Miguel Ángel TempranoEconomics, geopolitics and investment
ES·EN
The column by Miguel Ángel Temprano

The Thursday that cost big tech a trillion dollars

By · November 9, 2026· Investment· 9 minute read

On 5 November, Wall Street’s seven biggest tech companies lost more than a trillion dollars in a single session, and the blame fell on the midterm elections. The ballot box set the date, but the problem had been building for some time. Five of the companies investing most heavily in artificial intelligence now spend more than they generate, and their expansion depends on creditors, local councils and power companies.

On Wednesday 4 November, the Dow Jones rose 1,569 points, its best day since April 2025, celebrating a Democratic victory that made oil cheaper and promised a gridlocked Washington. On Thursday it lost 1,833. The Nasdaq fell 5.52% and entered correction territory, the S&P 500 shed 4.36% and the Philadelphia semiconductor index 12.3%. The seven giants lost more than a trillion dollars in market capitalisation, and more than $440 billion of that was Nvidia’s. The explanation arrived before the close. The Democrats have won, the Democrats want to curb the growth of data centres, and without data centres there is no artificial intelligence. It is a tidy story that, in my view, explains the day on which it happened and very little else. What we saw last Thursday is that an industry which until a year ago paid its way with its own cash has started to depend on people who can tell it no.

Buying the world means buying something that is almost three-quarters American.

It is worth looking at who fell, because the data spoils the headline, including the one on this article. Microsoft lost 3.6%, Alphabet and Amazon 2.2% and Meta close to 4%. The ones that really bled were the sellers. AMD plunged 17.3%, Micron 17.5%, Marvell more than 16% and Nvidia 9.26%. Shortly before the bell, in pre-market trading, Nvidia and Tesla, which makes cars, accounted for six of every ten dollars of the combined loss. If anyone feared that artificial intelligence was useless, they would sell the companies spending money on it. Instead, the market sold the companies being paid, and that changes the question, which becomes how much longer the customer will be able to keep buying. I admit there is another reading, because the chipmakers were also what had risen most, and they are the first thing sold when selling has to be done.

The elections matter, and it would be absurd to deny it. Pending certification of the results, the Democrats will control both chambers and have won the Texas governorship with Gina Hinojosa, who promised a moratorium on all data centre approvals. Of 41 local votes in ten states, the industry won only one, according to Ballotpedia’s count. Meta’s president admitted on Wednesday night that building will be harder, and Goldman Sachs estimates that the states likely to impose restrictions account for about 45% of planned capacity. But Goldman also believes those pauses will be short-lived, and the Republican governor Hinojosa defeated had already announced his own moratorium before the vote. In Washington, Bernie Sanders is calling for a national moratorium, while Trump holds the veto until January 2029 and the Democrats fall short of the 67 senators needed to override it. The real risk lives in town council meetings and in the commissions that set electricity tariffs, and that risk makes things dearer and slower, which is not the same as banning them. I find it hard to believe it justifies wiping $2.3 trillion off the Nasdaq-100 in a single afternoon.

A family that renovates its home on credit no longer decides alone when to stop.

Nor did the slump begin on Thursday. In October, the S&P 500 technology sector had fallen 6.8% while nine of the eleven sectors rose. In the five sessions to 15 October, hedge funds cut technology at the fastest pace since 2016, according to Goldman’s data. The most crowded trade in the market, in Bank of America’s fund manager survey, was being long semiconductors. On Wednesday night Qualcomm published weak guidance, and on Thursday the mechanics kicked in. Leveraged ETFs, which promise to multiply an index’s daily move and, to deliver on that, are forced to sell whatever is falling, offloaded more than $60 billion at the close, according to a JPMorgan note reported by Bloomberg. Retail investors sold individual stocks as they had not done since March 2020. That part of the fall says nothing about artificial intelligence and almost everything about how the money was positioned.

What does say something is in the accounts filed with the SEC a few days ago. Between January and September, Microsoft, Alphabet, Amazon, Meta and Oracle, the latter on its own fiscal calendar, invested $556.9 billion and generated $542.9 billion from their operations. A year earlier, after investing, they had $143.5 billion to spare. Now they are $14 billion short, something that had not happened since at least 2011, which is where the series Reuters uses begins. In the last quarter, Amazon generated $42,422 million and invested $75,878 million. Alphabet ended up $9,065 million in the red and Meta, for the first time, $2,120 million in the red. Only Microsoft is still paying for the works out of what it earns. Whoever pays for the renovation out of their salary can stop whenever they like. Whoever signs a loan has to answer to someone.

When the seller guarantees the buyer, it is worth asking what the seller knows.

And the bank has already turned up. Amazon has taken on $115.1 billion of long-term debt in nine months. Alphabet has sold $42.87 billion in new shares and bought back none, when a year earlier it repurchased $39.81 billion. On 2 November Oracle placed $10 billion in bonds, with the 2066 tranche at 8.4%, and at the same time registered a share sale for another $10 billion. In the same week as the fall, Microsoft came to market with $25 billion and Meta with at least $15 billion. LSEG calculates that these five firms have issued $370.7 billion this year, a fifth of all US investment-grade debt, although Goldman puts the figure at $296 billion. And that is only what is visible. Data centre leases already signed but not yet commenced add up to $1.3 trillion across these five companies, more than double their financial debt.

Those without a salary feel it first. CoreWeave, which rents out computing capacity and lives on debt, saw its CDS, the premium paid to insure against its default, rise above 1,289 basis points on Thursday, almost 13% a year of the insured amount. One of its bonds yields more than 16%, which is a credit card rate, not an infrastructure one. For the third quarter, the company itself expected to pay close to a quarter of its revenue in interest. Oracle’s CDS, which traded at 58 points a year ago, is now around 300, and S&P rates it one notch above junk.

And here is what strikes me as the most revealing part of the whole episode. The broad investment-grade credit index barely moved, from 0.91 to 0.93 points, and the yield on the ten-year Treasury fell from 4.90% to 4.84%. Nobody fled the United States or feared for its economy. The fear has an address, and it belongs to those who are building without enough cash.

The market punished the one being paid and forgave the one paying.

The relationships at that address are what I like least. Nvidia acknowledges in its quarterly report a guarantee of up to $105 billion for a campus that an OpenAI subsidiary will lease, and admits in writing that a default by OpenAI could trigger it. A chipmaker is guaranteeing its customer’s lease, like a car dealer co-signing the loan of the person buying a car from it. Microsoft booked $10 billion from OpenAI in the quarter, and Amazon holds $92.5 billion in Anthropic preferred shares and $146.2 billion in convertibles from the same company on its balance sheet. Each is at once supplier, creditor and shareholder of the other. That works as long as someone keeps putting in fresh money, and OpenAI has just closed about $11 billion in a round for which it was seeking more than $30 billion, according to reports this week.

I admit the opposing argument carries weight. The demand is real. Google Cloud grew 103% in the quarter, Azure 48% and AWS 39%. The S&P 500 trades at 17.7 times forward earnings, its lowest level since November 2023 according to FactSet, and technology at 17.3, below its 25-year average. Among the strategists Bloomberg canvassed on Saturday, none has cut their target, and Mike Wilson of Morgan Stanley talks of adding risk. I am not so sure this is cheap. Those forward earnings have risen 33.6% this year, and part of them are accounting gains on equity stakes, which at Alphabet reach $126,755 million in nine months and at Amazon $62,853 million, with their own customers in the portfolio. Nor has the full cost arrived yet. Alphabet has $145.27 billion in assets that have not yet been placed in service and which, therefore, have not started to be depreciated.

From Spain, all this has been watched with a certain relief, and the relief is misleading. The Ibex rose 0.59% on Thursday, partly because Madrid closes before New York, and ended the week at 19,369.50 points, up 3.84%. Three of those almost four points were gained on Wednesday, with Santander and BBVA rising more than 5% and Brent below $100. The Ibex won by being in a different business. Those who make what people stopped buying paid for it the next day, with South Korea’s Kospi falling 6.02% and the Nikkei 5.96%. The Spanish saver with a global index fund, however, is not in the Ibex. At the end of October the MSCI World had a 72.77% weighting in the United States and 29.84% in technology. Many people believe they have diversified when what they have actually done is spread the same bet across 1,248 stocks.

A lease that has not yet started does not count as debt, but it is paid all the same.

I have no idea whether Thursday was the beginning of something. The S&P 500 remains less than 6% below its 6 October high, which is hardly a catastrophe. What I do know is that money is getting dearer just when it is needed most. The Fed held rates in October with two votes in favour of raising them, payrolls grew by 165,000 against the 59,000 expected by the Dow Jones consensus, and the market puts the probability of a hike on 9 December at about 77%. Rates did not cause Thursday, but they will set the bill for everything that came before it.

Everyone has marked 17 November in bold, when Nvidia reports results. I would look first at Tuesday the 10th, when CoreWeave reports. Nvidia will tell us how much it has sold. CoreWeave will tell us at what price it is being allowed to keep buying. For three years we have asked when artificial intelligence would make money, and that question could be put off. Now it is the lender who is asking, and the lender has to be answered on the maturity date.

Share LinkedIn X Email

On the same subject

Get every column by email

You will be told whenever something new is published: a column, an analysis or a research note. No advertising, and your details are never passed on.

What would you like to receive?
Language

One click to unsubscribe in every email.