1998 VS 2000, Goldman Sachs executive: AI sell-off resembles LTCM crisis rather than the internet bubble, but key questions remain unanswered

The market is undergoing a fierce debate over historical precedent: Is this a liquidity shock similar to the 1998 Long-Term Capital Management (LTCM) crisis, or is it the prelude to the bursting of the Internet bubble in 2000? Rich Privorotsky, head of Delta One trading at Goldman Sachs, leans toward the former, but he also admits that a definitive answer has yet to emerge.

On Thursday, the Nasdaq plunged 5.5% in a single day, the Philadelphia Semiconductor Index plummeted 12%, and the S&P 500 dropped 4.4%, marking the largest single-day decline since April 2025. Goldman Sachs' Volatility Panic Index hit 9.3 out of 10 that day, and the VVIX soared nearly 45 points in one day—its second-largest single-day gain on record, second only to the "Volmageddon" event of 2018.

The trigger was the market’s repricing of data center construction prospects after the Democrats' overwhelming election victory, with the spark coming from potential regulatory restrictions on data center development in states like Texas.

The core question raised by this sell-off is now haunting every fund manager: Is the fundamental logic behind the AI investment boom still intact, or is the market exposing an economically unsustainable narrative?

Mechanically Programmed Selling Drives the Decline

Thursday’s plunge, from a market structure perspective, was a combination of mechanical events and fundamental shocks.

Privorotsky pointed out that leveraged ETFs in semiconductor sectors and the Nasdaq 100 triggered mechanical sell flows before close, compounded by negative market-maker Gamma positions, together forming a "self-reinforcing liquidation effect."

Charlie McElligott of Nomura estimates that leveraged ETF selling reached $47 billion that day, volatility-control funds supplied $61 billion, and market makers had about $7 billion in negative Gamma exposure for every 1% drop in the S&P 500.

Ariana Contessa, US equity sales trader at Goldman Sachs, added in the flow and positioning report that day that the VIX jumped 67% to 27.77 in one day, and several Goldman AI-related baskets posted their largest one-day declines since the launch of ChatGPT.

And the programmatic selling is not yet over—the futures close broke below the CTA mid-term threshold of 7,549 points. According to a Goldman scenario based on a 2.5 standard deviation move lower, the S&P 500 could face over $7.48 billion in systematic selling pressure in the week ahead; UBS warned that risk-control funds alone may have sold over $100 billion just that day.

The Real Issue: The Absence of Buyers

Programmatic selling created price pressure, but Privorotsky believes the real concern lies on the other side—the disappearance of buyers.

Goldman’s cash equities desk had a net sell of $3 billion that day, with tech stocks accounting for $2.2 billion, pushing cumulative net tech sales to $27 billion—exceeding the $21 billion technical adjustment seen in July 2025.

Contessa described, client inquiry intensity was “10 out of 10”, but actual trading activity was “only 5 out of 10”, and ETFs made up 32% of total volume for the day.

Notably, mega-cap tech stocks (Hyperscalers) that had stabilized the index on Wednesday showed “no corresponding buying activity at all.”

Privorotsky’s diagnosis is frank and pointed: “The real problem is the absence of underlying demand, especially from long-term funds... It feels like multi-strategy, quant, long-only, and retail are all moving in the same direction, and that is the core issue.”

1998 or 2000: Privorotsky’s Leaning and Reservations

Faced with the market’s most crucial qualitative debate, Privorotsky made his stance clear, but did not shy away from expressing his unease.

He prefers the 1998 LTCM crisis analogy: Russia’s default and the LTCM collapse were external shocks, but US economic fundamentals remained sound that year; once liquidity returned and the Fed cut rates, the market quickly regained lost ground.

He believes that today’s AI growth engine may be equally solid, and the market is simply repricing the pace and timetable of capital investment.

The core argument supporting this view is valuations and balance sheets. Privorotsky points out that, unlike in 2000, today’s tech giants have real revenues, cash flow, and robust balance sheets.

Peter Callahan, a Goldman TMT trader, notes that semiconductor index valuation multiples have compressed by about 50% from their June-July highs, Nvidia’s forward 24-month rolling P/E is about 10 times—the lowest since 2013. In contrast, Cisco’s P/E was over 100 times in 2000—this is the strongest argument for a 1998-style comparison.

However, Privorotsky also candidly lists what keeps him up at night:

"The unsettling similarity to 2000 is the sheer magnitude of capital expenditures and the extreme uncertainty of returns. Are we liquidating a healthy growth story, or discovering that the economic logic of the story itself is flawed? I lean towards the former, but demand, utilization, and returns must stand the test."

Four Major Concerns—the Market Is Focused On the Wrong Risk

Privorotsky breaks down worries behind the AI sell-off into four layers: first, insufficient financing capacity to support the scale of construction; second, inadequate revenue to justify capital investment; third, inadequate power and permit capacity to support compute expansion; fourth, an unsustainable overall capital expenditure trajectory.

In his view, the market is fixated on the third point—power and permits—“which is arguably the least important of the four.” Constrained supply does not mean demand has disappeared; the real issue is whether revenue can ultimately match the capital invested.

However, the first point—"financing"—is flashing a red warning. According to Goldman IR team statistics, mega-cap tech companies have already issued $296 billion in bonds this year, surpassing Goldman’s previous full-year forecast of $250 billion, and accounting for about 9% of total global investment-grade bond supply in all currencies.

Meanwhile, data center credit default swaps (CDS) breached 500 basis points on Thursday, a historic high, while CoreWeave bond spreads have topped 1,200 basis points.

What Privorotsky is highlighting is precisely a structural vicious cycle: permit delays for power (the third point) are directly driving up financing costs (the first point). Each quarter of permit delays means a half-built data center pays another quarter’s worth of interest.

Political Catalyst: Texas Regulatory Signal

The immediate catalyst for this market sell-off was a shift in regulatory expectations driven by election results. Goldman estimates that state-level moratoriums could freeze roughly 45% of planned US data center capacity, with Texas alone accounting for about a quarter of that.

Privorotsky thinks the market is taking Texas Governor-elect Gina Hinojosa’s statements too literally. He cites her remarks—“We can both promote economic growth and protect Texans' interests; the two are not mutually exclusive,” and “I don’t hate data centers”—to conclude:

"This sounds more like a regulatory reset than an ideological rejection. Actual spending might exceed what the current market repricing reflects."

The Goldman IR team adds that, according to CNBC, Hinojosa has stated that permitted data centers will be allowed to be completed, which "may protect much of the build scheduled for 2027 and even 2028." Goldman economists also expect the related moratoriums to be fairly brief, as data centers rank low on voters’ priority lists.

Still, public discontent should not be ignored. According to Jefferies citing Heatmap polling data, the proportion of Americans strongly opposed to data centers being built near their residences has soared from 24% in September 2025 to 61% in August 2026.

Georgia, Ohio, and Michigan are following Texas’s legislative direction. Texas’s legislative session ends May 31, and so-called "regulatory reset" is still a real reset.

The Flaw in the 1998 Playbook: No Fed Rate Cut Support This Time

Privorotsky wraps up with historical data as his confidence: since 1950, following 19 midterm elections, the S&P 500 gained over the ensuing 12 months in every case—"That's a 12-month observation, not a one-month forecast.” He sees events like this as historically buy opportunities, but timing in the short term is hard to grasp.

However, the 1998 analogy has a fatal internal contradiction, hidden in Privorotsky’s own description of 1998: “Once liquidity returned and the Fed cut rates.”

In 1998, then-Fed Chairman Greenspan cut rates three times in seven weeks. But in 2026, the market is still pricing a roughly 65% chance that the Fed hikes in December, while Cleveland Fed’s Hammack was still publicly expressing a preference for a restrictive stance on Thursday morning.

This time, there won’t be a Fed-convened LTCM rescue meeting, and the $296 billion in mega-cap tech bonds issued this year means this round of construction is more reliant on the bond market than ever before.

Liquidity is also not looking favorable. Equity long-short tech hedge funds are flat for the year, with strong motivation to lock in gains by year-end; programmatic selling is ongoing; and even Privorotsky admits the “real problem” is the absence of long-term fund buyers.

Overall, 1998 may be the right end point, but 2000 could be the necessary path along the way—at least until a material buying force finally enters. The only consolation is this comparison: Nvidia’s 10x forward P/E is still a far cry from Cisco’s more than 100x.