Weekly outlook · two weeks to October 16, 2026
Insurance on a Company You Don’t Own
The price of insuring one borrower’s debt went to distressed levels this month, and it dragged the whole semiconductor complex with it — while the Dow finished the same week higher. The likelier reading is that investors were repricing how the buildout gets financed, before any evidence of weaker demand.
If you hold anything built on artificial-intelligence infrastructure, the past fortnight felt indiscriminate. It was not. In the week ending October 16 the Nasdaq Composite fell more than 4%, its worst week since June, and the S&P 500 fell about 2%. Over the same five sessions the Dow Jones Industrial Average rose 0.7%.1
Those three numbers describe one market. A decline that leaves the thirty-stock industrial index higher is hard to read as a general retreat, and is more consistent with a rotation out of one kind of exposure — though index returns alone cannot separate rotation from a broader reduction in risk. The semiconductor complex absorbed most of it, posting its steepest eight-session fall since April 2025.1
The more useful question is what moved it. The evidence points somewhere more specific than the headlines about an “AI bubble” imply, though it falls short of a clean explanation.
1. The week in one picture
Start with the dispersion, because it tells you what kind of event this was. Broad selloffs take everything down together. This one did not.
Chart 1: One Week, Opposite Directions
The Dow finished higher in the same five sessions the Nasdaq had its worst week since June.
A five-point spread between the Dow and the Nasdaq inside one week is the signature of a rotation, not a general retreat. Whatever happened, it happened to a specific set of assets.
2. The financing came under pressure
One apparent trigger was a company most readers do not own. CoreWeave rents graphics processors to companies training AI models, and it has financed that build-out with debt and related data-centre obligations now exceeding $51 billion. In the week to October 16 the cost of insuring that debt against default moved to levels traders describe as distressed: its five-year credit default swap spread traded above 1,140 basis points before settling near 1,038, with that pricing implying, on the model’s own assumptions, a better-than-60% probability of default within five years — reported as an all-time high in the company’s intraday credit risk. The shares fell almost 12% on Friday.2
A credit default swap is protection against a borrower defaulting, and its price is a market-implied view of that risk rather than an objective forecast. When the cost of that protection rises this sharply, the question for shareholders in the chipmakers and equipment suppliers it buys from is whether funding is becoming expensive enough to constrain a major customer’s spending plans.
What is striking is what did not happen alongside it. No hyperscaler cut its capital-spending plans. CoreWeave’s own reported revenue backlog stood at $104.2 billion at the end of its second quarter, and its chief executive told staff the backlog more than covers its obligations. What that figure does not settle is the timing, profitability and funding needs of those contracts against the obligations they are meant to cover. J.P. Morgan argued the concerns are overdone, since the borrowing funds expansion rather than patching weak economics; D.A. Davidson countered that it is not clear the company will earn enough to repay what it has borrowed.2
Chart 2: Operating Indicators Against Credit Signals
What the selloff has, and has not, established.
Sorted this way the fortnight looks less like a verdict on artificial intelligence than a reassessment of the leverage behind it. That remains a working explanation rather than a settled one.
3. We wrote this mechanism down in August
Our August work on Micron asked a deliberately awkward question of its own thesis: what kills this without a single hyperscaler cutting spending? The answer written at the time was the financing layer — roughly $3 trillion of off-balance-sheet AI commitments growing faster than reported capital spending, set against a thirty-year Treasury yield at a nineteen-year high, with the note adding that capital-spending headlines would stay intact while the structures funding them repriced.
Naming a risk is not the same as knowing when or how it resolves, and writing it down did not help anyone avoid this drawdown. The value of having written it down is narrower and more practical: it tells you which numbers to watch now. If this is a financing event, it resolves through credit spreads and capital-spending guidance, not through chip demand data. So those are what the tests below measure.
4. What could clarify the outlook
The Federal Reserve meets on October 27 and 28, with the decision at 2:00 p.m. Eastern on the Wednesday. The target range is 3.75% to 4.00% after September’s increase, and the market prices a hold at roughly 86%. The median official saw one more increase this year, which leaves room to pause this month and move in December. The ten-year yield, which touched a 5.31% cycle high on October 5, sat near 5.1% on Monday.3
The larger event for anything AI-related is the hyperscaler reporting season and, inside it, the first real guidance for 2027 capital spending. That is the number that either confirms or refutes the market’s new suspicion. Estimates put those reports in the last days of October; confirm each company’s date on its own investor page rather than trusting a calendar aggregator, because we have been caught by exactly that before.
Chart 3: What Lands Next
Three weeks that decide whether this was a scare or a turn.
Stable or higher spending guidance alongside narrower credit spreads would support the financing reading. Broader spending cuts would challenge it. Continued stress in one borrower’s credit would be the least conclusive of the three.
- Hyperscaler 2027 capital spending (late October): if the combined guidance from the large cloud buyers points higher than 2026, the demand case survives the credit scare and we treat the drawdown as a repricing of leverage, not of orders. If the guidance points flat or lower across more than one major buyer, we treat the cycle as turning and say so; a single company guiding lower needs its size and reasons examined before it means anything about the cycle.
- CoreWeave credit (standing): a five-year default-swap spread back below 800 basis points says the credit scare is healing. At or above 1,140, the October high, says it is deepening and the equity complex has further to reprice.
- The Fed (October 28): a hold is priced at roughly 86%. A hold changes nothing in this framework. Any increase is a genuine surprise that tightens the same financing channel this note is about, and we would expect the AI complex, not the Dow, to take it.
- The ten-year (standing): a Friday close at or above 5.31%, October’s cycle high, keeps the pressure on long-duration equity. Below 5.31%, and sustained, one source of pressure is easing even if the credit question is not; simply sitting under a recent high proves little on its own.
One borrower’s credit is not an asset class. CoreWeave is unusually levered and unusually exposed, and reading the whole AI complex through its default swaps may be exactly the overreach that J.P. Morgan is describing. Monday’s session went the other way, with chip shares bouncing ahead of earnings, which is a reminder that this can reverse without becoming a spending slowdown. The dispersion argument also cuts both ways: an index that holds up because money rotated into defensives is not obviously healthier than one that falls together. And the author holds several of the names in this complex, so holding the sector makes it more important, not less, to separate evidence that supports this reading from evidence that would overturn it.
Sources
- Week ending October 16, 2026: the Nasdaq Composite fell more than 4% for its worst week since June, the S&P 500 fell about 2%, and the Dow Jones Industrial Average rose 0.7%; Friday's declines were 2.14% for the Nasdaq Composite and 1.09% for the S&P 500; the broad chip index recorded its steepest eight-session percentage decline since April 2025 (CNBC; Yahoo Finance markets wrap; Business Recorder). Monday, October 19: chip and AI-linked shares rebounded ahead of earnings (Detroit News / AP). ↩
- CoreWeave: debt and data-centre obligations above $51 billion; five-year credit default swap spread above 1,140 basis points before settling near 1,038, with swap pricing implying a better-than-60% probability of default within five years, reported as an all-time high in intraday credit risk; shares down almost 12% on October 16; $104.2 billion revenue backlog at the end of the second quarter; chief executive Mike Intrator's statement that the backlog more than covers obligations; J.P. Morgan's view that the concerns are overdone and D.A. Davidson's contrary view (CNBC; CNN Business; Yahoo Finance). ↩
- Federal Open Market Committee meeting October 27–28, 2026, decision 2:00 p.m. ET on October 28; target range 3.75%–4.00% effective September 17; market-implied probability of a hold around 86%; the median official projected one further increase in 2026 (CNBC; Chase market commentary). Ten-year Treasury yield: 5.31% intraday high on October 5; roughly 5.05%–5.15% on October 19 depending on source and time of day (Trading Economics; Forbes Advisor). ↩
- The August framing of the financing layer comes from our August research on Micron, quoted here from the note in which it was written.
Methodology: tests are graded against their written wording, not their intent. Index and credit figures are taken from dated contemporaneous reporting and are labelled where sources disagree. Company reporting dates marked estimated are not company-announced. Nothing here is a price target.
Nothing on this page is investment advice. We work in scenarios and trip-wires, not predictions. See disclaimer.
Disclosure: the author holds long positions in several of the semiconductor and AI-infrastructure companies whose shares are discussed in general terms in this note, including Micron and Nvidia, and holds no position in CoreWeave. Protective stop orders were triggered in several of these holdings on October 19, at the end of the period described, which reduces rather than increases the author's exposure to the outcome argued for here. Those exits are still being reconciled against broker records. This note is research and opinion, published to all readers alike, and is not investment advice for any person. bpleon research received no compensation from any company named. See disclaimer.