【TMGM Financial Recap】1998, Not 2000: Goldman Sachs Interprets the November 5 Tech Sell-Off as a Liquidity Flush Rather Than a Bubble Burst

The Nasdaq 100 lost 6.4% on November 5. That close erased approximately $2.3 trillion in market value in a single day, the index’s biggest daily fall since March 2020. The Philadelphia Semiconductor Index shed 12.25%, and a 70% surge in the VIX sent it to its highest level since early April. Nvidia’s 9.26% drop extended its retreat from the October high to 24.1%, officially placing it in a technical bear market. For Goldman Sachs, the investment question was not simply how far markets had fallen, but what had driven the selling. Was liquidity forcing the liquidation of positions in a fundamentally sound growth story, or were investors beginning to uncover flaws in its economic foundations? The bank’s review supported the first explanation.

1998, Not 2000: Goldman Sachs’ Preferred Comparison

Two historical episodes framed Goldman Sachs’ analysis of the latest decline. Its conclusion: 1998 provides the more relevant precedent.

The bank identifies three reasons for choosing 1998. The first is the external origin of the shock, rather than deterioration in fundamentals. US economic growth remained intact in 1998 despite Russia’s default and the collapse of LTCM. Uncertainty surrounding data centre approvals after the Democrats won both chambers of Congress is likewise a political shock, not a loss of AI demand. The second reason is financial strength. Today’s leading technology companies have actual revenues, cash flows and robust balance sheets, unlike the companies unsupported by earnings in 2000. Third, construction delays at data centres do not extinguish demand for computing power. Limits on capacity could even enhance the economics of existing assets.

Spending on an enormous scale and uncertain returns are the most concerning parallels with 2000, Goldman Sachs acknowledges. Even with those reservations, the bank favours a liquidation within a healthy growth story disrupted by financing schedules and construction timetables. It is less inclined to see the sell-off as evidence that the underlying economics are flawed.

Whether 1998 is the right comparison matters directly for investors: historically, similar liquidity shocks have led to buying opportunities. But the bank attaches a critical caveat elsewhere in its review. The market recovered quickly in 1998 because the Federal Reserve cut interest rates. Today, that support is absent, marking the largest divergence between the historical precedent and current conditions.

Forced pre-close liquidation by leveraged ETFs worsened the semiconductor and Nasdaq declines, Goldman Sachs believes. Negative gamma exposure among dealers then reinforced the liquidation cycle. Nomura’s Charlie McElligott put figures on these flows: leveraged ETF rebalancing generated approximately $47 billion in sell orders before the close, and volatility control funds contributed around $61 billion. Each 1% market decline brought approximately $7 billion of extra selling pressure from dealers’ negative gamma exposure. In aggregate, mechanical selling across the market far exceeded $100 billion.

The decisive shift from a “technical correction” to a “stampede”, however, came from positioning. Goldman Sachs’ prime brokerage data showed an absence of underlying demand from long-only institutions. At the same time, retail investors, long-only institutions, quantitative funds and multi-strategy funds were all selling. The difficulty was “Everyone being on the same side”.

Execution data told the same story, according to John Flood on Goldman Sachs’ trading desk. November 4, immediately after the election, was among the firm’s busiest trading sessions of the year. Long-only investors entered the election with reduced risk exposure. Wednesday’s rally took them by surprise, forcing them to rebuild positions by chasing higher prices. On November 5, they reversed course, selling the hyperscaler shares bought only a day earlier alongside hedge funds. Brokers whose client bases had a large retail component reported a substantial selling imbalance for the second consecutive session. Within Goldman Sachs, there was “almost no real buying”. Correlations climbed, the momentum factor underperformed and the residual leverage factor continued to face selling, while investors favoured low-beta and defensive assets. That combination matched the usual pattern of forced de-risking much more closely than a fundamental reassessment of valuations.

The slide also gathered force from “Front-running”. A midday estimate from Morgan Stanley indicated approximately $52 billion in mechanical sell orders before the close. News of that estimate travelled quickly through Wall Street. Discretionary traders sold ahead of those flows, intending to buy back between 2 pm and 3 pm at the “usual low”. The anticipated low never appeared. Their early selling instead added to the decline, and prices finished at the session’s lowest point.

45% Capacity Exposure: Assessing the Political Risk

The Democrats’ victory does not amount to an ideological rejection of AI in Goldman Sachs’ assessment of the political shock. The bank drew particular attention to Texas Governor-elect Gina Hinojosa’s remarks: “We can achieve economic development while protecting Texans. The two are not mutually exclusive,” and her direct statement that “I do not hate data centres.” Its reading favours a regulatory reset over ideological opposition. Actual spending could therefore come in above the expectations reflected in current market discounts.

Politics has nevertheless reset the parameters for market pricing, the bank accepts. Its trading desk explicitly described a change in the AI debate: the focus has moved away from “long-term demand for AI computing power” towards “the speed and cost of deployment”. Financing has become more expensive, data centres face political pressure, and post-election regulatory concerns have compounded the damage to sentiment. During the sell-off, Goldman Sachs warned that approximately 45% of planned US data centre capacity expansion could face moratoriums following the Democrats’ “blue tsunami”.

On that assessment, the November 5 repricing reflected “a longer timeline for AI infrastructure construction”, not “the disappearance of AI demand”. A longer wait warrants lower valuations for assets priced on future cash flows. It remains a fundamentally different shock, however, from demand collapsing to zero.

Four distinct concerns explain the selling in Goldman Sachs’ framework. There may be too little capital to fund construction at the required scale, or too little revenue to justify the investment. Power and permitting capacity may be inadequate for computing expansion. Alternatively, the overall spending trajectory may be unsustainable.

Power and permitting constraints, the third category, now command the most market attention in the bank’s assessment. Yet Goldman Sachs ranks them as the least important of the four concerns. Restrictions on supply do not establish that demand has disappeared. The ultimate test is whether revenues justify the invested capital.

For pricing, the implication is a renewed emphasis on “revenue validation” rather than “political noise”. Political shocks make delivery schedules less certain. Whether the revenues arrive will ultimately decide if the 1998 parallel remains valid.

1998 Comparison: Two Tests That Could Overturn It

Beyond its diagnosis of the selling, Goldman Sachs specifies three requirements for a different market narrative and two developments that would overturn its assessment. Rebuilding the narrative requires AI investment to generate a positive political feedback loop, financing markets to stabilise, and concrete evidence that underlying spending and demand remain intact.

The 1998 parallel would fail under two scenarios. Material delays to AI capital expenditure, combined with revenue outcomes that disprove the investment thesis, would make 2000 the more appropriate comparison. Refinancing failures appearing first in funding markets would instead alter the fundamental nature of the downturn.

Early dip-buying was already visible to Goldman Sachs’ trading desk during the sell-off. It confirmed interest in buying pullbacks in Nvidia, TSMC, Microsoft and CoreWeave. The desk nevertheless stressed that “it is difficult to know when selling pressure will ease, and the rebound when it does could be sharp”.

Rather than adopting the “AI bubble” explanation, Goldman Sachs frames the November 5 plunge as a position liquidation. Forced sales by leveraged ETFs, negative gamma feedback loops and missing demand from long-only institutions acted together as its principal drivers. A sudden loss of underlying AI demand was not the cause. The catalyst was political: the Democrats’ congressional sweep created uncertainty around data centre approvals. That shock lengthened construction timelines without eliminating demand.

Two apparently conflicting observations fit within this framework. Losses of 12% in semiconductors and 6.4% in the Nasdaq 100 were indeed crisis-scale moves. Yet Goldman Sachs’ cash equities desk recorded cumulative net technology selling of $27 billion, while positioning showed that “everyone was on the same side”. Together, those observations indicate the mechanics of a liquidity event rather than a fundamental one. Explaining both is the framework’s strength.

Its limitation is equally clear. The 1998 analogy implicitly relies on the assumption that “a backstop will arrive”. Federal Reserve rate cuts supplied that backstop in 1998; Goldman Sachs acknowledges that none exists today. Even a correct diagnosis would therefore not imply a recovery from the liquidity shock that follows the same pace or path as 1998. The bank’s stated requirements for recovery remain positive political feedback, stabilising financing markets and concrete revenue evidence.