Letter 183 was about what capital can produce. This one is about what capital costs to use — and for the first time in years, that second number is moving faster than the first.
The 30-year US Treasury yield has pushed past 5.25%, its highest level since November 2023. The 10-year has followed it up to 4.81%, its highest level since October 2023, after a five-session rally driven by a specific, dateable event: Fed Chair Kevin Warsh's commitment to combating inflation at the Jackson Hole Symposium. Markets responded by repricing the odds of a 25-basis-point Fed hike this month from roughly 40% to nearly 66% within a week. This letter follows directly from Letter 183's question about AI's productivity payoff, because the answer to that question is about to matter more than it has in years: capital is no longer free to be patient with.
Three forces are compounding into the same move. First, the renewed US-Iran confrontation has pushed oil and energy prices toward multi-week highs, feeding directly into inflation expectations at the exact moment markets were hoping for a cooling trend. Second, corporate bond issuance has surged, adding real supply pressure to a Treasury market already absorbing a growing federal deficit. Third, and most directly: Warsh's own Jackson Hole language was read by markets as confirmation that the Fed's next move is more likely a hike than a pause, let alone a cut — a genuine shift in the reaction function investors had been pricing for most of the year. None of these three forces required AI sentiment to shift at all. The cost of capital is rising for reasons entirely separate from the AI investment cycle Letter 183 examined, which is exactly what makes the two letters worth reading together rather than as competing theses.
For much of the past decade and a half, capital was cheap enough that the discount rate applied to a company's future cash flows was almost an afterthought — growth alone could carry a valuation, because the alternative use of that capital yielded very little. A 10-year yield near 4.8% and a 30-year above 5.25% change that arithmetic directly: every dollar of future earnings is worth less today, and every dollar of debt service costs meaningfully more. This is not a subtle shift. It is the return of a discipline markets had the luxury of mostly ignoring since 2009, with a brief, sharp reminder in 2022 that most of the market treated as a temporary spike rather than a regime change. The current move, layered on top of an economy already running elevated deficits and renewed geopolitical energy risk, reads less like a spike and more like a level.
The businesses most exposed are the ones whose entire investment case depends on that discount rate staying low: highly leveraged companies refinancing debt into a materially higher-rate environment, and long-duration growth stories whose value sits almost entirely in cash flows many years out, now discounted far more heavily than they were eighteen months ago. The businesses that become structurally more valuable in this environment are the mirror image — strong, current cash generation, manageable leverage, and pricing power that holds even as their own input and financing costs rise.
This is where the two letters meet. Letter 183 asked whether AI's productivity gains would arrive fast enough to justify the current pace of capital expenditure. A rising cost of capital raises the bar for that question directly: capital-intensive AI infrastructure — data centers, chip fabrication, grid buildout — was underwritten in large part against a lower discount-rate assumption than the one now in effect. None of that spending becomes unproductive overnight. But the return threshold it needs to clear to justify continued financing at scale has moved up, meaningfully, in a matter of weeks. Productive investment can still make sense even with expensive money — infrastructure that earns a real, durable return continues to earn it regardless of the discount rate applied to value it. What expensive money does is remove the tolerance for AI investment that is speculative rather than cash-generating, exactly the distinction Letter 183 already drew between AI as narrative and AI as infrastructure.
Cheap capital forgives an unclear investment thesis by giving it time. Expensive capital does not. The businesses and infrastructure projects that were always going to work on their own economics keep working. The ones that were depending on a low discount rate to look viable are the ones now being tested — by the bond market, not by sentiment.
A genuinely higher, more durable cost of capital reopens a debate that a decade of near-zero rates had mostly settled by default: whether a traditional 60/40 equity-bond portfolio construction still does the job it was built for. Bonds yielding close to or above 5% are, for the first time in years, a real competing use of capital against equities rather than a defensive afterthought — which changes the relative appeal of both sides of that split, and changes what "genuinely valuable" versus merely "cheap-looking" means across both asset classes. A stock trading at a depressed multiple because the market has re-rated growth expectations is not automatically cheap; it may simply be correctly priced for a higher discount rate. Distinguishing between the two is the actual work this environment demands, and it is a different kind of work than the multiple-expansion analysis that mostly sufficed for the prior decade.
This is checkable against specific, near-term events. Friday's US jobs report and the following weeks' inflation data will show whether the Fed's hawkish repricing holds or reverses. If the 10-year and 30-year yields settle meaningfully below their current multi-year highs over the next quarter without another Fed intervention, this letter's "regime change" framing weakens toward "spike" instead. If yields hold near or above current levels through the Fed's next meeting despite a hike already being priced in, that is confirmation the market believes this is structural, not transitory — and the businesses most exposed to refinancing risk are the ones to watch first.
The 30-year Treasury yield above 5.25% and the 10-year near 4.81% are real, dated, measured facts — not forecasts. They did not require an AI correction to happen; they arrived from oil, deficits, corporate debt supply, and a hawkish Fed stance moving together. What they change is the standard every investment now has to clear, including the AI infrastructure buildout Letter 183 examined. Cheap money forgave unclear theses by giving them time to prove themselves. Expensive money does not extend that grace period — which makes the distinction between genuinely valuable and merely cheap-looking the entire job from here.