As of the U.S. stock market close on November 20 Eastern Time, what began as a tech stock pullback on Thursday, with all three major U.S. indexes falling more than 2%, escalated into a global "Black Friday" for equities dominated by a crash in AI computing infrastructure themes and a broad selloff across technology stocks.
Following four extremely weak trading sessions during the week, the S&P 500 plunged 3.70% on Friday to 6,910.39 points, marking its seventh consecutive losing session, and along with the Dow Jones Industrial Average, which fell 1.93%, both indexes have retreated more than 10% from their record closing highs, confirming entry into technical correction territory. The Nasdaq Composite dropped 5.28%, erasing its year-to-date gains, while the Nasdaq 100 fell 5.61%, sitting approximately 19.84% below its October 6 record closing high, having briefly exceeded a 20% decline intraday to enter so-called "technical bear market territory."
For the Nasdaq Composite Index, which covers a broader range of tech stocks, constituent companies collectively shed more than $4 trillion in market capitalization during the week ending November 20 Eastern Time. Two companies that had been expected to turn around AI trading sentiment were simultaneously hit by selling: Anthropic fell as low as $83.05 on its second day of trading, just about 5 cents from its offering price, before closing at $88.91, down 8.50%; NVIDIA plunged 14.01% after delivering earnings showing revenue had doubled.
The VIX fear index surged to 40.03 intraday on Friday, meaning the intraday panic gauge briefly hit its highest level since the April 2025 tariff turmoil, while Goldman Sachs' trading desk fear indicator also spiked dramatically to 9 out of a maximum of 10 on Friday. The core logic behind this global tech stock selloff, including "Black Friday," is that global investors have begun to distinguish between revenue already realized by chipmakers and investment returns still pending from large AI computing buyers such as AI neoclouds and cloud computing giants.
For data center operators, GPU delivery is only the beginning of capital investment; only when clusters are commissioned on schedule and paid utilization rates and rental income meet expectations will projects generate sufficient cash flow to cover electricity costs, service debt, and recoup equipment investments. The interest rate futures market still prices a 25 basis point Fed rate hike in December as nearly 100% certain, and expectations for a 50 basis point hike in December and continued hikes in early next year are steadily rising, further raising the return threshold for new projects and compressing the valuation premium investors are willing to pay for future profits.
Additionally, regarding global capital flow direction, as persistently crowded leveraged AI bulls sharply cut exposure and trend-following "fast money" quantitative forces, represented by CTA strategies, mechanically reduced positions on a massive scale as prices weakened, doubts about capital returns spread from model companies and chipmakers to cloud computing leasing and data center power equipment chains, rapidly amplifying the brutal selloff across the entire AI supply chain. With the market having become excessively fearful this week, especially with AI computing supply chain leaders and broad tech stocks falling too hard, many institutions are anticipating a sudden violent rebound.
Several top Wall Street strategists have explicitly expressed bullish views on a rebound in U.S. and global tech stocks during Thanksgiving week (November 23-27, i.e., next week). From Thursday's breach to Friday's stampede-like plunge, the "AI computing bellwether" Philadelphia Semiconductor Index dropped 17% in a week. In measuring the depth of this stock market correction, the Philadelphia Semiconductor Index, known in recent years as the "AI computing bellwether," is more representative than the Nasdaq.
The Philadelphia Semiconductor Index plunged 10.73% on Friday to 8,408.59 points, retreating 42.54% from its June 22 record closing high. Its weekly decline reached 17.13%, the worst week since September 2001, with constituent companies collectively losing approximately $2.3 trillion in market capitalization. U.S.-headquartered memory chip leader Micron Technology plummeted 13.89% in a single day to $594.51, having crashed 51.01% from its record closing high set in June, experiencing what amounts to a halving-level collapse.
During the same week, the S&P 500, Dow Jones, and Nasdaq Composite fell approximately 8.6%, 8.0%, and 10.0% respectively, with the Nasdaq posting its largest weekly decline since March 2020. U.S. stock trading volume on Friday reached 25.19 billion shares, about 29% above the average level of the past 20 trading days, and declining stocks on the Nasdaq outnumbered advancing ones by 4.27 to 1, with selling pressure spreading wildly from a handful of AI mega-caps to the entire stock market, panic sentiment spreading to every corner.
Asian stock markets had already experienced a round of repricing before New York officially opened for trading on Friday. Taiwan's weighted index, which has feasted on the AI infrastructure boom in recent years, fell 6.2% that day and 17.5% for the week, while South Korea's composite index and Japan's Nikkei 225 fell approximately 5.2% and 5.1% respectively. Subsequently, Europe's Stoxx 600 fell 2%, and the MSCI global stock index dropped 8.1% for the week, the largest weekly decline since early 2020.
Taiwan's wafer foundry and server manufacturing chain, South Korea's memory chips, and Japan's vast semiconductor equipment chain all have different revenue recognition points, yet all are exposed to the same variable: whether North American cloud computing companies' next batch of ultra-large-scale AI data center construction projects can secure substantial financing on schedule, get powered up, and continue aggressively purchasing core AI infrastructure equipment after Democrats control both chambers of Congress.
As noted above, the core of this wild market selloff and the brutal "Black Friday" crash is that investors have begun to distinguish between revenue data already realized by chipmakers and investment returns still pending from AI computing buyers. For data center operators, GPU delivery is only the beginning of capital investment; only when clusters are commissioned on schedule and downstream model companies' AI application-related revenue data, paid utilization rates, and rental data meet expectations will large AI infrastructure projects have sufficient cash flow to cover electricity costs, service debt, and recoup equipment investments.
The market's doubts about AI capital expenditure returns center on whether enterprise paid demand for AI applications can continue to grow at a significantly better-than-expected pace. For model and application companies, if subscription and API revenue growth can consistently outpace the computing costs required to provide services, there will be a more solid commercial foundation for renewing leases and expanding GPU purchases. For data center operators, only when clusters are commissioned on schedule, customers pay rent as agreed, and leases are renewed at sufficient rental rates upon existing contract expiration or taken over by new customers can projects be expected to cover electricity and operational costs, service debt on schedule, and recoup investments within the equipment's economic lifespan.
Why is the entire AI infrastructure supply chain, from GPUs to backup power, crashing even as NVIDIA's revenue doubled? NVIDIA's third quarter revenue reached $115.8 billion, up 103% year-over-year, exceeding Wall Street expectations of approximately $109 billion, with data center revenue up 110% year-over-year to $107.6 billion, and the company significantly raised its growth guidance for the next quarter and next fiscal year beyond expectations.
The earnings confirmed how sought-after already-delivered computing power is, yet the stock is reassessing how high a return customers can generate to continue expanding. On Friday, AMD fell about 13.3%, Broadcom dropped 12.2%, cloud computing operators CoreWeave and Nebius fell approximately 12% and 16% respectively, and even Caterpillar, which sells power generation and backup power equipment, dropped about 5%.
For chipmakers, the risk is customers delaying their next round of purchases; for operators who have already borrowed to build data centers, the same delay means rental income is pushed back while interest and equipment costs have already begun accumulating. NVIDIA's stock price also faces pressure independent of the industry cycle. Following Jensen Huang's departure, management succession is still underway. On Friday, Reuters cited two people familiar with the matter saying that U.S. SEC enforcement personnel are reviewing potential accounting issues at NVIDIA; before the report was published, the stock had already fallen more than 6%, and the decline widened after the news emerged.
The specific matters under review have not been made public, and the regional revenue attribution and customer financing arrangements that investors have been continuously monitoring cannot currently be understood as major issues already identified by U.S. regulators. Horizon Investment Services CEO Chuck Carlson had originally hoped that NVIDIA's earnings and Anthropic's listing would allow the market to digest pent-up doubts, but Friday's close showed that neither positive catalyst resolved investors' concerns about the next round of capital returns.
Cetera Chief Investment Officer Gene Goldman believes the two test different questions: the top chip company's earnings have already proven strong current profitability, while Anthropic's listing tests how high a valuation investors are willing to pay for future growth. When the 10-year Treasury yield remains near 5% and the market expects the Fed to continue hiking rates, even if AI demand remains strong, investors will ask how much of future years' growth has already been priced into stocks.
Ameriprise Financial Chief Market Strategist Anthony Saglimbene noted in a Friday closing report that investors are reassessing the financing capacity of computing power buyers: borrowing costs are elevated, whether credit funding can continue to flow remains uncertain, and NVIDIA's management transition adds another variable, namely how much financing support the company will be willing to provide to small and medium-sized customers going forward. Treasury Partners Chief Investment Officer Richard Saperstein pointed out that the same round of data center investment slowdown has different earnings impacts on different parts of the supply chain.
Tighter financing for new projects will lower tech companies' earnings growth expectations, but hyperscale cloud service providers that already own substantial operating data centers can still serve customers using existing facilities; if new capital expenditures slow, their free cash flow may actually improve. U.S. Treasury bonds rose and the Fed remains hawkish; who determines the next "selling storm"?
Bill Northey, Senior Investment Director at U.S. Bank Wealth Management, said selling pressure was still accelerating near Friday's close, with short-term prices more driven by sentiment, and that economic resilience and corporate earnings support for stock prices will take longer to manifest. Goldman Sachs' trading desk positioning model further explains why selling pressure may persist even if indexes stabilize: in a scenario where stocks trade sideways over the coming week, trend-following strategies such as commodity trading advisors (CTAs) could still net sell approximately $65.8 billion in stocks, and if the decline continues, potential selling could reach $80.6 billion.
This means that even if corporate earnings expectations do not deteriorate in tandem, systematic deleveraging triggered by earlier declines could still amplify short-term price volatility. On Friday, safe-haven funds bought U.S. Treasuries, with the 10-year yield falling about 10.4 basis points to 4.948%, while gold and silver fell 2.81% and 4.22% respectively, and Bitcoin briefly dropped below $60,000. Although Brent crude fell about 4% for the week, it was still insufficient to change near-term rate pricing.
Fed Governor Cook believes previous rate hikes were insufficient, while Philadelphia Fed President Paulson advocates cautious progress while remaining open to two consecutive meetings of rate hikes. Fed funds futures still price a 25 basis point December hike as nearly certain. The decline in long-term bond yields reflects investors buying safe-haven assets, while the actual financing cost of AI projects also depends on credit spreads. The U.S. October PCE data to be released at 21:30 Beijing time on November 25 will test whether the market has reason to lower rate hike expectations after the selloff.
Goldman Sachs' calculations explain why selling could continue even if the market stabilizes. The firm estimates that over the past month, trend-following strategies have sold approximately $57 billion in stock positions, with U.S. equity positioning turning to a net short of $19 billion. In the coming week, if stocks rise, the model still expects trend strategies such as CTAs to potentially sell $28.6 billion in stock positions; the sideways scenario shows $65.8 billion, and under a continued decline scenario, buying expectations are approximately $80.6 billion.
Goldman Sachs research team said the core reason is that a downtrend accumulated over multiple weeks will not immediately reverse on a single day's rebound. Investors should note that $80.6 billion is a quantitative model estimate under a declining scenario, and public disclosures also do not limit it entirely to U.S. stock selling. Five trading observations to distinguish an oversold rebound from a true stabilization of AI valuations.
On Friday, Anthropic traded just 5 cents from its $83 offering price intraday, and the Philadelphia Semiconductor Index plunged 10.73%. After such a sharp drop, short covering or trend funds slowing their position reduction could both bring about a rebound.
The following five trading observations are meant to answer whether, if an oversold rebound in the AI supply chain and broad tech stocks occurs, most institutional investors are beginning to expect global tech stocks to stage a technical oversold rebound next week after the massive plunge, and whether there is sufficient buying power and earnings foundation to sustain it. Charles Schwab Financial Research Center's Director of Derivatives Research and Strategy Nathan Peterson explicitly turned bullish on U.S. stocks next week in a trading outlook published Friday.
He noted that at the time of writing, the relative strength index for both the S&P 500 and Nasdaq Composite had dropped to 26, while the VIX had risen above 30, indicating short-term selling was approaching extremes and the probability of a technical rebound was higher than usual. He is particularly focused on the follow-through strength after Monday's open: if indexes open lower but close higher, or if Monday continues under pressure but Tuesday quickly recovers losses, that would indicate more than a simple gap-up that buyers are willing to catch the selling.
TheStreet Pro technical analyst Helene Meisler also pointed to next week as a rebound window. She said that even if the Nasdaq continues to close slightly lower, the momentum indicators she tracks could turn upward by midweek next week. In her experience, when downward momentum decays and investor sentiment has clearly turned pessimistic, a tradeable rebound is more likely to occur. The firm's portfolio manager Chris Versace noted after Friday's close that next Wednesday's U.S. Personal Consumption Expenditures inflation data could serve as a catalyst: if inflation comes in weaker than expected, market concerns about further Fed rate hikes may ease, providing rebound conditions for tech stocks that have already entered oversold territory.
In the short term, as underwriters complete overallotment covering, who is willing to commit real money near the offering price to absorb shares, and whether systematic selling pressure can ease as volatility subsides, will determine whether the market can first stabilize prices. Further out, whether the NVIDIA CEO controversy gradually subsides and the new management lineup and product line outlook gain market recognition, how much revenue increased enterprise customer model usage can generate, whether gross profit per unit of computing power can improve, and whether computing operators can finance the next batch of GPUs at affordable interest rates will determine whether investors have reason to re-rate the AI supply chain upward.
First, Anthropic's $83 reveals price disagreement between two types of buyers. Investors who received allocations at the offering price were still up 7.12% as of Friday's close; investors who bought at the $111.50 opening price on the first day and held on have lost 20.26% over two trading days. What Forte Securities calls a "buyers' strike" refers to some institutions that did not receive allocations and have no index-tracking obligations being willing to wait for lower quotes. Vega IS even explicitly stated it would only consider buying near or below the offering price. Demand remains, but the premium buyers are willing to pay has narrowed significantly.
Second, the record-breaking fundraising amount did not leave an equivalent-scale support for the secondary market. After underwriters fully exercised the 147 million share greenshoe, gross proceeds from the IPO and concurrent private placement totaled $119.5 billion, including proceeds from old share sales, with the company itself receiving approximately $117.6 billion in gross proceeds. Underwriters acquired additional shares at the offering price, already covering the overallotment position. If the stock price approaches $83 again, the buying that might previously have come from covering this position through the open market no longer exists, and whether the offering price holds depends more on genuine quotes from new investors.
Third, model usage growth can only support the next round of chip orders if it translates into sufficient per-unit computing revenue. Anthropic disclosed that as of the end of October, its annualized revenue run rate had reached $81 billion, but the company typically purchases computing power 12 to 18 months in advance, making short-term costs relatively fixed. If the additional usage volume brought by model price cuts is sufficient to improve utilization, more paid tasks can spread idle costs; if each task consumes more inference resources, or if per-call pricing declines faster than per-unit inference costs, revenue expansion may come with margin pressure. This is precisely why the market has been questioning from Claude's enterprise demand all the way to GPU and high-bandwidth memory orders.
Fourth, the credit market has already begun pricing different AI projects differently. A CoreWeave bond issued in June at a 9.625% yield has seen its secondary market yield approach 15% by mid-November, directly affecting the pricing reference for new financing and refinancing. Meanwhile, Apollo and Goldman Sachs on Friday still finalized approximately $44 billion in GPU financing for SpaceX, with one tranche of senior secured notes maturing in 2033 carrying a yield of 8.259%, with related leases expected to commence upon equipment delivery.
The collateral and tenor of the two financings differ, but the transaction results reveal what the capital is still willing to support: when equipment will be delivered, who bears the rent, and what recourse creditors have to assets and cash flow all directly affect financing terms. Fifth, the VIX touching 40 can explain why a rebound might be violent, but cannot alone determine a bottom. The VIX ultimately closed at 35.89 on Friday, up 26.20%, the highest close since the April 2025 tariff turmoil.
CBOE's Mandy Xu estimates that more than 80% of this increase can be explained by the S&P decline and its knock-on effects on options pricing. Under high volatility, even a slight weakening of selling could bring about a sharp bounce. What better tests the quality of a rebound is whether the number of stocks hitting 52-week lows significantly declines from Friday's elevated level when indexes recover, whether computing operators' credit spreads narrow in tandem, and whether new data centers can be delivered on schedule and begin collecting rent. The three respectively measure breadth of buying, financing capacity, and cash flow realization.
Comments