaq endured before the 2000 bubble burst," CICC strategists wrote in a July 27 research note.
South Korea's stock market, which had the highest concentration of leveraged retail AI bets, has seen margin debt fall 15% and leveraged ETF assets drop 53%, according to Kofia and EPFR data. The unwind follows a period when the Philadelphia Semiconductor Index's 14-day relative strength index exceeded 75, a level that historically preceded each of the four dot-com corrections.
The question for investors is whether this correction follows the historical pattern — where each of the four dot-era pullbacks was followed by a renewed rally — or marks the beginning of a structural unwind. The answer hinges on three variables: earnings catalysts, Federal Reserve policy, and the pace of AI monetization.
Four corrections, three common triggers
Between 1995 and early 2000, the Nasdaq Composite suffered four significant drawdowns before the bubble finally burst in March 2000. The declines ranged from 13% to 26% and lasted two to three quarters each. In every case, the selloff was triggered by the same three factors: overheated valuations and sentiment, a macro shock, and an industry-specific disappointment.
The first correction in mid-1996 saw the Nasdaq fall 15% after a wave of internet IPOs pushed sentiment to extremes. Semiconductor capacity had expanded 70% year over year, driving DRAM prices down 62% in the first half of 1996. The selloff ended only after the Federal Reserve held rates steady against hawkish expectations and tech earnings proved resilient outside the DRAM-exposed names.
The second correction in early 1997 was triggered by a price war in network cards, with Intel cutting prices 40% and 3Com issuing a profit warning. The Nasdaq fell 13% before Cisco's earnings showed the damage was contained to standardized hardware, and the broader tech sector delivered 60% net income growth.
The third and fourth corrections — 14% in late 1997 and 26% in mid-1998 — were driven by the Asian financial crisis and the LTCM collapse, respectively. In both cases, the selloffs reversed after the Federal Reserve eased policy and tech earnings rebounded.
What's different this time
The current correction shares the same DNA but with one critical distinction: the macro environment is tighter. The Federal Reserve faces pressure to raise rates, not cut them. CICC's strategists noted that "the existing environment cannot completely rule out the possibility of a 'preventive rate hike' in September," even as the bank's own analysis suggests inflation data does not support tightening.
Brent crude oil has returned to the $90 to $100 per barrel range as Iran tensions escalate, adding another layer of cost pressure. Meanwhile, Google's latest quarterly filing showed free cash flow turning negative, highlighting the tension between massive AI infrastructure spending and the slower-than-expected revenue conversion from token-based products.
"Demand is still growing rapidly, but the economics of token replacement are expensive, slowing the pace of revenue conversion," the CICC strategists wrote. This means cash flow pressure will persist until either a breakthrough in demand materializes — similar to Anthropic's strong performance in early 2025 that reignited the AI rally — or hardware prices fall enough to make substitution economical.
Where investors can look
CICC recommends focusing on "bottleneck assets" — areas where supply constraints create pricing power regardless of the macro environment. The two largest themes are the US power shortage driven by data center demand and China's semiconductor self-sufficiency push.
For investors seeking lower risk while the AI trade resets, the report suggests rotating toward internet platforms and innovative pharmaceuticals, where the fundamental阻力 is smaller. Cyclical and export-oriented sectors should wait for US Treasury yields to decline before re-entering.
Nvidia shares, which have been the bellwether of the AI trade, now trade at roughly 28 times forward earnings, down from a peak of more than 40 times earlier this year. Advanced Micro Devices and Broadcom, two other key AI infrastructure plays, have seen similar multiple compression. Whether the current valuation reset is sufficient to attract buyers will depend on the earnings reports due in the coming weeks from Microsoft, Amazon, and Alphabet.
This article is for informational purposes only and does not constitute investment advice.
