Has the AI bubble burst? Goldman Sachs: It's more like 1998 than 2000 right now

Tech stocks have faced a sharp sell-off, but whether this is due to leveraged liquidation or the invalidation of business models will dictate drastically different strategic approaches.

According to the Trend Chasing Trading Desk, in a November 6 report, Goldman Sachs FICC and equities trading desks noted that the Nasdaq plunged 5.5% in a single day, the Philadelphia Semiconductor Index fell 12%, and the S&P 500 dropped 4.4%. Goldman Sachs’ cash equities data show that U.S. cash equities saw net outflows of $3 billion that day, with the Information Technology and Communication Services sectors accounting for $2.2 billion. The cumulative net sell-off in tech stocks has now widened to $27 billion, surpassing the $21 billion of technical selling seen in July. Meanwhile, VVIX’s daily move was the largest since "Volmageddon" on February 5, 2018.

Rich Privorotsky, head of Goldman Sachs FICC and Equities, made it clear in the daily update referenced above that he tends to categorize this sell-off as akin to the 1998 liquidity crisis rather than a prelude to the bursting of the dot-com bubble in 2000. His core assessment is: The AI growth engine itself may still be intact; what’s being dramatically repriced is construction financing and the pace of deployment, not underlying demand. This judgment is crucial for investors—if the analogy holds, past episodes like this have ultimately turned into buy opportunities.

Self-reinforcing Deleveraging, Not Valuation Reset

The mechanism of this sell-off points to forced deleveraging, rather than a valuation reset driven by fundamentals.

Privorotsky pointed out that leveraged ETFs' forced liquidations toward the close amplified losses in semiconductors and the Nasdaq, and negative gamma exposures among market makers triggered a self-reinforcing liquidation cycle. From fund flows, Goldman Sachs' prime brokerage data show that the core pressure on semiconductors stems from the absence of long-only institutional (LO) foundational demand. Multi-strategy funds, quant funds, long-only institutions, and retail investors are all selling in the same direction—this "everyone on one side" scenario is itself the problem.

Correlation rising, momentum factors underperforming, residual leverage factors persistently selling off, while low-beta, defensive, and quality assets are sought after—this mix is highly consistent with classic forced risk-off behavior. In the two days following the election, first, AI stocks underperformed while cyclicals rebounded on political centralization expectations; then, there was a violent catch-down: if data center construction slows markedly, nominal growth expectations will be downgraded, placing renewed repricing pressure even on the rebound in cyclicals.

1998 or 2000: Two Templates, Two Outcomes

Privorotsky compared this drop to two historical episodes and ultimately leans toward the 1998 template.

Reasons supporting the 1998 analogy include: The Russian default and LTCM crisis were external shocks, but underlying U.S. economic growth remained robust; once liquidity returned and the Fed turned dovish, the market quickly recovered lost ground. In today’s context, the AI growth story may also hold; what’s being repriced is construction financing and the implementation timeline.

Additionally, today’s tech leaders generate real revenues, cash flow, and have strong balance sheets—a fundamental difference from companies during the 2000 bubble. Delaying data center construction does not extinguish demand for computational power; restricted capacity may even improve the economics of existing assets.

What most closely resembles 2000 is the sheer scale of spending and uncertainty about returns. The core question is: is the market unwinding a sound growth narrative or discovering a flaw in the economic logic itself? Privorotsky leans toward the former, but emphasizes that demand, utilization, and returns data must remain solid.

Four Concerns Over AI Infrastructure—Market Focusing on the Wrong Risk

Goldman Sachs outlined four market concerns driving the current sell-off: First, capital may be insufficient to support the scale of build-out; second, revenues may not justify investment; third, power and permit capacity may be insufficient to support compute expansion; and fourth, overall spending may be unsustainable.

Privorotsky believes the market’s current focus—power and permitting issues—is actually the least important among the four. A supply constraint does not mean demand vanishes. The real question is: can revenues ultimately justify the capital being deployed.

On political factors, Texas Governor-elect Gina Hinojosa’s remarks were milder than market interpretations. She made clear: "We can achieve economic development while protecting the people of Texas; these are not mutually exclusive," and reiterated, "I do not hate data centers." Privorotsky sees this as more of a regulatory reset than ideological rejection, and actual spending may turn out better than the market’s current discounted expectations.

Technical Levels and Preconditions for Bottom-Fishing

From a technical perspective, the Nasdaq has broken below its 200-day moving average, while the S&P 500 is just above it—whether this key support level holds is critical. The negative gamma environment means there is scope for sharp swings in either direction.

Privorotsky specifically named TSMC and ASML as quality preferences, suggesting an overweight toward high-quality tech assets amid current uncertainties.

He also listed three conditions required to alter the market narrative: AI investment forms a political positive feedback loop, financing markets stabilize, and empirical evidence confirms underlying spending and demand remain sound. The contrary is also clear: if AI capital expenditure is substantially delayed and revenue fails to validate the investment thesis, the analogy with 1998 fails and the 2000 template becomes relevant; if refinancing stress appears first in financial markets, the nature of the sell-off will fundamentally change.

On longer horizons, Privorotsky cited historical data that in the 12 months following the 19 midterm elections since 1950, the S&P 500 has always risen—but he stresses this is a 12-month perspective, not a one-month forecast, and short-term timing remains difficult. On energy, oil rebounded about 4% in a single day, and earlier market pricing of a rapid easing in Iran's situation was overly optimistic; negotiations are unresolved, and the complex impact of energy shocks on rates, growth expectations, and risk appetite remains an important variable in the stabilization process.