Author: Chasing the Trend Trading Desk, Wall Street News
The AI market rally is not a simple replay of the 1999–2000 bubble. Goldman Sachs believes the more critical question now is that while earnings and capital expenditures are still being revised upward, market prices have already priced in a large amount of optimistic expectations, and investor sensitivity to narrative changes is increasing.
According to the Chasing the Trend Trading Desk, Goldman Sachs noted in a June 22 research report that the AI investment boom may continue, with recent market expectations for its scale even requiring further upward revisions. However, the report also pointed out that much of the value has already been priced in, making the market more vulnerable to any news challenging the optimistic AI narrative.
The main risk in AI trading is no longer just a "valuation bubble." Forward P/E ratios have not significantly run out of control, due to simultaneous upward revisions in earnings expectations. What truly needs to be tested is whether the current strong earnings can be sustained after the capital expenditure cycle peaks.
For investors, strong earnings may continue to outweigh valuation concerns until the peak of the AI investment cycle appears. However, as incremental market value becomes increasingly reliant on optimistic assumptions, stock volatility may rise further, and the value of downside protection is also increasing.
AI is Not 1999, but the Market Has Run Ahead of the Macro Picture
Goldman Sachs's core assessment is that today's AI cycle is not like 1999–2000, which was built on extreme valuation expansion, macroeconomic overheating, and financing imbalances.
Current fundamentals are not significantly deteriorating; they are even strengthening. AI-related companies have strong earnings, capital expenditure plans continue to be revised upward, and the market therefore has reasons to keep buying related assets. Compared to the late 1990s, forward valuations have not shown the same degree of runaway expansion.
But this does not mean the risk is lower. The market cap growth of AI-related companies has clearly outpaced baseline macroeconomic calculations. To justify current prices, one must assume AI winners can secure a higher-than-normal share of productivity gains for an extended period.
In other words, the core bet of the current market is not that "valuations can expand infinitely," but that "exceptionally high earnings can persist."
What Resembles the 90s is Investment Intensity; Other Bubble Signals Have Not Yet Appeared in Unison
The late stages of the 1990s tech bubble had four typical signals: investment remaining at abnormally high levels, declining macroeconomic profit margins, rapidly rising corporate financing needs and leverage, and a widening current account deficit.
Currently, the only clearly evident signal is the first one: accelerating AI capital expenditures. The report states that tech investment as a share of GDP has already surpassed the 1990s peak, and its growth rate is faster. Hyperscale cloud providers' expectations for 2026 capital expenditures have increased by nearly 80% compared to six months ago. On the current trajectory, AI-related investment could approach or even exceed the peak of the 1990s tech investment boom in the coming years.
However, this capital expenditure cycle still differs from that of the past. First, its duration has not yet reached the length of the late 1990s. Second, its coverage is not as broad. The 1990s tech investment resembled a broad-based economic expansion, whereas today's AI capital expenditures are more concentrated among hyperscale cloud providers, semiconductors, and the related infrastructure chain.
The most crucial macro-level contrast lies in profits.
In the late 1990s, corporate profit margins peaked and began declining after 1997, eroded by rising wages and unit labor costs. The current situation is different. The corporate profit share of GDP remains near highs, and productivity growth has not been completely offset by wage acceleration similar to back then.
Corporate financing has also not followed the same path. Free cash flow for hyperscale cloud providers has declined noticeably, and the share of capital expenditures to operating cash flow has risen sharply. However, for the entire corporate sector, the gap between savings and investment has not significantly deteriorated because profit growth has largely offset the rising investment rate.
External imbalances are also different. In the late 1990s, the U.S. current account deficit widened; currently, the deficit is actually narrowing. At least from the perspective of macroeconomic imbalances, the current AI cycle has not yet developed the typical cracks seen at the end of the previous bubble.
$27 Trillion in Market Cap Increase, Exceeding the Baseline Macro Ledger
Changes at the market level are more aggressive.
Since the end of November 2022, the incremental value of AI-related companies is approximately $27 trillion, higher than the level of around $19 trillion in November 2025. Meanwhile, traditional U.S. equity valuations remain near historical highs; the Shiller CAPE ratio has only been higher at the end of 1999 and in 2000.
However, there is a key difference between this rally and 1999: earnings expectations are also being revised upward rapidly. Because EPS expectations have risen, even as stock prices continue to climb, forward P/E ratios have not increased in parallel this year. Recent gains have been driven more by earnings than by pure valuation expansion.
The problem is that the macroeconomic ledger does not provide support of comparable magnitude. Baseline calculations show that the present value of new capital income for the U.S. economy from AI productivity gains is about $9 trillion. Even using a more conservative market definition, focusing only on "pure AI" companies, the related value increase is about $14 trillion. Adding 25% of the incremental value from other AI-related companies brings the total to about $17 trillion, still above the baseline calculation.
To Justify Current Prices, One Must Bet on Winners Keeping a Larger Long-Term Profit Share
Current market prices are not entirely inexplicable, but they require more optimistic assumptions.
These assumptions include: faster AI adoption, higher productivity gains from AI, capital capturing a larger share of the economic benefits, or U.S. companies securing a larger portion of global AI revenues.
One optimistic scenario outlined in the report is: U.S. companies capture 50% of global related revenues, the capital income share is significantly above the economic average, AI adoption is faster, and the discount rate is lower. Only if multiple conditions hold simultaneously does the potential value more easily cover the current market cap increase.
The most compelling optimistic narrative is that AI-related companies can maintain a higher share of productivity gains over the long term. So far, this narrative has indeed been supported by earnings. Strong profits and high margins for semiconductor companies, cloud providers, and infrastructure beneficiaries are precisely what are supporting the market.
But this is also the point of vulnerability. In the early stages of a productivity acceleration, profit shares typically rise; over a longer horizon, competition, investment expansion, and a new wave of innovation may erode excess returns. While the AI industry has high concentration and technical characteristics that may favor capital owners, how long the barriers for current winners can last remains an open question.
The Greatest Risk Shifts from "Valuation Bubble" to "Earnings Bubble"
The AI investment boom itself is generating substantial profits. Companies selling chips, computing power, and building data centers directly benefit from rising capital expenditures. As long as the investment peak is not yet in sight, upward earnings revisions may continue to outweigh valuation concerns.
However, if the market directly extrapolates the strong profits of the next two or three years far into the future, risks will rise. Capital expenditures cannot grow at the current intensity forever. Once the investment cycle peaks, it may become harder to gauge the earnings trajectory for the companies currently benefiting most directly.
This is also why "forward P/E not being expensive" does not necessarily mean cheap. Cyclical industries and commodity companies often appear inexpensive at the peak of their cycles because the earnings denominator is too high. Whether the AI infrastructure chain will face a similar issue depends on how long the investment intensity can last, how quickly AI benefits materialize, and whether technological innovation reduces reliance on high-intensity capital spending.
AI May Be Masking Weakness in the Non-AI Economy
Compared to the 1990s, there is another important difference in the current macroeconomic backdrop.
In the late 1990s, U.S. domestic demand was extremely strong; in the final two years, real domestic demand grew at an annualized rate of nearly 6%, with robust consumption, residential investment, and non-tech investment. Capital inflows from the Asian and emerging market crises, a strong dollar, and global commodity price deflation actually masked overheating within the U.S., prolonging the cycle.
The current situation is the opposite. The U.S. economy outside of AI is not that strong. Non-tech investment is weak, consumption growth is far below the late 1990s level, and real disposable income grew at an annualized rate of about 1% over the past two years, compared to 5%–6% in the late 1990s.
This suggests that the AI boom may not be adding fuel to an already overheated economy, but rather offsetting weakness in areas outside AI. Consequently, the kind of extreme bubble seen in 1999–2000 and the typical imbalances before the 2001 recession might be less likely to appear. However, if the AI narrative faces setbacks, the non-AI parts may not provide sufficient support.
Volatility Shifting Gears, Portfolios Need More Downside Protection
Market structure has already begun to change.
Credit spreads remain tight, differing from the path of gradually rising credit pressure in 1998–2000. But stock volatility has begun to rise more noticeably. Over the past few months, single-stock implied volatility has increased, U.S. single-stock option skew has moved lower, and demand for call options relative to put options has risen.
At the same time, implied correlation has fallen to very low levels, suppressing index volatility, but long-term index volatility has also been creeping higher. Gains are also more concentrated. Broad index performance remains more moderate than in the late 1990s, but the gains in the semiconductor index over the past few years have approached the performance of the Nasdaq in its later stages. In April and May, the consecutive two-month gains for the Nasdaq, South Korea, Taiwan, the SOX semiconductor index, and a basket of unprofitable tech stocks all reached multi-year highs.
As long as the investment cycle peak has not yet arrived, strong earnings may continue to dominate the market. But as prices become increasingly dependent on optimistic assumptions, the value of downside protection increases. In terms of strategy, it may be more about staying in the trade while using put protection or replacing some spot exposure with call options to control drawdowns.
There is also a countervailing risk on the interest rate side: if the non-AI economy's vulnerabilities are exposed after the AI investment peak passes, the probability of a significant decline in interest rates at that time may be higher than usual.
