The Quarter in Brief

The U.S. enters 4Q with the soft-landing label doing a little too much work. Real GDP slowed to a 1.5% annualised pace in Q2 from 2.1% in Q1, but the labour market is not rolling over which is the part that keeps the market hanging: August payrolls rose 162,000, unemployment held at 4.1%, and June-July revisions added back 55,000 jobs. Wage growth eased to 3.1% y/y.  

‍

The awkward part is inflation – swing back and forth with no clear signs of direction.  July headline PCE was still 3.7% y/y and core PCE 3.34%, while real consumer spending was essentially flat in July and the saving rate fell to 3.0% - let’s stay positive on this – not negative saving rate – meaning spending tomorrow’s income today. That is not recession data; neither is it the clean disinflationary glide path that would give the Fed an easy exit.

Central-bank policy therefore remains restrictive for longer than the market wanted to believe. The Fed held the funds rate at 3.50-3.75% in July by a 9-3 vote, with three dissents preferring a 25bp hike. With inflation still above target and the Middle East supply shock contaminating the price data - Hormuz clearly is not helping.  

We expect September meeting to be dynamic rather than a ceremonial one. Our focus is less on whether the first move comes in September or October but more on the direction of travel: the burden of proof has shifted back toward inflation, so front-end yields can stay sticky even as growth cools – part of the curve most sensitive to Fed’s action.

Trade and fiscal policy add another layer of complexity. The July goods-and-services deficit widened to $88.6bn as exports fell to $310.7bn and imports rose to $399.3bn; capital-goods imports reached a record $140.3bn. Year-to-date the deficit is still 29.6% smaller than a year earlier, but the monthly reversal matters because it shows how investment-heavy the economy has become.  

‍

Fiscal supply is equally important for markets: Treasury expects $739bn of privately held net marketable borrowing in Q3 and another $628bn in Q4. That does not guarantee higher long yields, but it means the long end must keep absorbing very large supply while inflation risk and term premium are already elevated.  

Government policy is pulling in different directions – alignments in discord we would say. Tariffs and industrial policy support domestic production and supply-chain security – MAGA slogan here we go (a side note: now a new slogan, MAHA – Make America Healthy Again), but they also raise input-cost and retaliation risks. The suspension of heightened reciprocal tariffs on Chinese imports runs only through 10 November, so trade policy remains a Q4 event risk rather than a settled backdrop and as we can see, Trump administration will find ways to tackle the sensitive subject with fresh tactics. At the same time, the Middle East remains the macro switch and such switch has been on and off for longer than anticipated. A durable reopening of Hormuz would relieve oil, inflation and the Fed; renewed disruption would do the opposite and right now it’s touch and go. In this environment, geopolitics is not a separate chapter at the back of the report — it sits directly inside the inflation and rates forecast.

Since the world first learned about ChatGPT in November 2022, the AI race is now large enough to be a macro variable in its own right and to be fair, a dominant factor. The biggest AI providers/hyperscalers are expected to spend roughly $755bn in 2026 (and trillions more in the coming years), and individual budgets have become extraordinary: Meta guides to about $130-145bn of 2026 capex, while Alphabet raised $49.6bn of equity and $20.3bn of senior notes in Q2 partly to scale AI infrastructure and global compute. The race is now about who can secure power, data-centre capacity, networking, memory, grid access and — increasingly — cheap enough financing. The GPU race has quietly become a power-and-balance-sheet competition. Music will not stop for a while.  

‍

Our base case for 4Q is therefore - supply-shock stagflation-lite rather than a conventional late-cycle slowdown – we are not there yet. Growth remains positive, consumption is softer but not broken, and AI investment continues to provide a powerful private-capex floor. At the same time, let’s do not take our eyes off the obvious - the price of capital. That is why our convictions below favour the companies receiving the capex, remain cautious on leveraged AI financing, and treat the Treasury long end with more respect than the front end. The growth story remains intact; the financing environment is where the regime has changed.

Review of Our Previous Convictions

As always, we start by checking our own homework. The 18 June H2 framework got the Fed hold right but not on the front-end trade; broadening, short-duration IG and gold worked modestly; the soft-dollar call was mixed; and the biggest macro miss was assuming Hormuz normalisation would persist. The scorecard below is the one we prepared against market data through 4 September.

The lesson is not to become less directional; it is to be more explicit about the transmission mechanism. We were right that the long end and AI-capex beneficiaries mattered, but too early in front-end duration and too optimistic about the geopolitical supply shock.  

Q4 starts from that correction.

Our 4Q Convictions

1. Fed — a September 25bp hike is the base case; the path matters more than the first 25bp – either way

‍

The Fed held 3.50-3.75% in July by a 9-3 vote, with three dissents for a 25bp hike. Warsh then used Jackson Hole to make the inflation priority unusually clear. With core PCE still at 3.34% and the August jobs report removing much of the immediate labour-market case for patience, our view is that the Fed hikes in September as a precautionary move rather than waiting for the trend to fully confirm itself. The higher-conviction call is at least one hike before year-end given the lingering effect from the supply shock in the Middle East. The September dots and press conference matter more for Q4 than whether the first move comes this month or next.

‍

2. Front-end rates — the 2-year stays sticky; cheap is still the wrong adjective

The 2-year was 4.59% on 11 September, above the 3.50-3.75% policy range. The market has been pricing a live tightening path across the remaining meetings, not one isolated decision.  We see a September 25bp hike is the more likely precautionary move. A hold only becomes durably bullish for the 2-year if it comes with clearly softer inflation and renewed payroll weakness – conditions we don't yet see. Until then, a dovish surprise can create convexity, but the base case remains sticky and elevated yields – a tug of war stance.

3. Treasury curve — the long end remains the dangerous end

The 10-year closed 8 September at 4.80% and the 30-year at 5.25%. The August 30-year auction cleared at 5.216%, the highest auction yield since 2001, even with respectable demand. Treasury's larger long-end buybacks should improve market functioning, but such action will not remove fiscal supply, inflation risk or term premium – and the way we see it, the effect and impact to the market will be short-lived.  

Our take here: long-duration Treasuries remain difficult to own without a convincing inflation break – which we do not expect one soon.

‍

4. FX — tactically firm dollar, structurally less comfortable dollar

DXY was around 99 on 11 September, above the mid-90s we expected in our last call. With a Fed hike now in September, the dollar remains tactically firm; large deficits and high long-end financing costs argue against extrapolating strength. Q4 base case: 97–101 — tactically firm, structurally challenged – we expect both forces to pull against each other and as a result, DXY will be range bound in 4Q even with the September hike delivered.

5. Equities — follow the capex dollars, not the index label

The S&P 500 can still grind higher on AI-linked earnings, but the better signal is breadth: what we have seen - equal weight was up 14.07% YTD through 4 September versus 12.75% for the cap-weighted index.  

Our conviction is rotate away from pure mega-cap beta toward companies monetising the buildout — electrical equipment, grid and power infrastructure, data-centre infrastructure, industrial automation, selected financials and memory/semiconductor names with real pricing power. Put in simply, own the companies cashing the capex cheques. Backlog is useful but need to be mindful on real backlogs vs commitments (one can be 100% committed without delivery); free cash flow still pays the interest bill (money talks, and talks loud).

‍

6. Credit / AI financing — the canary is still in spreads, not the VIX (very calm)

Amazon, Alphabet, Meta and Oracle had issued about $194bn of bonds through 7 July, 79% more than their roughly $108bn issuance in all of 2025. Median spreads widened across maturities, and let’s pounder here, 78 of 91 comparable 2026 hyperscaler issues were trading at higher yields than at issuance by late July. Some of that is supply fatigue (too crowded or too full). The more important signal is the weaker balance-sheet tail: Oracle was cut to BBB-, while CoreWeave's secured borrowing costs stepped from SOFR +225bp to +450bp and then +550bp.  

‍

Our highest-conviction remains on financing: equities can ignore it for a while; credit rarely does.

‍

‍

Conclusion

Into year-end, avoid heroic calls on the S&P or one Fed meeting. The test is whether inflation breaks, the long end stabilises and AI financing stays orderly.  

Until then, own the capex receivers, respect the cost of capital, and let data—not narrative—earn the right to change the duration view.

‍