A Trillion Dollars Left Semiconductors and Nobody Missed a Number

A Trillion Dollars Is a Strange Amount to Lose

Semiconductor stocks have lost more than $1 trillion in market value in this sell-off and the companies that lose it are the ones that supply the artificial intelligence that has driven the market for two years

Intel fell 21 percent over the course of July. The damage was not limited to any particular name and was not limited to the United States

Put the figure in proportion before continuing. A trillion dollars is greater than the total market value of all but a handful of companies on the planet and in a matter of weeks it left a single industry group without any of those companies reporting a bad quarter losing a major customer or issuing a warning. Businesses that existed in June were still existing in July selling the same products to the same buyers under the same contracts

A number this large deserves a moment of context. This is not a rotation between sectors where money leaves one place and appears in another. Losses of this magnitude in a concentrated group of companies are a reassessment of an assumption and the assumption in question is the one on which all commerce is based: that the current pace of spending on artificial intelligence infrastructure continues

The Selloff Crossed Borders

The most useful detail is where the damage fell. This was not a sell-off of American technology. The Korean and Japanese names were trapped along with the domestic ones

That geography matters because it tells you what was selling. A rotation out of expensive U.S. growth stocks would hurt U.S. growth stocks. A repricing of the AI ​​supply chain affects whoever supplies it and the supply chain runs through memory makers in Korea foundry and equipment makers across Asia and holding companies with big positions in all of it

It also means that the usual explanations don't add up. Currency movements domestic policy and rate expectations differ between those markets. What those companies have in common is a customer base and the market repriced that customer base

The Question Is Not Whether Demand Exists

Here's the distinction that most comments blur. No one involved in this sell-off argues that demand for AI chips has fallen. Order books have not been emptied. The question being asked is different and more difficult

Chip demand right now is not consumer demand. It is capital spending by a small number of very large technology companies building data centers. Capital spending is a discretionary decision made annually by a board of directors and is the most cyclical line of corporate finance because it can be reduced to almost zero without shutting anything down

So the market doesn't ask if people want the product. It asks whether the companies that finance construction will continue to finance it at this rate which is a question of their willingness to spend more than anyone's need for computing

Customer Concentration Is the Real Exposure

A supplier whose revenue comes from millions of customers has a demand curve. A supplier whose revenue comes from a handful of customers has a relationship and relationships are renegotiated

Buyers of advanced AI infrastructure are concentrated to an unusual degree. That concentration is what produced the extraordinary revenue growth because when a small number of very large buyers decide to build at the same time orders arrive faster than any diversified customer base could deliver them

It works identically in reverse. If two or three of those buyers decide to moderate spending in the same planning cycle read the same analyst reports and face the same shareholders revenue doesn't decline smoothly. There isn't a long tail of smaller customers cushioning it

This is why the market can reprice a trillion dollars upon a change in sentiment about capital budgets that have not actually been cut. The exposure is real whether the cut occurs or not

The Part of the Cycle Everybody Forgets

Memory is where the cyclicality is most brutal and memory makers stood out in this sell-off

The mechanism has been repeated for decades. Prices rise because supply is scarce. High prices justify huge investments in new capacity. Capacity takes years to build and arrives all at once usually after the demand that justified it has moderated. Prices collapse the industry registers losses investment stops and the cycle restarts

What makes it reliable is that plants are extraordinarily expensive and cannot be built quickly or partially. A manufacturer decides years in advance based on a forecast and the decision cannot be reversed when the forecast turns out to be wrong. All participants know this and the cycle happens anyway because the alternative to construction is to see a competitor take the market

The open question is whether the demand for artificial intelligence is large enough and long-lasting enough to absorb the capacity that is being built for it. If so this cycle is behaving differently than all the previous ones. This is a substantial statement and it is the statement implicit in the prices that these stocks had before the sell-off

Depreciation Is the Bill That Arrives Later

There is a second mechanism that acts on buyers and not sellers and it operates with a delay that makes it easy to ignore

When a company purchases data center equipment the cash comes out immediately and the cost hits the income statement over the assumed useful life of the asset. Therefore an accrual produces a period in which cash flow looks tight and reported earnings look good followed by a much longer period in which the cash has stopped going out and the depreciation charge is still there

That second period is when boards ask what the return on investment really was. If the revenue attributable to the spend has increased the charge is absorbed and no one thinks about it. If not the depreciation shows up on the income statement as a visible drag that finance teams must explain and the natural response is to slow the next round of spending

The assumed useful life is the variable to watch. Extending it spreads the cost over more years and reduces the annual charge which favors current profits and says nothing about whether the hardware will still be productive in the end. Shortening it does the opposite. Either way the accounting choice made during construction determines how painful the digestion phase seems

Where the Money Comes From Is Part of the Question

A construction financed with operating cash flow and one financed with loans behave differently under stress and the distinction decides how quickly spending can be cut

Cash-financed spending is the resilient type. A company generating huge profits from an existing business can continue investing during a downturn because the decision is all about allocation. It will slow down if returns disappoint and nothing forces it

Debt-financed spending is not resilient in the same way. Interest must be paid whether the investment worked or not and the obligation is fixed while the income it was supposed to produce is not. A company in that position facing disappointing returns does not have the option of waiting because lenders have a schedule. With a central bank that keeps rates high and leans toward increases refinancing that debt is more expensive than the original borrowing

The variety worth taking a closer look at is the arrangement in which a supplier helps finance its own customers whether through investment credit or favorable payment terms. It is a legitimate practice and favors the supplier's revenue as long as the arrangement is in place because part of what is recorded as a sale was financed by the seller. It also means that the supplier's credit exposure and its revenue exposure are the same exposure which is exactly the concentration that no one wants when the cycle turns

None of this is visible in a stock price. It is visible in cash flow statements in the gap between reported income and cash raised and in the footnotes describing related party agreements which is where a reader with an afternoon can learn more than a month of commentary provides

What Would Confirm Each Story

The two readings of this sell-off make different predictions making it worth separating them rather than arguing about them

If this is a capital cycle that is changing the signs are that capex targeting to buyers is being cut order lead times are shortening inventory is being built in the channel and pricing power is shifting from the supplier to the customer. These appear in disclosures rather than comments and they appear before revenue declines

If this is a valuation correction within a story of intact growth spending guidance remains the same order books remain full and the stock rally comes as earnings confirm what companies have been saying. In that version the sell-off is a drawdown rather than a turnaround and the trillion dollars returns

Both things have happened before in this industry more than once and the early stages seem identical. Anyone who claims to know which one it is within weeks of moving is guessing

How I Read a Selloff in a Capital Cycle

The habit I try to maintain is to separate the customer's willingness to spend from the end user's desire for the product because those are the two things that get conflated in every discussion about a supplier

The second can be perfectly healthy while the first collapses. A company can be selling in the face of genuine growing and lasting demand and still have a terrible year because its handful of customers chose the same quarter to digest what they already bought

The other habit is to distrust my own framework when a stock has already fallen a lot. A 21 percent drop in one month invites the conclusion that the bad news is already priced in and in an equity cycle that conclusion has historically been early by a wide margin because earnings revisions come after multiple compression rather than with it

The Case That This Is Not a Cycle At All

The cyclical framework has one obvious weakness which is that it has been incorrect so far throughout this development

Each previous stage of the AI ​​trade has produced a group of people who argue that the spending could not continue and the spending continued. If the technology really changes the productivity of a large part of the economy the capital that is deployed against it is not a cycle at the top it is infrastructure at the beginning and the correct comparison is with the early construction of electricity or telecommunications rather than with a memory glut

The awkward thing is that both the early stage of a genuine infrastructure build and the late stage of a speculative one feature huge capital expenditures justified by a story about the future. The development of telecommunications in the late 1990s produced fiber that turned out to be genuinely needed and many investors did not survive to see it used. Getting the technology right and the timing wrong has been a costly combination before

The Bottom Line

More than a trillion dollars has flown out of semiconductor stocks the damage crossed borders because it followed a supply chain rather than an index and the question driving it is whether a small number of very large buyers are still spending at the current rate. This is a question of capital budgets rather than whether anyone wants artificial intelligence and the two things are constantly mixed. Watch for guidance on capital expenditures order lead times and channel inventory rather than comments because these are thefirst to move.And hold both readings at once as an early infrastructure build and a late speculative build look the same from here