The 30-year US yield hit its highest since 2007 — and the government responded by doubling buybacks of its own long-dated debt, an intervention that lasted one day before yields rebounded. The bond stress is now the main transmission channel into AI valuations.
In five trading days, global bond markets put governments on notice — and for the first time in this cycle, a government answered back directly. The events are separate in form: a yield spike here, a central bank’s minutes there, an oil shock in the Gulf — but they describe one structure: fiscal stress being priced into every long-duration asset at once.
| Date | Event | Level / Size |
|---|---|---|
| Aug 17 | US 30-year yield rises to 5.31%, highest since 2007 (Bloomberg); Japan’s 10-year JGB touches 2.93%, a three-decade high not seen since 1996 (FT)[3] | 5.31% / 2.93% |
| Aug 18 | US 30-year peaks at 5.34%; global selloff spreads to Canada and Japan as fiscal worries take hold (Reuters, NYT). UAE halts all trade and financial ties with Iran after accusing Tehran of ballistic missile strikes[5]; Brent rises ~1% to $91.89[5] | 5.34% / $91.89 |
| Aug 19 | Treasury Secretary Bessent doubles long-end buybacks to at least $4B per operation (from $2B), effective Sept 9 – Nov 4; quarterly repurchase ceiling rises from $69B toward $83B.[1][11] Same day, Fed July minutes show “several” policymakers ready to raise rates and “many” saying a hike would be needed if inflation does not fall to target[2] | $4B+ per op / ~$14B added |
| Aug 20 | Yields rebound, wiping out the intervention’s effect — 10-year back above 4.70% (CNBC)[3] | +5bp in a day |
| Aug 21 | Ray Dalio calls Bessent’s move “a sign that a debt crisis is getting closer” and recommends gold and bitcoin (CNBC)[4]; Fed officials publicly cautious about the buyback program | — |
| Aug 22 | Dollar at a three-month low as markets balk; rotation out of crowded US AI infrastructure names into software and emerging markets accelerates (Schwab Network, BT)[6] | — |
Last week’s memo covered the equity side of this cycle — how AI companies are pricing themselves against 2028 forecasts. This one covers what is happening underneath: the discount rate itself is moving, and for the first time in years, the issuer of the benchmark asset has stepped into its own market to stop it.
The mechanics are simple; the precedent is not. The Treasury’s quarterly refunding program already includes repurchase operations — buying back older bonds to manage the maturity mix of outstanding debt. What changed on Wednesday was scale and intent: buyback sizes for 10- to 30-year securities doubled from $2 billion to at least $4 billion per operation. Bessent said later that day the program “could end up being more than the $4 billion” announced (Reuters, CNBC).[4]
The timing was deliberate. Evercore ISI described it as an “activist Treasury secretary — hitting bond shorts with a surprise announcement of an increased buyback program on an August day with thin liquidity and a lull in prior one-way bets on yields higher.” It is also the second direct market intervention this month: on Aug 1, Bessent joined Japan in a currency operation to reverse the yen’s slide toward four-decade lows.[7]
The result lasted exactly one session. The announcement drove the 30-year from its 5.34% peak down to about 5.18%; by Thursday, yields had climbed back above their pre-announcement levels, with the 10-year up more than five basis points to 4.70%.[10] Politico’s framing — “drop in the bucket” — captures the arithmetic: at least $14 billion of added liquidity support against a debt pool measured in trillions.
Source: Channel NewsAsia — “CNA Explains: Why have US Treasury yields surged, and why does it matter to Asia?” (27 Aug 2026).
The market read was unambiguous: this is a signal, not a solution. And signals only work while the underlying problem stays unchanged.
Three forces are pushing long rates higher simultaneously, and none of them is temporary in the way a one-week selloff would be.
Fiscal. Bloomberg’s own framing this week: “AI is driving up Treasury yields — it just touches everything.” Surging government spending and a flood of long-dated issuance are the base case. The AI buildout itself has become part of the supply story — data centre construction financed by public debt markets raises the volume of long-duration paper investors must absorb. Inflation that has been stuck above the Fed’s 2% target for five years means the real-yield component is doing work it should not be.
Policy. The July FOMC minutes, released Wednesday under Chair Warsh, showed a committee moving hawkish: “several” policymakers were ready to raise rates at the meeting itself, and “many” said a hike would be needed if inflation does not decline.[2] Talk of potential cuts has vanished from the document entirely (Reuters, CNBC). A Fed that is debating hikes while the Treasury doubles buybacks is two branches of the same government pulling in opposite directions — and markets price that contradiction into the term premium.
The oil kicker. Iran’s threat to go on the offensive in the Strait of Hormuz if diplomacy fails, its strikes on UAE tankers earlier this month, and the UAE’s decision to halt all trade and financial ties with Tehran. Together they have Brent sitting near $92.[5] Oil is the one variable that can turn a fiscal story into an inflation story — which would force the Fed’s hand in exactly the direction the minutes point.
The Japan parallel. Tokyo’s 10-year yield touched 2.93% on Monday, its highest since 1996, as a weak yen fuels inflation and DBS now expects the BOJ to hike in September while accelerating its tightening pace.[8] When two of the world’s largest bond markets stress at once — one on fiscal grounds, one on currency-inflation grounds — it is no longer a single-country repricing. It is a global re-rating of what long-duration safety is worth.
This is where the bond story stops being a macro footnote and becomes an equity problem. The transmission channel is the discount rate, and it hits long-duration growth assets first.
1. Every AI valuation from last week’s memo just got more expensive to justify. Anthropic’s IPO hinges on $190–200 billion of 2028 revenue[9]; CoreWeave’s backlog is $104 billion of contracted future cash flows; OpenAI’s run rate is being priced against compute commitments that mature over a decade. All of those numbers are discounted at long rates. When the 30-year moves from ~5% to 5.34%, the present value of cash flows ten years out falls by double digits — before any change in the underlying business. The equity story and the bond story are not two markets; they are one valuation with two inputs.
2. The rotation is already visible. Schwab Network reported this week that “markets are rotating out of AI infrastructure names, with memory and optical stocks seemingly handing the baton to software companies” — a shift from capital-intensive, rate-sensitive assets toward shorter-duration earnings. Business Times’ podcast framing was more direct[6]: bond stress is rotating capital out of crowded US AI positions into emerging markets. That matters for anyone holding an Asia-tilted book (EIMI, CNYA): the same force that pressures US mega-cap AI is a tailwind for EM equity flows.
3. The fiscal channel cuts both ways. If the AI buildout keeps expanding, Treasury issuance keeps rising, yields keep pressure — and the discount rate on the very assets driving the capex keeps climbing. That is not a contradiction to be resolved; it is a feedback loop with a ceiling. The question for portfolio construction is where that ceiling sits. If long rates settle at 5%+, the AI trade has to earn its multiple on cash flow, not forecast — which is precisely the standard Anthropic’s IPO will be tested against.
4. For an equity-heavy FI book, this is the main risk channel right now. Rising real rates compress equity multiples across the board, but they do so unevenly: long-duration growth first, short-duration value last. A book tilted toward Asian equities with meaningful cash and fixed income has a structural hedge against exactly this scenario — the rotation out of US AI into EM is the mechanism by which that hedge pays.
Singapore’s reserves are a monetary policy instrument before they are an investment portfolio. Official foreign reserves stood at S$550.7 billion, or US$433.0 billion, at end-August 2026[12]. MAS’s own accounting note is explicit about what moves the stock. Foreign exchange intervention to implement the exchange rate policy, transfers of assets in excess of what it deems necessary to maintain confidence in the Singapore dollar. Changes in the FX swap book held as part of money market operations to manage banking-system liquidity[12].
That is the same instrument class now being used at the long end of the US curve, put to a different purpose. One sovereign buys and sells to keep a basket exchange rate on its path and uses swaps to manage domestic liquidity; the other intervenes directly in its own long-dated debt. For a Singapore-based portfolio, the transmission runs through the reserves and the SGD curve, not through a domestic buyback facility.
The bear case on bonds has a real answer, and it comes from the intervention itself — plus three supporting points.
1. Buybacks are not QE, and they do not create the same problems. Repurchasing outstanding debt with existing balances is maturity management, not money creation. The Fed’s balance sheet is not expanding; the Treasury is reshaping its own liability mix. That distinction matters for inflation expectations — which have stayed anchored even as nominal yields rose.
2. The market digested it in one session and moved on. If investors believed the intervention signaled fiscal desperation, the 30-year would not have dropped 16 basis points on the announcement. It did — and then gave some of it back because $4 billion per operation is small relative to daily issuance. That is a liquidity response, not a confidence collapse. The two are different diagnoses with different implications.
3. The fiscal trajectory has room to improve. Revenue growth from the AI-driven expansion, plus the administration’s spending discipline rhetoric, means the deficit-to-GDP path is not locked in at current levels. Yields price a five-year fiscal forecast; one year of elevated issuance does not set it.
4. Dalio’s read is worth taking seriously — and he is not alone in it. His “debt crisis getting closer” framing[4], with the gold-and-bitcoin recommendation, is the tail-risk version of this memo: if buybacks become a standing feature rather than a one-off signal, the term premium will reprice permanently higher. The base case is that they don’t; the option value of being prepared for the case where they do is cheap.
Four near-term events will tell you whether this was a one-week wobble or the start of a new regime for long rates:
1. The Sept 9 buyback start. The first doubled-size operation is the real test — Wednesday’s announcement worked in thin liquidity; September will show whether it works when volume returns. If yields respond durably, the intervention has credibility. If they don’t within two operations, expect the program to be scaled up again — which is Dalio’s scenario becoming policy.
2. The BOJ meeting in September. A Japanese hike at 10-year yields near three-decade highs would test whether global bond stress is synchronised or US-specific. A BOJ move that triggers JGB outflows into Treasuries would be the transmission mechanism no one currently has a model for.
3. The next FOMC under Warsh. The minutes showed a hawkish contingent larger than the dissents indicated. Any signal of an actual hike — not just readiness — would be a new regime for AI valuations, because it would mean the discount rate is rising while the equity story still depends on 2028 cash flows.
4. Hormuz and oil. Brent at $91.89 with Iran threatening offensive action in the strait is the one variable that converts a fiscal story into an inflation story within weeks, not quarters. A sustained move above $100 would force the Fed’s hand regardless of what the Treasury does with buybacks. That would make the rotation out of US AI into EM and energy-producing markets a structural shift rather than a tactical one.
The government has now told you, in the most direct way possible since the 1940s, that it considers its own long-end yields too high. Whether that statement is true depends on what happens after September 9 — and on whether the AI buildout keeps adding to the supply of exactly the debt the Treasury is trying to buy back.
This analysis is based on publicly available data as of 2026-08-22. For related coverage, see The Thirty-Year Line and The Debasement Trade.