【TMGM Financial Recap】November 5 Tech Sell-Off Echoes a 1998-Style Liquidity Flush, Not a 2000-Style Bubble Burst, in Goldman Sachs’ Assessment

A 6.4% closing fall in the Nasdaq 100 on November 5 removed approximately $2.3 trillion in market value in a single day, its sharpest daily loss since March 2020. Elsewhere, the Philadelphia Semiconductor Index dropped 12.25%, while a 70% rise in the VIX carried it to its highest reading since early April. Nvidia lost 9.26%, leaving the shares 24.1% below their October high and officially in a technical bear market. Against these extreme figures, Goldman Sachs focused its review on a distinction central to investment decisions. Had liquidity pressures forced investors out of an otherwise sound growth story, or was the market starting to recognise that the underlying economic rationale was unsound? The bank favoured the liquidity explanation.

Goldman Sachs Sees a Closer Parallel With 1998 Than 2000

Of the two historical downturns considered in its analysis, Goldman Sachs judged 1998 to be the better guide to the current sell-off.

The bank's preference rests on three arguments. One concerns the source of the disruption: an external shock, rather than weakening fundamentals. Russia's default and the collapse of LTCM set off the turmoil in 1998, without undermining US economic growth. In the present case, the Democrats' capture of both chambers of Congress has created uncertainty around data centre approvals. That is a political disruption, not evidence that demand for AI has disappeared. Another distinction lies in corporate finances. Real revenues, cash flows and strong balance sheets support today's technology leaders, unlike the businesses without earnings backing in 2000. The third argument concerns capacity: postponements to data centre construction leave demand for computing power in place, while supply constraints could even make existing assets more economically attractive.

There are uncomfortable echoes of 2000, Goldman Sachs accepts, particularly the sheer size of expenditure and the uncertain returns. On balance, though, it sees investors unwinding positions in a sound growth story hampered by financing and building schedules. It is less persuaded that the selling reveals a defective economic premise.

For investment decisions, much turns on whether that historical parallel is valid: liquidity shocks of this kind have previously produced opportunities to buy. An important caveat appears elsewhere in Goldman Sachs' analysis, however. Federal Reserve interest rate cuts underpinned the rapid recovery in 1998, and equivalent support is missing today. No other difference between that precedent and the present market is as substantial.

The bank attributes part of the weakness in semiconductors and the Nasdaq to compulsory selling by leveraged ETFs before the close. It also argues that negative gamma exposure among dealers produced a liquidation loop that reinforced itself. Quantifying those flows, Nomura's Charlie McElligott estimated that pre-close rebalancing by leveraged ETFs produced approximately $47 billion of sell orders, with around $61 billion coming from volatility control funds. For each 1% fall in the market, dealers' negative gamma exposure contributed approximately $7 billion of further selling pressure. Combined, these mechanical flows were comfortably above $100 billion across the market.

Positioning helps explain why the liquidation became a “stampede” rather than remaining a “technical correction”. In Goldman Sachs' prime brokerage figures, there was no underlying demand from long-only institutions to absorb the sales. Multi-strategy funds, quantitative funds, long-only institutions and retail investors were instead all sellers. The problem was precisely “Everyone being on the same side”.

An account of trade execution from John Flood on Goldman Sachs' trading desk supported that interpretation. Activity on November 4, the day after the election, made it one of the bank's busiest sessions that year. Having lowered risk ahead of the vote, long-only investors were wrong-footed by Wednesday's advance and obliged to rebuild positions at rising prices. The following day, November 5, brought a reversal: alongside hedge funds, they sold the hyperscaler shares purchased just a day earlier. For the second session running, brokers serving a large share of retail clients described a marked imbalance in favour of sales. The bank itself reported “almost no real buying”. With correlations increasing, the momentum factor lagging, continued selling in the residual leverage factor and appetite for low-beta and defensive assets, the pattern closely resembled forced risk reduction. It was less consistent with valuations being reconsidered on fundamental grounds.

Another source of downward pressure was “Front-running”. Morgan Stanley put mechanical sell orders due before the close at approximately $52 billion in a midday estimate. The figure quickly circulated on Wall Street, prompting discretionary traders to sell in advance with the intention of buying back between 2 pm and 3 pm at the “usual low”. No such low emerged. By selling first, those traders helped push prices down further, and the session ended at the day's lowest level.

Political Developments Shift the Terms of the AI Trade

A notable feature of Goldman Sachs' political analysis is its refusal to equate the Democrats' election victory with ideological opposition to AI. In support, it singled out comments by Texas Governor-elect Gina Hinojosa: “We can achieve economic development while protecting Texans. The two are not mutually exclusive,” together with her unambiguous statement that “I do not hate data centres.” The bank interprets these remarks as pointing towards a regulatory reset, rather than ideological rejection. On that basis, actual expenditure could prove higher than the expectations reflected in current market discounts.

Politics has nonetheless changed the terms on which the market prices assets, Goldman Sachs acknowledges. As its trading desk explicitly observed, “long-term demand for AI computing power” is no longer the centre of the AI trade debate; attention has moved to “the speed and cost of deployment”. Sentiment has suffered from rising financing costs, political pressure on data centres and regulatory uncertainty after the election. Goldman Sachs warned during the sell-off that approximately 45% of planned US data centre capacity expansion could be subject to moratoriums in the wake of the Democrats' “blue tsunami”.

Under this interpretation, what investors discounted on November 5 was “a longer timeline for AI infrastructure construction”, not “the disappearance of AI demand”. Extending the wait for future cash flows is itself grounds for marking down assets valued on those flows. Even so, a delay is fundamentally different from a shock that takes demand to zero.

To organise the concerns driving the selling, Goldman Sachs set out four categories. Capital may be insufficient to finance construction at the envisaged scale; revenues may be insufficient to warrant the investment; power and permitting capacity may be insufficient to accommodate computing expansion; or the overall expenditure trajectory may be unsustainable.

Although the third concern, constraints on power and permitting, currently dominates market attention in the bank's view, it ranks that issue last in importance among the four. Its argument separates supply from demand: limiting the former does not establish that the latter has gone away. Ultimately, the decisive test is whether revenues can justify the capital committed.

The consequence for pricing is a return to “revenue validation” from “political noise”. Politics leaves delivery schedules uncertain, but the arrival or absence of revenues will ultimately establish whether 1998 remains an appropriate comparison.

Revenue and Funding Risks Could Undermine the 1998 Parallel

As well as explaining the sell-off, Goldman Sachs specified what would be needed to rebuild the market narrative and what would overturn its diagnosis. Restoring that narrative requires three conditions: AI investment must generate a positive political feedback loop, financing markets must stabilise, and tangible evidence must show that underlying expenditure and demand are still intact. Separately, it identified two clear developments that would invalidate its assessment.

A material delay in AI capital expenditure, accompanied by revenue results that refute the investment thesis, would make 2000 the more suitable precedent in place of 1998. Alternatively, should refinancing failures appear first in funding markets, the downturn would take on a fundamentally different character. Either development would undermine the 1998 parallel.

There were already tentative signs of buying on weakness during the sell-off, according to Goldman Sachs' trading desk, which confirmed interest in pullbacks in Nvidia, TSMC, Microsoft and CoreWeave. Its caution on timing remained explicit, however: “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 presents the November 5 slump as principally a liquidation event. Forced sales by leveraged ETFs, negative gamma feedback loops and missing demand from long-only institutions combined to drive the unwinding of positions; underlying AI demand did not suddenly vanish. The catalyst was political: uncertainty about data centre approvals after the Democrats swept Congress. In the bank's account, that shock lengthened the expected construction timetable without removing demand.

This interpretation can reconcile two facts that appear at odds. Losses of 12% in semiconductors and 6.4% in the Nasdaq 100 were severe enough to qualify as crisis-level 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”. Those features suggest liquidity mechanics, rather than a fundamental event, were at work.

The interpretation has a distinct weakness, nevertheless. Implicit in the 1998 analogy is the expectation that “a backstop will arrive”; Federal Reserve rate cuts provided it then. Goldman Sachs concedes that no such backstop is available today. Even a correct diagnosis would therefore not mean that recovery from this liquidity shock follows the pace or course seen in 1998. The bank's requirements for a recovery remain explicit: positive political feedback, more stable financing markets and concrete revenue evidence.