BlackRock Bought Treasuries Before the AI Trade Cracked

The Nasdaq Composite lost $2.5 trillion of market value on Thursday, the largest single-session markdown in the index's history, and it took a 5.5% decline to get there. The bond market had already made the opposite call five weeks earlier. The 10-year Treasury yield peaked at 5.36% on October 7, the highest since April 2002, and had fallen back below 5% by the end of October, before any midterm ballot was counted. BlackRock Investment Institute's October 12 commentary is the document that explains that sequence, and it reads as a narrow call about the price of money rather than a broad call about artificial intelligence.

What BlackRock Actually Changed in October

Four moves landed in the opening paragraph of that note. The institute upgraded U.S. Treasuries and said it preferred short-term bonds. It funded the position by downgrading U.S. equities to neutral. It trimmed an underweight in Japanese government bonds. And it kept its conviction in AI fundamentals inside U.S. equities, which is the part most of the coverage skipped.

The reasoning was narrower than the headline. The note argued that "potentially excessive Fed tightening expectations" had opened room to unwind a Treasury underweight that had been a mainstay of the house view for most of six years. That is a judgment about the price of money, not a verdict that the AI cycle had run its course. The context behind it: 10- and 30-year Treasury yields had touched 24-year highs the week before, then retreated after strong auctions.

Timing deserves a fair reading. The 10-year peaked four days before the note was published, which puts the upgrade into the retreat rather than at the top. That is a more useful thing to study than a victory lap. It shows that a house the size of BlackRock will move off a six-year underweight when auctions start clearing strongly at high yields, and that the signal it reads is the auction, not the narrative.

The error investors make is bundling two trades. Believing the Fed is too tight and believing AI is finished are separate positions, and BlackRock held the first while explicitly declining the second. The Fed trade paid first, and it is now the crowded one, which is what makes the collapse of the hike trade instructive rather than merely topical.

Five Weeks When the Two Markets Disagreed

To understand how bond yields and stock valuations moved in opposite directions, look at the timeline across October and early November:

• October 7 (10-Year Peak at 5.36%): The 10-year Treasury yield reached its highest level since April 2002, signaling that the bond market was demanding maximum compensation for borrowing.
• October 12 (BlackRock Upgrade): BlackRock Investment Institute published its guidance upgrading Treasuries to overweight, favoring the short end, and downgrading U.S. equities to neutral.
• October 22 (First Big Yield Drop to 4.95%): The 10-year yield plunged 19.3 basis points in a single session, its largest single-day retreat since March 2023, while equity markets kept pushing to new highs.
• October 30 (Month-End Close at 4.93%): Treasury yields stabilized under 5%, yet stock indexes ignored the repricing and continued trading at elevated multiples.
• November 4 (Post-Election Relief): Equity averages experienced a brief post-election bounce across the Dow and S&P 500, setting up the sudden unwind.
• November 5 (Nasdaq Correction to 24,315): The Nasdaq tumbled 5.5% in a single trading session, shedding $2.5 trillion in market capitalization as the financing disconnect caught up.

Between the yield peak in early October and the equity break in early November, rates and stocks told opposite stories. The 10-year fell 19.3 basis points in a single session on October 22, its largest one-day drop since March 2023, per CNBC's bond coverage. Equities kept climbing. The S&P 500 set a record in the week of October 5, its 28th of the year, and the Nasdaq's own record came later in October, the one the index now sits more than 10% below.

Rates were repricing while equities were not, and that gap survives as long as earnings can absorb it. AI capex growth supplied the cover: spending by the largest technology companies is on pace for $799 billion in 2026, a 92% increase on 2025, and the flow of memory and networking orders that feeds it is the same flow the chip complex trades on. Anyone tracking the memory supply chain saw pressure building through lead times long before it showed up in the index.

What changed is the math underneath that plan. The 10-year climbed from about 4.4% at mid-year to 5.36% at the October peak, which is roughly 90 basis points of extra financing cost on every dollar of data center debt raised against it. A buildout that works at 4.4% is a different business at 5.4%, because the same 2031 revenue stream gets discounted at a materially higher rate. The mechanism sat in the yield, not in the narrative.

Where the Two and a Half Trillion Went

Thursday's selling did not stay inside the AI complex, which is the detail that separates a positioning unwind from a repricing. All 11 S&P 500 sectors fell and seven of them lost at least 2%. Caterpillar dropped 6.4% and JPMorgan 4.2%. The Russell 2000 fell 4.45% and joined the Nasdaq in correction. Inside tech the damage concentrated: Nvidia fell 9.3% and is now down for the year, Broadcom 10%, Oracle almost 7%, as the Journal's live coverage recorded through the session.

The trigger was political. Investors read the midterm results as a Democratic sweep of both chambers and began pricing a slower data center buildout. Senator Bernie Sanders of Vermont had called for a national data center moratorium, writing on X that "Americans don't want Big Tech oligarchs wrecking the environment, gutting jobs and threatening humanity's future." Jay Hatfield, chief executive at Infrastructure Capital Advisors, told the Wall Street Journal: "This is a true panic. This is 100% about the chip stocks and the potential slowdown in the AI trade."

Put the two headline numbers side by side and the size of the move explains itself. The index erased $2.5 trillion in a session against an annual AI capex bill of $799 billion, a markdown roughly 3.1 times the entire year's spending. Nobody liquidates 3.1 years of a capex programme in an afternoon because they have recalculated demand for servers. They do it because the rate dividing those future cash flows has moved, and because permitting politics became a variable inside a week.

The run rate matters too. Growth of 92% implies roughly $416 billion of spending in 2025, which means about $383 billion of new money has to be found in 2026 alone, split between operating cash flow, equity issuance and debt. That debt competes in the same market absorbing record sovereign supply, which is the plainest explanation for why the 30-year Treasury reached a 24-year high in the first place.

What It Means for Cash in Hand

At 4.93%, $100,000 in short-dated Treasuries pays $4,930 over the next twelve months, contractually, with no view required on data center permitting. The same $100,000 in the Nasdaq absorbed roughly $10,000 of loss on a 10% drawdown, and the index went past 10% this week. That comparison does not make anyone clever in a bull market. It is the one you want the morning after a record markdown.

The long end is where the care belongs. A 30-year bond bought here is a wager that AI borrowing demand and sovereign supply will both ease, and neither is showing it. Even a Fed that stops tightening will not pull mortgage rates down if term premium stays where it is, because long Treasury yields set the reference price for home lending. Anyone waiting on a cut to buy or refinance is waiting on the wrong variable. Meanwhile Brent crude traded near $118 a barrel at the end of October on Middle East supply disruption, raising household energy costs at the moment the case for cuts got harder.

The practical ordering is unglamorous. The front end is where the return gets paid, the long end is where the risk gets warehoused, and equities are the piece of the portfolio you rent rather than own. BlackRock's stated preference for short-term bonds over long bonds with high sensitivity to rates is the same instruction, written for institutional size.

The Variable That Settles This

Everything above turns on one number, and it is not the Nasdaq. It is the spread on data center and hyperscaler debt. If credit spreads widen from here while the 10-year holds near 4.9%, the market is repricing the cost of financing AI and the equity drawdown has further to travel, because a slower buildout means a smaller revenue base for every supplier in the chain. If spreads hold and yields fall alongside equities, this is a growth scare the Fed can cut into, and Thursday becomes the low.

Next week's calendar offers partial answers. Fed speakers have to work through a divided committee while carrying an energy shock into their inflation forecasts. The post-election auction cycle then puts real duration in front of real buyers within days of a 5.5% equity decline, and the clearing yield on that paper will say more about the next six months than any earnings call.

BlackRock's call was right and its shape still holds: collect the front end, rent the equity exposure, and demand a wider spread before lending anyone money for thirty years to build a data center. A 4.93% short Treasury is the only figure in this market you can bank, and every other number on the screen is somebody's forecast.

Disclaimer: This article is for informational purposes only and does not constitute financial advice. Past performance is not indicative of future results. Consult a qualified financial advisor before making investment decisions.