English · 简体中文 · 日本語

Seasons Do Not Service Debt

On 16 October 2026, Shikiho Online published a premium column by Daiwa Securities strategist Eiji Kinouchi arguing that CoreWeave's credit anxiety was “seasonal.” The visible subtitle attributed the autumn decline in high-yield debt to loss-taking and argued that it was not a problem for AI as a whole.[1]

The useful half of that claim is obvious: one leveraged neocloud is not the entire AI economy. The unsupported half is the leap from “not systemic” to “seasonal.” Seasonal tax-loss selling and fund-year positioning can amplify a move. They do not establish why a particular issuer was selected for sale, whether its refinancing economics remain viable, or whether the market is merely wrong.

The Numbers Autumn Cannot Explain

Reuters reported roughly USD 50 billion of CoreWeave debt, CDS near 1,000 basis points and an implied five-year default probability near 63%. A follow-up placed CDS around 1,200 basis points and yields on some bonds near 15–16%.[3][4]

CoreWeave's own first-half filing shows the funding equation. Second-quarter revenue was USD 2.575 billion, operating loss USD 49 million, net loss USD 626 million and net interest expense USD 640 million. First-half operating cash flow was positive USD 3.663 billion, but property-and-equipment purchases consumed USD 14.117 billion while financing activities supplied USD 13.985 billion. Net interest expense rose 121% year on year to USD 1.176 billion. Its top three customers accounted for 72% of quarterly revenue.[5]

This is a real company with real demand and real infrastructure. That is precisely why the risk is important. Real demand can still be financed badly. A backlog is valuable, but debt service occurs on a calendar while customer deployment, power delivery, construction and utilization can slip.

The Strong Countercase

A fair critique must include the bull case. Nvidia has a commitment covering unsold capacity, major customers have signed long-duration contracts, and CoreWeave reports a very large backlog. Evercore maintained an outperform rating, and other market participants warned against treating CoreWeave as the poster child for all AI.[6]

Those facts may support a bond rebound. They do not prove that a 15–16% required yield is a seasonal illusion. The relevant test is whether cash generation begins to fund more of the build-out before external capital becomes too expensive.

The Nikkei 88,000–94,000 Illustration

At roughly 65,324 on 16 October, 88,000 requires about 34.7% appreciation and 94,000 about 43.9%.[7]

A chart that repeatedly labels advances “30% higher” can illustrate a conditional pattern. It cannot supply a probability. At an unchanged valuation multiple, index earnings must rise by the same 35–44%. If earnings rise 15%, reaching 88,000 still requires multiple expansion of about 17%; if earnings rise 20%, reaching 94,000 requires roughly 20% multiple expansion.

Reuters' February strategist poll had a median of 58,500 for end-2026 and 60,750 for mid-2027. Kinouchi's range is therefore not consensus but an aggressive upside tail.[8]

A tail scenario may become reality. A paid publication should still show the earnings path, valuation assumption, rates, currency, probability and invalidation conditions.

Was This Nepotism?

I found no evidence of nepotism or an improper editorial relationship. Kinouchi is an established technical analyst with a long record, multiple number-one rankings and two QUICK forecast awards.[2]

That makes the broader problem more interesting, not less. Prestige is a reason to read a hypothesis. It is not evidence that the hypothesis is true.

A Declaration

This is not a personal attack and does not allege fraud, misconduct or bad faith. It is a declaration against a method: converting titles into evidence, charts into probabilities, and confidence into a substitute for valuation.

Shikiho literally evokes the four seasons. Seasonality suits the masthead.

But seasons do not service debt.

If someone forecasts Nikkei 94,000, ask: How much earnings growth? What multiple? What discount rate? What invalidates the call?

Without those answers, the reader is not buying analysis.

The reader is buying a priced belief.

Original Notion archive