The $2.3 Trillion Panic
The Best Earnings Week of the Year Ended in the Worst Day Since 2020
Apple set a record. Palantir grew 96%. AMD guided to its strongest year ever. Two days later the Nasdaq had its steepest fall since the pandemic crash. Nothing in the earnings explains it.
November 2–6, 2026 | Ieva Bružaitė, Orion Capital
What happened
On Thursday 5 November roughly $2.3 trillion of US equity value disappeared in a single session. The Nasdaq Composite fell 5.52%, its worst day since March 2020. The Nasdaq 100 fell 6.36%. The S&P 500 fell 4.36%, its worst since April 2025. The Philadelphia Semiconductor Index fell about 12%. This was a panic, and it is worth being precise about what caused it. [1][2][3]
It was not the earnings. Three of the most closely watched AI-adjacent companies reported in the first two days of the week and all three were excellent. Apple posted a record September quarter. Palantir grew revenue 96% and raised guidance. AMD reported a record quarter and told investors to expect revenue growth above 80% in 2027. [4][5][6]
AMD fell 17.26% on Thursday. It had given the strongest guidance in its history on Tuesday. [1][7]
There was panic in the positioning.
There was none in the numbers.
Two things changed during the week that had nothing to do with company results. Democrats won control of both chambers of Congress on Tuesday, putting data-centre permitting and electricity costs into political play. And oil reversed violently, heading for a weekly loss of roughly 12% as reports emerged that crude flows through the Strait of Hormuz had recovered. Friday then delivered a labour market far stronger than anyone forecast, which pushed December rate-rise odds back up. [2][8][9][10]
Friday brought a partial recovery that did not rescue the week. The Nasdaq closed up 1.09% at 24,579, the S&P 500 up 0.32% and the Russell 2000 up 0.89%, while the Dow slipped 0.46%. The semiconductor index regained about four of the twelve points it had lost. But the Nasdaq still finished the week down 3.37%, its fourth consecutive weekly decline, and the S&P 500 fell 2.68% in its worst week since March 2025. A panic that takes one day to arrive and more than one week to clear is not yet finished. [11][13]
The week in numbers
| Index | Thu 5 Nov | Fri 6 Nov | Week | Friday close |
| Nasdaq Composite | −5.52% | +1.09% | −3.37% | 24,579 |
| S&P 500 | −4.36% | +0.32% | −2.68% | 7,376 |
| Dow Jones | −3.44% | −0.46% | −0.94% | 51,286 |
| Russell 2000 | — | +0.89% | — | 2,754 |
| PHLX Semiconductor | −12.25% | ~+4% | — | — |
| Elsewhere | Level | Note |
| VIX | 28.18 | +70% on Thursday, its highest since March; eased about 13% on Friday |
| 10-year Treasury | 4.86% | Up slightly Friday on the jobs report, but down about 7 bp on the week |
| 2-year Treasury | 4.52% | +4.5 bp Friday; the short end did the work |
| Brent crude | $98.32 | Down about 13% on the week from above $117 in late October |
| WTI crude | $90.83 | Energy was the worst S&P sector of the week, down about 12% |
| Gold | $4,074.90 | Spot gold up about 5.2% on the week — a genuine haven bid |
| Dollar index | 102.32 | +0.75% Friday after the employment report |
| December hike odds | 78% | Up from 70% before the jobs report |
Closing figures for Friday 6 November; weekly changes are for the full week. [1][2][3][10][13]
The earnings were the best part of the week
| Company | What it reported | What happened next |
| Apple (FQ4) | Revenue $114.5bn, +12%. Diluted EPS $2.10, +14%. A September-quarter record, with iPhone at a quarterly record and Services at an all-time high. First results under new chief executive John Ternus. | A strong quarter from the one mega-cap that is not spending heavily on AI infrastructure. |
| Palantir (Q3) | Revenue $2.31bn, +96%. US commercial revenue grew 148% and exceeded US government revenue for the first time. Adjusted EPS $0.44 against $0.39 expected. Full-year guidance raised. | The clearest evidence yet that someone is paying for AI software rather than only for the hardware beneath it. |
| AMD (Q3) | Revenue $13.4bn, +45%. Data Center $8.5bn, +97%. Q4 guided to roughly $17bn. Management guided 2027 revenue growth above 80%. Free cash flow $554m against roughly $1.97bn expected. | −17.26% on Thursday. |
Hold those three results next to Thursday’s tape and the ordinary explanations stop working. A market that has lost confidence in artificial intelligence does not sell a company 17% lower two days after it guides to 80% growth. Something else was being repriced.
Anatomy of a panic
The scale first. The Magnificent Seven lost more than $1 trillion of market value between them in the session, Nvidia alone shedding over $440 billion. Tesla fell 10.8%, SanDisk more than 22%, and Qualcomm had its steepest one-day fall since 2020 after a weak smartphone-chip forecast. By Friday morning the shock had travelled: South Korea’s Kospi fell more than 8% and SK Hynix with it. [15]
The most useful account of Thursday is mechanical rather than fundamental, and the evidence for it sits in positioning data rather than in any earnings release.
- Crowding. Global equity funds took in a reported $265.6bn in the four weeks to 4 November. Private-client cash balances at Bank of America sat at a record low of 9.3%. In the 13F data, the more hedge funds that owned a stock, the harder it fell on Thursday. Stocks held by five or more of nine tracked growth funds fell 7.3%, against 4.4% for the S&P 500 and 2.9% for the equal-weighted index. [2]
- No hedges. Leveraged funds had cut their net short position in Nasdaq 100 futures from roughly 101,000 contracts in August to about 5,600 by 27 October. The market walked into the fall with its protection sold. [2]
- Forced selling. Ten-day realised volatility on the S&P 500 jumped from 15.8% to 26.4%, which mechanically obliges volatility-targeting funds to reduce exposure regardless of any view. Street estimates of that potential selling ran as high as $135bn. [2]
- Dealer positioning. Options dealers had been short gamma since mid-October, meaning their own hedging amplifies moves rather than damping them. The S&P 500 was well below its gamma flip level when the selling began. [2]
- The signature of panic itself. Cboe recorded 11.23 million put contracts on 5 November, an all-time high, and the VVIX rose 49.3%. A 3x leveraged semiconductor ETF fell 38.7%, its worst day since 2010. [2]
None of that is an argument about whether AI will earn its capital. It is an argument about what happens when a great many people own the same thing, with the same thesis, having sold their insurance. The fundamental questions get answered eventually. Positioning gets answered in an afternoon.
A crowded trade does not need bad news to fall.
It only needs the absence of new good news.
The variable that was not in any model
Democrats won control of both the House and the Senate on Tuesday 3 November, a result called by CBS News, NBC News, PBS and CNBC among others. Markets initially rose on the outcome; by Thursday the reading had changed. [8]
The link to the AI trade runs through electricity and permitting, and two specific items did the damage. Texas governor-elect Gina Hinojosa pledged to halt new data-centre approvals — Texas being the single largest destination for US data-centre construction. And an NBC exit poll found that 76% of voters oppose new data-centre construction in their own communities. That second number is the one I would write down. It means the political cost of approving a data centre has now flipped sign in most districts, regardless of which party holds them. OpenAI responded during the week with a “community compact” promising public reporting of water and electricity use, which tells you how seriously the industry is taking it. [15]
One estimate circulating during the week put roughly 52 gigawatts of planned 2028 data-centre capacity at risk of pauses or new conditions, while 2027 capacity is largely beyond the reach of new rules. I would treat that specific figure as an estimate rather than a fact, but the direction is not in doubt. [2]
This is a genuinely new risk and it is not priced in any obvious way. For two years the constraint on AI infrastructure was supply of chips. Then it became supply of power. It may now become permission to use the power. A permitting delay does not reduce demand for compute; it changes when revenue is recognised, and for companies valued on 2028 cash flows, timing is most of the valuation.
The hardest evidence of the week arrived on Friday
While the market was busy recovering, Bloomberg reported that Microsoft has decided not to extend its 2023 agreement to rent AI cloud capacity from Oracle, and has been switching that capacity off over recent months. Separately, Microsoft is reviewing more than $80 billion of contracts with external cloud providers including CoreWeave, Nscale and Nebius, with executives reportedly frustrated by rising costs and project delays. Talks to lease additional Oracle capacity had already collapsed in June. Oracle fell about 3% on the news and has more than halved from its June peak. Oracle’s response was that demand for its AI infrastructure remains strong. [14]
This matters more than Thursday’s 5.52% did, and it received a fraction of the attention.
Thursday was investors changing their minds about AI spending.
Friday was a buyer actually cancelling some.
Every argument in this briefing and the last one has circled the same question: when does AI capital expenditure convert into cash, and what happens if the answer is later than expected. A hyperscaler declining to renew capacity and reviewing $80 billion more is the first hard datapoint on the other side of that question. It is one contract and one company, and Microsoft is building its own capacity rather than retreating from AI. But it establishes that these commitments are revisable, which is not how the market has been treating them.
Oil ran the same film backwards
Last week’s briefing described a war premium driving Brent above $117 and freight rates to records. This week the premium came out almost as fast as it went in. Brent fell toward $100 and crude is heading for a weekly loss of roughly 12%, with reports that outflows through the Strait of Hormuz have recovered towards and possibly above pre-war levels, alongside signals that Washington and Tehran have room for diplomacy. Fresh tanker attacks during the week did not reverse the move. [9]
Two observations. First, this is a reminder of how quickly a geopolitical premium can unwind, and it should make anyone cautious about building a long-term energy thesis on a conflict. Second, it matters for rates: falling oil removes one of the arguments for the Federal Reserve to tighten further, which is precisely why Friday’s employment report carried as much weight as it did.
Friday: a labour market that refuses to cooperate
October payrolls rose 165,000 against a consensus near 65,000. The unemployment rate fell to 4.1% from 4.2%. August was revised up by 53,000 to 186,000 and September by 31,000 to 60,000, adding 84,000 jobs to the record. Average hourly earnings rose 0.4% on the month and 3.1% over the year. Gains were led by professional and business services, health care, transportation and warehousing, and manufacturing. [10][12]
Treasury yields rose on the release, with the two-year up six basis points to 4.52% and the ten-year up three to 4.87%. Market-implied odds of a December rate rise moved from 70% to 78%. The dollar gained 0.7%. [10]
Two details deserve more attention than the headline. Involuntary part-time employment rose by 339,000 to 4.8 million, which is not what a tightening labour market usually produces. And with wage growth at 3.1% against headline inflation above it, real hourly earnings fell. A strong payroll number and a squeezed household are not contradictory; they are what late-cycle inflation looks like. [10][12]
The tightening is not only American. The Bank of England raised its policy rate by a quarter point to 4% during the week, voting 8–1, its first increase in more than three years, citing higher energy prices and inflation at 3.5%. Two major central banks moving the same direction at the same time is a materially different environment for long-duration assets than the one most portfolios were built for. [15]
Orion Capital lens
Last week this briefing argued that the market had stopped accepting growth as sufficient evidence and started asking when AI spending converts to cash. This week extends that, and adds a second question that did not exist a fortnight ago.
- Positioning is now a first-order risk, not a footnote. When ownership is this concentrated, the quality of a company’s quarter does not protect it over a horizon of days. AMD is the proof. That cuts both ways, which is why forced selling tends to create opportunity rather than destroy it.
- Permission has joined power and silicon as a constraint. Watch state-level permitting decisions, utility rate cases and any federal legislation on data-centre electricity pricing. These will move revenue-recognition timing across the entire infrastructure chain.
- Palantir is the most interesting data point of the week and the least discussed. US commercial revenue overtaking US government revenue, growing 148%, is evidence of customers paying for AI outcomes rather than AI capacity. That is the monetisation proof the hyperscaler cash-flow debate has been missing.
- Apple is quietly the control case for the whole argument. It grew 12% and set records without a hyperscale capital-expenditure programme. If AI infrastructure spending is the source of the market’s anxiety, the company spending least on it is worth watching as a relative-performance signal.
The week ahead | 9–13 November
- Whether Thursday’s low holds. A retest that fails to make a new low would suggest the panic was positional and has cleared. A lower low would suggest something more durable.
- The October CPI release, which lands into a market already pricing a 78% chance of a December rate rise. Given Friday’s wage data, this is the most consequential number of the month.
- Federal Reserve speakers. After two hawkish dissents in October and a strong payroll print, any shift in tone will move the curve immediately.
- First legislative signals from the new Congress on data centres, energy pricing and permitting.
- Nvidia reports later in the month and is now carrying the weight of the entire thesis. Positioning into that print is worth watching more closely than usual.
Verify all dates and earnings schedules against issuer and official sources before acting; calendars move.
Three scenarios
The panic clears. Friday’s bounce extends, volatility normalises, forced sellers finish, and the market re-rates on the actual earnings, which were strong. Palantir and Apple are the evidence. This requires CPI to cooperate, and it requires the Oracle contract to stay an isolated case.
The panic was early. Permitting restrictions move from campaign pledges into actual state decisions, CPI runs hot, December brings a rate rise, and more hyperscaler capacity commitments get quietly revised. In this path AMD’s 2027 guidance becomes the number to doubt rather than the number to own, and Friday’s rebound reads as the first bounce rather than the bottom. Four consecutive weekly declines argue this deserves more weight than it would have a month ago.
Narrowing, not falling. The index stabilises while leadership changes hands — from companies building AI capacity to companies selling AI outcomes and to businesses generating cash now. Palantir up, pure capacity plays flat, Apple and defensives steady. This is the path I would still weight most heavily over the next quarter.
Bottom line
The best week of AI earnings this year produced the worst day for technology stocks since the pandemic. That is not a paradox once you accept that price and fundamentals answer to different clocks. Earnings decide where a stock belongs over years. Positioning decides where it trades on a Thursday.
I do not think Thursday told us anything new about whether AI will earn its capital. It told us that a great many investors had the same view, expressed in the same stocks, with their hedges sold, and that this is a fragile arrangement regardless of whether the view is correct. Friday’s bounce supports that reading, and the fourth consecutive weekly decline qualifies it.
Two pieces of genuinely new information came out of the week, and neither was a price. Voters have turned against data centres in their own neighbourhoods by a margin of roughly three to one, which changes the politics of building them. And Microsoft let a capacity contract lapse while reviewing $80 billion more, which changes the assumption that these commitments are fixed. Those two facts will outlast the panic that happened to accompany them.
The panic was real. The reason given for it was not.
That gap is usually where the opportunity is — and occasionally where the warning is.
Ad Astra.
— Ieva Bružaitė, Orion Capital
Disclaimer
This publication is for informational and research purposes only. It does not constitute investment advice, an offer, a solicitation or a recommendation to buy or sell any security, and it takes no account of any individual’s circumstances, objectives or risk tolerance. Company names appear as research subjects, not recommendations. Market conditions, company fundamentals and valuations change rapidly. Figures are drawn from the sources listed and should be verified against issuer filings and official statistical releases before being relied upon. Readers should conduct their own research and, where appropriate, consult a licensed adviser.
Sources and verification notes
[1] The Motley Fool, “Stock Market Today, Nov. 5: Nasdaq Suffers Worst Day in 18 Months as AI Trade Comes Under Fire.”
https://www.fool.com/coverage/stock-market-today/2026/11/05/stock-market-today-nov-5-nasdaq-suffers-worst-day-in-18-months-as-ai-trade-comes-under-fire/
[2] FinzData, “Why Did the Stock Market Crash on Nov 5, 2026?”
https://finzdata.com/research/ai-unwind-selloff-november-2026
[3] Seeking Alpha, “Nasdaq ends -5.5%, S&P 500 sees worst day since April 2025,” 5 November 2026.
https://seekingalpha.com/news/4667196-nasdaq-ends–55-sp-500-sees-worst-day-since-april-2025
[4] Apple, “Apple reports fourth quarter results,” 2 November 2026.
https://www.apple.com/newsroom/2026/11/apple-reports-fourth-quarter-results/
[5] StockTitan, “Palantir Earnings Show Revenue Up 96% and a Higher Revenue Forecast for 2026.”
https://www.stocktitan.net/articles/palantir-q3-2026-earnings-us-commercial-revenue-passes-government
[6] AMD, “AMD Reports Third Quarter 2026 Financial Results,” 3 November 2026.
https://newsroom.amd.com/news/amd-reports-third-quarter-2026-earnings/
[7] StockTitan, “Chip Stocks Bounce Back in the Stock Market Today Live, November 6, 2026.”
https://www.stocktitan.net/articles/stock-market-today-live-november-6-2026
[8] NBC Washington, “Live updates: Democrats win control of the House and Senate,” 4 November 2026.
https://www.nbcwashington.com/elections/election-day-after-2026-midterms-congress-results-live-updates/4177248/
[8b] CBS News, “2026 House midterm election results as Democrats capture majority with major wins.”
https://www.cbsnews.com/live-updates/house-election-results-2026/
[9] OilPrice.com, “Oil Prices Fall as Reports Say Hormuz Crude Flows Top Pre-War Levels.”
https://oilprice.com/Latest-Energy-News/World-News/Oil-Prices-Fall-as-Hormuz-Crude-Flows-Top-Pre-War-Levels.html
[10] Reuters via Yahoo Finance, “Strong October jobs report sends yields higher,” 6 November 2026.
https://finance.yahoo.com/economy/policy/articles/strong-october-jobs-report-sends-134858909.html
[11] US Bureau of Labor Statistics, “Employment Situation — October 2026,” 6 November 2026.
https://www.bls.gov/news.release/empsit.nr0.htm
[12] Investrade, “Market Review: November 06, 2026.”
https://investrade.com/market-review-november-06-2026/
[13] “Microsoft ends Oracle AI cloud rental deal as it reassesses $80 billion in contracts,” 6 November 2026, reporting a Bloomberg story.
https://globetv.app/news/microsoft-ends-oracle-ai-cloud-rental-deal-as-it-reassesses-80-billion-in-contracts-6aee32edc0ff3
[14] Daily market brief, 6 November 2026 — Thursday session detail, Texas data-centre pledge, NBC exit poll, Bank of England decision.
https://noctaaa.substack.com/p/daily-brief-friday-6-november-2026
Editorial notes. Index figures are closing values for Friday 6 November; weekly changes cover the full week. Several items rest on a single source and should be confirmed before publication: the VIX level of 28.18, the 52-gigawatt data-centre capacity estimate, the NBC exit-poll figure and the Texas pledge. Midterm results are reported as called by the named news organisations, not as official certified counts. The Microsoft–Oracle report originates with Bloomberg and was accessed through a secondary outlet; cite Bloomberg directly if the original is available. AMD’s free-cash-flow figure should be checked against the company’s own cash-flow statement. Analytical interpretations are Orion Capital’s own and are not attributed to the sources cited.
About the author
Ieva Bružaitė is the founder of Orion Capital, an independent investment research platform based in Vilnius, Lithuania, and the author of the children’s book Marso pasaka (Mars Tale). Orion Capital publishes research on artificial-intelligence infrastructure, semiconductors, energy and capital allocation under the principle that credentials are claims and research is proof.
orioncapital.lt | Investments with a Soul.