Updated: 21 hours ago
The question we kept getting after Friday's close was the same one: what actually happened this week?
We were in an unusual position to answer it. We measure the buyers and the sellers separately rather than inferring them from price, so we can show what changed hands rather than what the tape settled at. Most flow data infers direction from price; we classify the two sides directly. So we are publishing it.
Here is the three-session sequence across roughly 3,000 US names.
Wednesday: the policy trade
The morning after the midterms, institutions bought $5.6 billion of financials and sold $8.4 billion of technology in a single session.
That is a clean, legible trade: buy what a new Congress might favour, sell what looked crowded. Market-wide, institutions added $4.0 billion. Nothing in the price action that day looked unusual. What was unusual was the size of the sector rotation underneath it.
Thursday: everything sold
Thursday was the worst session for technology stocks since March 2020. The Nasdaq 100 fell more than 6 percent, roughly $2.3 trillion of index value, and Nvidia alone lost about $440 billion.
The tape tells you who did it. Institutional net flow across all eleven sectors was minus $6.8 billion. Technology, which had been sold hard the day before, was roughly flat to slightly negative. The selling was broad: financials lost another $3.1 billion, energy $2.6 billion, health care $1.3 billion.
Retail sold as well, minus $553 million in the AI complex, concentrated in memory and semis. On Thursday, both sides of the market were selling. That is what a genuine de-risking event looks like, and it is why the day felt different from an ordinary pullback.
Friday: the money came back, but not to the same places
Friday reversed most of it. Institutions bought $3.3 billion market-wide. Technology flipped to plus $2.2 billion. Health care added $2.7 billion. Real estate and utilities, which had held up all week, kept taking money.
But two things did not reverse.
Financials lost another $3.0 billion on Friday, their second straight session of distribution after Wednesday's $5.6 billion inflow. And over the full three sessions, energy is down $5.5 billion with no recovery at all.
The same story in single names
Zoom in from sectors to individual stocks and the pattern holds. Nvidia saw the largest institutional exit of the week at $3.7 billion, but Amazon at $3.7 billion and Apple at $2.6 billion were sold almost as hard.
The money that stayed went to software: Microsoft took in $3.5 billion, the single largest destination in the market. This was not one crowded name unwinding. It was broad de-risking, with the survivors concentrated in software rather than hardware.
The answer to the question
Add the three days up. Across all eleven sectors, institutional net flow was plus $437 million.
Essentially flat.
That is the answer our clients are looking for, and it is not the one the headlines suggest. Almost nothing was added to or removed from US equities over the most dramatic week in recent memory. What changed was ownership.
Money left energy and financials. Money arrived in real estate, health care, utilities and, on Friday, technology. The rotation was large and deliberate, and it happened while the aggregate market did nothing.
Why the aggregate hides it
A market-wide net of plus $437 million sounds like nothing happened.
It is what remains after a $6.7 billion exit from technology and a $5.5 billion exit from energy, partly absorbed by $4.4 billion into real estate and $3.8 billion into health care.
Those are enormous position changes that net close to zero because every trade has two sides.
Price can only show you the net. Flow shows you the composition. When the two disagree, a flat market on enormous internal rotation, the composition is the story, and it is invisible in the index.
The coincidence nobody connected
Here is the detail that reframes the whole week.
What the productivity data showed
On Thursday, the same morning the market decided AI was worth $2.3 trillion less, the Bureau of Labor Statistics published preliminary third-quarter productivity. Nonfarm business output per hour rose 3.6 percent at an annual rate. Output rose 3.7 percent. Hours worked rose 0.1 percent.
Read that again. The economy produced 3.7 percent more with essentially the same labor. Unit labor costs fell 0.7 percent. Year over year, productivity is up 2.1 percent, and this business cycle is running at 2.2 percent annualized against 1.5 percent for the previous cycle, back to the long-run rate of the post-war period.
The market sold the AI trade on fears that datacenter buildouts might stop. On the same morning, the official measure of American productivity came in strong enough to be consistent with those buildouts continuing. The data does not establish cause, but it does not support the fear either.
What GPU rental rates show
Our own data points the same way. GPU rental rates have been climbing across the board, per the SemiAnalysis and Silicon Data rental trackers, not just at the frontier.
Ordinary users are running open-parameter and near-frontier models on commodity hardware that was written off as obsolete, and doing real work on it.
That is where a 3.6 percent productivity gain with flat hours comes from: not fewer people, but the same people finishing more.
Demand for compute is physical before it is financial. The equities repriced on a narrative about the next two years. The productivity data, published the same morning, described what is already happening.
What to watch next
Two things.
Financials
First, whether financials stop bleeding. Two straight sessions of institutional distribution into a sector that supposedly benefits from the policy outcome is the most interesting divergence on the board. Either the policy read is wrong, or something else is being priced.
Technology
Second, whether the Friday bid in technology is follow-through or a dead-cat bounce. One session of plus $2.2 billion does not undo a $6.7 billion three-day exit. We will know within a week whether institutions are rebuilding or covering.
Tracking it from here
We publish the split daily. When flow stops explaining price, that breakdown is itself the signal, and it is time to look at news, macro surprises, or options-implied data instead.
Method
Method: XTech US Equity Sector Flow, Russell 3000 universe, approximately 3,000 names across eleven GICS sectors, 370 in information technology. Trades are classified as institutional or retail using multiple independent inference models applied to consolidated-feed tick data.
Direction is inferred from trading behaviour, not disclosed identities. The approach is validated against 13F filings: institutional inference shows significant agreement with reported position changes, retail inference shows essentially none, which is the expected result.
Net flow is buys minus sells, regular and extended hours. Sector aggregates are not index-weighted.
Exponential Technology Inc. — real-time supply and demand for the entire market.
Comments