Treasuries Just Did Their Oldest Job for the First Time in Months — Skiba Names the Bond-Equity Divorce the Tape Has Been Hiding

On the Friday close of October 16, Andrzej Skiba — Head of U.S. Fixed Income at RBC Global Asset Management and the BlueBay team lead — went on CNBC's Closing Bell: Overtime and made a market-structurally unusual call: \"First time in awhile we've seen Treasurys rally into a weaker market … bonds are once again a 'safe haven' for investors.\" The framing matters because 2026 has not been a year in which bonds protected capital. The benchmark 10-year had climbed roughly 130 basis points since late February, the 30-year 100 basis points, and the 30-year touched5.73% on October 7 — the highest level since June 2002. For most of the year, bonds and stocks have sold off together; the safe-haven trade has been gold and cash, not duration.

What changed in the week ended October 16 was that the 10-year yielded from5.36% to 5.085%, the 30-year from 5.73% to 5.434%, while the SPDR S&P 500 ETF Trust slipped on the week (Nasdaq −1.77%) and a $2 trillion chip rout emptied risk appetite. That move is what Skiba was identifying: not the rally itself, but the divorce — bonds and stocks decoupling at last, with the 3-month correlation between the 10-year yield and the SPDR S&P 500 ETF Trust at −0.62, the lowest in at least fifteen years.

For most of 2026, the bond-equity correlation has been positive — both assets falling together as inflation expectations re-anchored higher, Big Tech borrowing competed with Treasury supply, and the federal deficit pushed the annual interest bill above $1 trillion for the first time. That is the IMF's \"less protection from selloffs\" regime: in stress, both legs of a 60/40 portfolio bleed.

The Oct 16 break is therefore a regime signal, not just a price move. If bonds resume their safe-haven function — rallying into equity weakness — the marginal allocator (the hedge fund, pension, RIA) gets a new tool. It also rewrites the calculus behind the iShares 20+ Year Treasury Bond ETF trade, which had been a -22% drawdown story for most of the year and now holds a 5.49% 30-day SEC yield against an SPDR S&P 500 ETF Trust earnings yield of just ~3.8%. The \"TINA\" frame that supported equities — there is no alternative to risk assets at sub-1% cash yields — has flipped: cash and Treasuries now offer a real yield again.

U.S. Treasury Secretary Scott Bessent spent August and September trying to engineer a long-end rally the conventional way. On August 19, 2026, the Treasury doubled the maximum size of long-dated buyback operations from $2 billion to at least $4 billion each, effective September 9 — the so-called 'Treasury twist.' Bessent later signaled the ceiling could go higher, citing a near-$1 trillion Treasury General Account available to deploy. The move was explicit supply management: thin out long-bond float, compress term premium, restore the 'regular and predictable' issuance framework.

It didn't work. Yields kept climbing into October, with the 30-year hitting 5.73% on October 7 — the highest since June 2002. Bloomberg's verdict: the intervention 'really did work at first, but Bessent's intervention upended the Treasury's regular and predictable policy framework and introduced uncertainty.' In other words, by stepping outside the auction calendar to manage curve shape, Bessent spooked the very buyers — pension funds, foreign reserve managers, insurance balance sheets — whose long-dated demand he was trying to unlock. Short-dated debt already accounts for more than 22% of total government borrowing; the term-premium channel was already broken before the buyback.

What finally moved the tape was risk-off in equities, not bond-market plumbing. Reuters' Oct 15 dispatch — chip stocks set to lose $2 trillion in market value after Broadcom's weaker-than-expected AI revenue forecast triggered a global selloff in NVIDIA, Advanced Micro Devices, Taiwan Semiconductor Manufacturing, and ASML — was the demand-side shock Bessent's buyback could not manufacture. Investors sold the AI complex and bought duration; the iShares 20+ Year Treasury Bond ETF rallied from multi-year lows near $77.55 to ~$80, the iShares 7-10 Year Treasury Bond ETF closed at $90.42. By Oct 16, the 10-year had yielded 28 basis points off its peak.

Layer the bond market onto the equity-market diagnosis and the picture tightens. The same corporate borrowing binge that helped push the 10-year to 5.36% — hyperscaler capex, AI-infrastructure debt issuance — is also the engine keeping S&P 500 earnings growth positive. Per the Seeking Alpha Oct 11 piece, AI infrastructure spending accounts for over half of S&P 500 EPS growth in 2026. The trade that broke last week was not just 'risk on / risk off' — it was the funding assumption underneath the AI capex cycle repricing.

That has direct implications for which equity pockets carry the most risk if the rotation continues. The chip complex — NVIDIA, Broadcom, Advanced Micro Devices, Taiwan Semiconductor Manufacturing, ASML — is the most direct beneficiary of capex assumption and the most obvious funding casualty if Treasury yields rise further. Their combined drawdown of roughly $2 trillion in market cap on Oct 15 is the single largest evidence point for Skiba's decoupling call: when AI funding stress hits, the bond market is the absorber, not the amplifier.

Near-term, the marginal allocator is watching two screens: the AI tape and the refunding announcement. Whichever blinks first will tell us whether the Oct 16 bond rally was a regime change or a one-week reflex. The Bessent constraint matters here — having already intervened, Treasury's next move is more likely to be a refunding tweak (composition, not size) than another buyback escalation. The buyback tool has been used; the marginal bidder is now the foreign reserve manager and the US pension fund, not the Treasury.

If the Oct 16 decoupling persists, three structural shifts follow over the next 1–3 years. First, the asset-allocation consensus re-rates: with the10-year yielding 5%+ and the SPDR S&P 500 ETF Trust earnings yield stuck near 3.8%, the historical '60/40' benchmark tilts back toward fixed income — and active managers who can pick duration (Skiba's whole pitch) earn their keep in a way they haven't since the 2010s.

Second, the AI capex cycle has to clear a higher hurdle rate. With BlackRock reporting record assets under management and AI infrastructure spending accounting for half of SPDR S&P 500 ETF Trust EPS growth, the marginal funding source for the chip complex shifts from cheap balance-sheet cash to 5%-ish term debt. That compresses the IRRs on the hyperscaler buildout and pressures the valuations of NVIDIA, Broadcom, Advanced Micro Devices, Taiwan Semiconductor Manufacturing, and ASML. The bear case is the 2000 dot-com analogue: the index holds up via the megacaps while the median S&P name bleeds; the bond is the only liquid hedge.

Third, the policy reaction function gets harder. Bessent's intervention, the Goldman Sachs Group trading-revenue beat (a direct beneficiary of bond volatility), and the JPMorgan Chase earnings miss together tell the same story — fixed-income volatility is now a permanent fixture of bank earnings, and the Treasury's tools to manage the long end are limited. The 'regular and predictable' framework is gone; the marginal Treasury buyer in 2027 will price optionality, not auction size.