#美股史诗级抛售 #Gate股票畅聊季 #每周来晒 The “Deleveraging” Moment in U.S. AI Stocks: A 1998 Replay Without Fed Rate Cuts to the Rescue!
This article analyzes the plunge in U.S. AI stocks and the wave of programmatic selling: at the market-structure level, leveraged ETFs and dealers’ negative Gamma triggered a mechanical selloff; at the valuation level, uncertainty over capex and returns created a rift; and at the macro level, the Fed was absent as a backstop through rate cuts. It examines whether the AI trade is replaying the 1998 LTCM crisis or heading toward a 2000-style bubble.
As buyers of AI computing power start doing the math, sellers need to rewrite their valuation logic; and when buyers disappear across the entire market, a stress test of the AI narrative’s survival officially begins. Thursday’s selloff gave U.S. stocks their worst single-day performance since April 2025: the Nasdaq fell more than 5%, the Philadelphia Semiconductor Index dropped by double digits, and the S&P 500 also plunged. Volatility markets were even more extreme—the Goldman panic index neared a perfect score, VVIX jumped nearly 50% in a single day, its largest point gain since the “Volmageddon” of 2018. The spark that set it all off was a sharp rise in expectations for regulations restricting data center construction in Texas and elsewhere after Democrats swept the elections.
After the plunge, Wall Street was engulfed in an intense debate over historical precedents: was this a liquidity shock like the 1998 Long-Term Capital Management (LTCM) crisis, or the prelude to the bursting of the dot-com bubble in 2000? Goldman’s head of Delta One trading, Rich Privorotsky, leans toward the former, but acknowledges that a definitive answer has yet to emerge.
I. Market Structure: The Absence of Buyers Is Scarier Than Programmatic Selling
From a trading perspective, Thursday’s plunge was a combination of a mechanical event and a fundamental shock. Leveraged ETFs triggered mechanical selling ahead of the close, and dealers’ negative Gamma positions compounded the pressure, creating a “self-reinforcing liquidation effect.” Nomura estimates that leveraged ETFs and volatility-control funds sold on the order of $100 billion that day. Meanwhile, Goldman’s trading desk was a net seller of $3 billion, with tech stocks accounting for the bulk of it. Even more unsettling for traders, the mega-cap tech stocks that had consistently acted as “buyers of last resort” saw no meaningful buying this time.
If it were merely programmatic selling, markets could usually rebound quickly once liquidity returned. But the mass exodus by multi-strategy funds, quant funds, long-only funds, and retail investors in the same direction means the market is experiencing a “collapse of consensus.” The real question is not “who is selling,” but “why is nobody willing to buy?” The absence of underlying demand has exposed the extreme fragility of the AI trade’s market microstructure.
II. Valuation: 1998-Like Fundamentals, but Watch Out for 2000-Style Capex
On the crucial question of how to characterize this selloff, Privorotsky offered data to support his view that the current situation looks more like 1998. Unlike in 2000, today’s tech giants have real revenue, cash flow, and strong balance sheets.
More importantly, valuations have fallen: the semiconductor index’s valuation multiple has contracted by about 50% from its high for the year, while NVIDIA’s forward price-to-earnings ratio over the next 24 months is around 10x, its lowest level since 2013.
By comparison, Cisco’s P/E ratio exceeded 100x in 2000.
However, Privorotsky also pointed to concerns: AI capex is currently enormous, while returns are highly uncertain. NVIDIA at 10x and Cisco at 100x are the strongest reasons to believe that “this time is different.”
This means the market is currently punishing the “valuation premium” and “leverage,” not “fundamentals.” But the sheer scale of capex means that if revenue fails to match these investments, today’s 10x P/E could become the starting point for the next leg down. Is the market liquidating a healthy growth story, or discovering that the story’s underlying economics are wrong? That will take time to determine.
III. The Macro Impasse: An “LTCM Moment” Without Fed Rate Cuts to the Rescue
Privorotsky divided the concerns behind the AI selloff into four layers: financing capacity, revenue alignment, power and permitting, and the sustainability of capex.
He believes the market is overly fixated on “power and permitting,” the least important of the four, while the real impasse lies in the second point—whether revenue can match investment.
At the same time, the first point, “financing,” is flashing red. Credit markets have sounded the alarm before equities: the data center CDS basket has breached 500 basis points, and CoreWeave’s bond spread has exceeded 1200 basis points. Meanwhile, according to Goldman’s IR team, tech giants have issued $296 billion in debt this year, far more than forecast at the start of the year. Worse, delays in power permits are directly driving up financing costs, creating a vicious cycle.
The 1998 playbook has one fatal flaw. That year, the Fed cut rates three times in seven weeks, swiftly resolving the crisis; in 2026, however, markets are still betting on rate hikes. There will be no Fed convening an “AI rescue meeting,” even as this wave of construction is more dependent on the bond market than ever before.
1998 may be the right end point, but before substantial buying emerges, the path to a 2000-style bubble burst may be an unavoidable bout of pain.
IV. The Regulatory Variable: Texas’s “Regulatory Reset” and Public Backlash
The immediate trigger for this selloff was a regulatory signal from Texas. Goldman estimates that state-level moratoriums could freeze about 45% of planned U.S. data center capacity, with Texas accounting for a quarter of it. But Privorotsky believes the market has taken the remarks of Texas’s governor-elect too literally: this looks more like a “regulatory reset” than ideological opposition. Data centers that have already obtained permits will likely be allowed to finish construction, protecting the buildout through 2027 and 2028. Still, the public backlash cannot be ignored: a Heatmap survey found that the share of respondents strongly opposed to data centers being built near their homes rose from about a quarter to more than 60% in one year.
Markets often overreact to political signals. But this “regulatory reset,” combined with intense public backlash, means data center construction costs—including time and compliance costs—will rise systematically. This also reinforces the logic outlined above: permitting delays directly raise financing costs, becoming a heavy burden on the balance sheets of AI infrastructure assets.
Conclusion: From “Valuation Reset” to “Liquidity Washout”
From the selloff triggered by Texas regulations, to buyers’ absence amid programmatic selling, to fading expectations of Fed rate cuts, the U.S. AI trade is enduring a painful upheaval, shifting from a “valuation reset” to a “liquidity washout.”
NVIDIA’s 10x forward P/E suggests this may not be a repeat of the 2000 bubble; but the steadily widening CDS spreads on data centers and the absence of long-term buyers warn us that the 1998 liquidity crisis may not yet have bottomed.
For investors, what matters is not the war of words among macroeconomists over historical comparisons, but when substantial buying will emerge and when AI revenue will deliver. Until those two questions are answered, any rebound could be no more than a mechanical recovery driven by short covering, rather than a reversal of the trend.
This article analyzes the plunge in U.S. AI stocks and the wave of programmatic selling: at the market-structure level, leveraged ETFs and dealers’ negative Gamma triggered a mechanical selloff; at the valuation level, uncertainty over capex and returns created a rift; and at the macro level, the Fed was absent as a backstop through rate cuts. It examines whether the AI trade is replaying the 1998 LTCM crisis or heading toward a 2000-style bubble.
As buyers of AI computing power start doing the math, sellers need to rewrite their valuation logic; and when buyers disappear across the entire market, a stress test of the AI narrative’s survival officially begins. Thursday’s selloff gave U.S. stocks their worst single-day performance since April 2025: the Nasdaq fell more than 5%, the Philadelphia Semiconductor Index dropped by double digits, and the S&P 500 also plunged. Volatility markets were even more extreme—the Goldman panic index neared a perfect score, VVIX jumped nearly 50% in a single day, its largest point gain since the “Volmageddon” of 2018. The spark that set it all off was a sharp rise in expectations for regulations restricting data center construction in Texas and elsewhere after Democrats swept the elections.
After the plunge, Wall Street was engulfed in an intense debate over historical precedents: was this a liquidity shock like the 1998 Long-Term Capital Management (LTCM) crisis, or the prelude to the bursting of the dot-com bubble in 2000? Goldman’s head of Delta One trading, Rich Privorotsky, leans toward the former, but acknowledges that a definitive answer has yet to emerge.
I. Market Structure: The Absence of Buyers Is Scarier Than Programmatic Selling
From a trading perspective, Thursday’s plunge was a combination of a mechanical event and a fundamental shock. Leveraged ETFs triggered mechanical selling ahead of the close, and dealers’ negative Gamma positions compounded the pressure, creating a “self-reinforcing liquidation effect.” Nomura estimates that leveraged ETFs and volatility-control funds sold on the order of $100 billion that day. Meanwhile, Goldman’s trading desk was a net seller of $3 billion, with tech stocks accounting for the bulk of it. Even more unsettling for traders, the mega-cap tech stocks that had consistently acted as “buyers of last resort” saw no meaningful buying this time.
If it were merely programmatic selling, markets could usually rebound quickly once liquidity returned. But the mass exodus by multi-strategy funds, quant funds, long-only funds, and retail investors in the same direction means the market is experiencing a “collapse of consensus.” The real question is not “who is selling,” but “why is nobody willing to buy?” The absence of underlying demand has exposed the extreme fragility of the AI trade’s market microstructure.
II. Valuation: 1998-Like Fundamentals, but Watch Out for 2000-Style Capex
On the crucial question of how to characterize this selloff, Privorotsky offered data to support his view that the current situation looks more like 1998. Unlike in 2000, today’s tech giants have real revenue, cash flow, and strong balance sheets.
More importantly, valuations have fallen: the semiconductor index’s valuation multiple has contracted by about 50% from its high for the year, while NVIDIA’s forward price-to-earnings ratio over the next 24 months is around 10x, its lowest level since 2013.
By comparison, Cisco’s P/E ratio exceeded 100x in 2000.
However, Privorotsky also pointed to concerns: AI capex is currently enormous, while returns are highly uncertain. NVIDIA at 10x and Cisco at 100x are the strongest reasons to believe that “this time is different.”
This means the market is currently punishing the “valuation premium” and “leverage,” not “fundamentals.” But the sheer scale of capex means that if revenue fails to match these investments, today’s 10x P/E could become the starting point for the next leg down. Is the market liquidating a healthy growth story, or discovering that the story’s underlying economics are wrong? That will take time to determine.
III. The Macro Impasse: An “LTCM Moment” Without Fed Rate Cuts to the Rescue
Privorotsky divided the concerns behind the AI selloff into four layers: financing capacity, revenue alignment, power and permitting, and the sustainability of capex.
He believes the market is overly fixated on “power and permitting,” the least important of the four, while the real impasse lies in the second point—whether revenue can match investment.
At the same time, the first point, “financing,” is flashing red. Credit markets have sounded the alarm before equities: the data center CDS basket has breached 500 basis points, and CoreWeave’s bond spread has exceeded 1200 basis points. Meanwhile, according to Goldman’s IR team, tech giants have issued $296 billion in debt this year, far more than forecast at the start of the year. Worse, delays in power permits are directly driving up financing costs, creating a vicious cycle.
The 1998 playbook has one fatal flaw. That year, the Fed cut rates three times in seven weeks, swiftly resolving the crisis; in 2026, however, markets are still betting on rate hikes. There will be no Fed convening an “AI rescue meeting,” even as this wave of construction is more dependent on the bond market than ever before.
1998 may be the right end point, but before substantial buying emerges, the path to a 2000-style bubble burst may be an unavoidable bout of pain.
IV. The Regulatory Variable: Texas’s “Regulatory Reset” and Public Backlash
The immediate trigger for this selloff was a regulatory signal from Texas. Goldman estimates that state-level moratoriums could freeze about 45% of planned U.S. data center capacity, with Texas accounting for a quarter of it. But Privorotsky believes the market has taken the remarks of Texas’s governor-elect too literally: this looks more like a “regulatory reset” than ideological opposition. Data centers that have already obtained permits will likely be allowed to finish construction, protecting the buildout through 2027 and 2028. Still, the public backlash cannot be ignored: a Heatmap survey found that the share of respondents strongly opposed to data centers being built near their homes rose from about a quarter to more than 60% in one year.
Markets often overreact to political signals. But this “regulatory reset,” combined with intense public backlash, means data center construction costs—including time and compliance costs—will rise systematically. This also reinforces the logic outlined above: permitting delays directly raise financing costs, becoming a heavy burden on the balance sheets of AI infrastructure assets.
Conclusion: From “Valuation Reset” to “Liquidity Washout”
From the selloff triggered by Texas regulations, to buyers’ absence amid programmatic selling, to fading expectations of Fed rate cuts, the U.S. AI trade is enduring a painful upheaval, shifting from a “valuation reset” to a “liquidity washout.”
NVIDIA’s 10x forward P/E suggests this may not be a repeat of the 2000 bubble; but the steadily widening CDS spreads on data centers and the absence of long-term buyers warn us that the 1998 liquidity crisis may not yet have bottomed.
For investors, what matters is not the war of words among macroeconomists over historical comparisons, but when substantial buying will emerge and when AI revenue will deliver. Until those two questions are answered, any rebound could be no more than a mechanical recovery driven by short covering, rather than a reversal of the trend.
