The question went from framework to front page in seven days. On September 18, the counter-signal window opened: oil was deflating on a de-escalation valve, the yen had sunk on BOJ gradualism, and bond-demand repair chatter was building. This morning the FT runs "Are Big Tech bonds crowding out the US Treasury?" — and the tape answers before the article does.
THE TAPE: The 10-year Treasury closed at 5.12% — through the 5% cusp that had held as a ceiling, a 19-year high. The 30-year at 5.43% is the highest since 2004. Germany's 10-year Bund re-broke its 3.55% regime trigger, Japan's 10-year JGB at 3.06% is the highest since 1996, and the UK gilt sits at 5.32%. This is not a US-specific repricing. It is synchronized tightening round two, with the long end leading.
THE DEMAND LEDGER: The bear steepener was never only an inflation story — it was always a competition-for-capital story. The 2026 investment-grade supply wave runs to roughly $2.3 trillion, about a quarter of it from AI issuers. China has cut its Treasury stockpile to 2008 lows; Japan sold $128 billion in a single month; foreign holdings sit at nine-month lows. Into that demand hole, this week printed two markers: a weak $70 billion Treasury auction, and SoftBank's $11 billion junk bond — a record — clearing at up to 9.75% yields to fund an OpenAI bet.
THE MECHANISM: Crowding out is not a metaphor here; it is an auction dynamic. Capital is finite. AI issuers are relatively price-insensitive — they are competing for strategic compute capacity, not yield. Treasury buyers are price-sensitive. When the marginal dollar has an alternative that pays 9.75% junk or 5.4% hyperscaler IG, the Treasury must clear where that dollar is, not where term-premium models say it should. The FT's question is the right one because the answer determines whether this is a rates cycle or a regime change.
THE 1987 ANALOGY — AND ITS LIMITS: Larry McDonald's warning that bond yields delivering equity-like returns mirror the pre-Black Monday configuration deserves attention, but not for the crash-mechanics reason usually cited. The 1987 parallel that matters is the fracturing of the natural hedge. The ledger already shows it: TLT fell on equity risk-off days this month — September 14 and September 23 both saw long bonds sell alongside stocks. That is the stagflation signature: no flight-to-quality. When bonds stop hedging equities, every balanced portfolio is structurally longer risk than its label says, and the de-risking cascade — vol-targeting funds, CTAs, balanced funds — sells both at once. Portfolio insurance did not cause 1987; it determined the shape of the day. The modern equivalents are bigger and faster.
WHAT WOULD FALSIFY: Two valves exist. First, demand repair: Treasury buyback operations absorbing long-duration supply, foreign buyers returning, or a Fed signaling a pause that collapses the term premium. Second, oil de-escalation: Brent back below $100 and holding would drain the inflation premium doing most of the work. Both valves were open a week ago. Both closed this week. That is the entire story of the last seven days — and the reason the crowding-out question is no longer academic.
POSITIONING NOTE: The TLT-short expression remains intact with trim and exit triggers unchanged — this is not a fresh entry signal, and the equity leg of the transmission is the risk to monitor, not a short entry. The trade that matters now is whether the 10-year consolidates above 5% or mean-reverts. If it holds, the next referenda are the October FOMC (one-more-hike odds near 70%) and the IG spread complex against the 250bps stress threshold.