Research — 28 August 2026

Verdict

Constructive on the AI complex with dispersion: NVDA's beat and margin transparency ease the most acute input-cost concerns, and the trade broadens beyond semis into software and hyperscalers.

Analysis

The paper-led thesis is mechanical: NVDA beat on every line that matters (revs $96.2bn vs $92.4bn cons, gross margin 75.0% vs 74.8%, data center $89.0bn vs $85.9bn) and guided 3Q to $108bn vs $105bn, with GS's Jim Schneider arguing the CY27 72-73% medium-term gross margin guidance 'is likely to put to rest the most acute investor concerns on input costs' and that management transparency on customer financial guarantees helps investors size risk. The counterargument is embedded in the guide itself — 3Q gross margin guided lower at 74.0% vs 74.8% cons — and in the lackluster immediate reaction across parts of the Asian AI complex, suggesting positioning, not fundamentals, is the marginal driver. JPM's thematic work adds the portfolio consequence: the tape is returning to previous highs with 'broader participation i.e., no longer by the + Semis vs - Software pairs trade', so expression should favor breadth (hyperscalers plus software) over a single-semi bet.

What changed

NVDA's guide and margin disclosure converted an input-cost debate into a demand-visibility debate.

Desk views

GS's Schneider sees a clearer path to outperform: margin guidance de-risks the input-cost bear case and guarantee transparency improves risk assessment.

Evidence IDs · P003

JPM thematic frames the consequence as broadening: Tech moves higher with more dispersion, retiring the narrow semis-vs-software pair.

Evidence IDs · P007

GS Asia sales flags the lukewarm post-earnings reaction in Japan/Asia AI names — positioning, not fundamentals, is the near-term swing factor.

Evidence IDs · P005

Investor implication

Favor broadened AI expression (hyperscalers, software, selected semis) over concentrated single-name semis beta; expect dispersion to rise even as the theme grinds higher.

Trade expression

Long broadened AI basket / short narrow semi concentration; MRVL earnings as the next catalyst.

Invalidation

A MRVL guide miss or evidence that AI capex orders are decelerating would undercut the broadening thesis.

Verdict

Bearish/neutral the long end structurally; steepeners remain the durable rates expression, with only shallow relief available.

Analysis

GS's rates team is explicit: 'ongoing energy volatility and the AI borrowing boom are keeping G10 bonds under pressure, especially at the long-end', with inflation and fiscal drivers of higher term premium 'persistent' and steeper curves a durable feature. Crucially, they push back on the supply narrative — upward yield pressure 'has more to do with the macro environment than an excess of long-end bond supply', so WAM reductions and Treasury micro-measures won't arrest the backdrop. TS Lombard agrees the 10-year's rise reflects a new macro regime still not fully priced, and warns that fighting the weight of global capital demanding higher yields 'is not going to be sustainable', with the risk case being markets testing Treasury resolve if the Fed shows no hawkish potential. The disagreement is one of emphasis, not direction: GS Marshall calls Treasury news 'a little bit more micro than macro', TS Lombard worries about procyclical fiscal policy amid energy shocks. The near-term catalyst is Warsh at Jackson Hole — ING notes FX expects little (a ~35 pip USD/JPY straddle) and he may avoid monetary policy entirely, but any hawkish framing could jolt rates and risk assets.

Tension

Treasury attempts to support the long end vs. a macro regime demanding higher term premium — GS says micro, TS Lombard says the market wins eventually.

What changed

Treasury intervention attempts have shifted the debate from supply mechanics to whether authorities can suppress a macro-driven term premium.

Desk views

GS: term premium drivers are persistent; supply-side fixes won't lower yields; 10y UST to 4.40% end-2026 only if benign inflation/below-trend growth fundamentals hold.

Evidence IDs · P013

TS Lombard: most of the 10-year rise is a new macro regime not fully priced; fighting global capital flows at the long end is unsustainable.

Evidence IDs · P021

GS rates research (Marshall): Treasury news is micro not macro — it can support cheaper/less liquid curve parts but won't arrest the broader backdrop.

Evidence IDs · P006

Investor implication

Maintain steepeners as the core rates position; treat long-end rallies as tactical, not regime-changing; size equity beta knowing a hawkish Warsh surprise hits both legs.

Trade expression

Curve steepeners; fade long-end rallies; low-risk long-end selloff structures into post-Warsh clarity.

Invalidation

A Warsh speech explicitly endorsing Treasury long-end suppression with a credible mechanism would force a rethink.

Verdict

Neutral USD near term; the desk splits between markets fully pricing a December hike and houses (UBS) expecting the Fed to hold in September — data, not Warsh, decides.

Analysis

Natixis frames the anchor: 'money markets are fully pricing a December hike, and any hawkish framing on inflation could jolt rates and risk assets' at Warsh's first major speech. UBS takes the other side of the near-term debate — 'Our base case is that the Fed holds rates steady at its September meeting, as recent inflation and labor-market data have not materially strengthened the case for a further increase' — making next week's labor releases and the speech the key USD catalysts. GS's payrolls benchmark work adds a data wrinkle: the preliminary benchmark revision is likely the first upward revision since 2022 (+50-450k), lifting the measured pace of payroll growth from ~25k/month to 30-65k/month — a mechanical hawkish tilt to the labor narrative. JPM's global work tempers the hawkish read: global employment rose at a sluggish 0.5%ar last quarter, and their recoupling call expects above-trend GDP to eventually deliver stronger hiring. The portfolio consequence: USD direction hinges on whether the labor re-acceleration is real or a benchmark artifact, with Warsh as the near-term volatility gate.

Tension

Fully-priced December hike vs. UBS's September hold base case; upward benchmark revision vs. still-sluggish global employment.

Desk views

UBS CIO: Fed holds in September; labor data hasn't strengthened the hike case; favors selling USDCHF and EURNOK upside as hike expectations fade.

Evidence IDs · P022

GS: preliminary benchmark revision likely +50-450k, first upward revision since 2022, lifting measured payroll growth to 30-65k/month — a hawkish data tilt.

Evidence IDs · P014

JPM Kasman: global employment still depressed at 0.5%ar; recoupling to stronger job growth anticipated but not yet delivered.

Evidence IDs · P016

Investor implication

Stay neutral USD into the speech; express the divergence via crosses (short USDCHF, long NOK) rather than outright dollar direction.

Trade expression

Short USDCHF; long EURNOK on Norges hawkishness and carry.

Invalidation

A hawkish Warsh speech endorsing near-term tightening would validate the December pricing and break the UBS base case.

Verdict

Constructive Korean rates via curve flatteners; one more hike to 3.25% expected (potentially Q426), then done.

Analysis

CACIB's read of the 27 August meeting: the BOK hiked 25bp to 3.00% in back-to-back meetings to address inflation, supported by the semiconductor cycle and related income growth — with MPC member Hwang Kun il the lone dissenter for no change. The trade-relevant signal is the guidance: 'The BOK's forward guidance was more dovish than what the market expected with six-month forward guidance signalling one more rate hike until Q127', and CACIB expects one more hike to 3.25% potentially in Q426. The mechanism is the semiconductor cycle feeding income growth and inflation — meaning Korea's tightening is a byproduct of the same AI boom lifting global equities. Portfolio consequence: with the KTB curve 'quite steep relative to recent years', CACIB sees value in flattening positions, and the revised USD/KRW forecasts (1,385 end-2026, 1,360 end-2027, from 1,435/1,420) imply a firmer won path as the cycle matures.

Desk views

CACIB: dovish forward guidance caps near-term rates; one more hike to 3.25% then pause; KTB flatteners offer value; USD/KRW forecasts revised stronger.

Evidence IDs · P001

Investor implication

Position for the end of the Korean tightening cycle: KTB flatteners, and treat won strength as a second-order beneficiary of the semi cycle.

Trade expression

KTB curve flatteners; tactical long KRW exposure.

Invalidation

A re-acceleration of Korean inflation forcing more than one further hike would invalidate the pause thesis.

Verdict

Range-bound USD/JPY around 158-160; UBS holds 160 (Dec 2026) and extends 158 through September 2027 — no JPY catalyst without Fed cuts.

Analysis

UBS's thesis is that US factors dominate: 'a fairly resilient US economy and a still-hawkish Federal Reserve bias continue to keep USDJPY supported', and a decline toward the low 150s 'would likely require markets to price in Fed rate cuts, which remains a high bar for now'. On the domestic side, the BoJ may tighten slightly to contain yen weakness but policymakers won't welcome disorderly appreciation. The evidence: after the coordinated US-Japan intervention in late July drove the pair from ~164 to a low of 155.2, it has 'gradually recovered to the 158-159 range' — intervention effects fading on schedule. The tension with the rates theme is direct: if the Fed stays hawkish (December hike priced), USD/JPY stays supported; if UBS's September-hold call lands and cuts get priced, the pair breaks lower.

Desk views

UBS CIO: prolonged range-trading base case; intervention effects fading; JPY strength needs Fed cut pricing, a high bar.

Evidence IDs · P011

Investor implication

Treat USD/JPY as a Fed-expectations expression, not a BoJ trade; range strategies favored until cut pricing emerges.

Trade expression

Range trades around 158-160; buy JPY only as a hedge against a dovish Fed repricing.

Invalidation

Fed cut pricing or a BoJ hawkish surprise (rate hike + tolerance for strength) breaks the range.

Verdict

Constructive on China AI monetization — enterprise/token economics are scaling fast — but the equity reaction remains positioning-constrained.

Analysis

GS Asia sales lays out the mechanism: Chinese labs are converting AI demand into real revenue with a B2B tilt. MiniMax H1 revenue grew 283% YoY to $116.6m with Enterprise/API revenue up 703% and now 63% of total; 'July token consumption was 20x January levels, while August ARR has reportedly already exceeded $800m, with around 80% coming from B2B'. SenseTime shows the same pattern (generative AI revenue +28% to nearly 80% of group, recurring revenue +124%, margins expanding), and Z.AI's GLM-5.3-Flash at $0.15/$0.50 per million input/output tokens pushes ultra-low-cost, high-efficiency models alongside DeepSeek. The counterpoint is price action: 'The move post Nvidia is a touch lackluster across a lot of the AI complex, particularly in Japan' — and China AI equities broadly haven't broken out, with China consumer discretionary (CHIQ) sitting in RS breakdown territory per the internal snapshot (context, not a paper call). The portfolio consequence: China AI is a fundamentals-improving, sentiment-lagging lane — the divergence between token economics and equity performance is the opportunity and the risk.

Tension

Rapid enterprise monetization vs. lackluster equity reaction across the Asian AI complex.

Desk views

GS Privorotsky: Chinese AI labs show explosive B2B/token growth (MiniMax ARR >$800m, 20x token consumption) with ultra-low-cost models — demand is real even if equities lag.

Evidence IDs · P005

Investor implication

Watch China AI names for a catch-up re-rating as enterprise revenue compounds; size positions for continued sentiment lag.

Invalidation

Token/ARR growth stalling or renewed US export restrictions on China AI compute.

Verdict

Constructive industrials as the enabler sector; industrial metals supported by AI/electrification; agriculture offers an El Niño kicker; gold stays strategic.

Analysis

UniCredit's Short View makes the sector thesis: AI infrastructure, electrification, defence spending and supply-chain diversification are 'collectively driving capital expenditure across developed and emerging markets', with geoeconomics now 'a major earnings driver' — the industrial sector is 'increasingly emerging as one of the most important enablers of structural change'. The defence spillover is becoming visible in Eurostat manufacturing data, though UniCredit's Coffee Break cautions the contribution 'remains limited as long as Europe's defence industry operates through fragmented national markets' — a real constraint on the thesis. UBS's commodities work extends the same capex logic into metals: 'Industrial metals should stay supported by investment in AI and electrification', while wheat (+6.6%) and corn (+2.7%) hit three-year highs on Russia-Ukraine export disruption and El Niño weather fears, and gold remains 'a useful strategic diversifier'. The desk color corroborates: copper has been 'hovering around the same level' with FCX at $70 'a major standout' and retail demand outsized. The counterargument is fragmentation — the defence cycle's macro payoff is diluted by national markets — and the risk that AI metals demand is already priced.

Tension

Structural capex tailwinds vs. fragmented defence markets and already-priced copper breakouts.

Desk views

UniCredit: multiple structural investment cycles (AI, electrification, defence, supply-chain) converge on industrial demand; geoeconomics is now an earnings driver.

Evidence IDs · P027

UBS: broad commodity exposure justified beyond energy — industrial metals on AI/electrification, agriculture on El Niño, gold as strategic diversifier; favor active management as leadership rotates.

Evidence IDs · P025

UniCredit Coffee Break: defence spillovers visible in manufacturing data but contribution limited while Europe's defence industry stays fragmented nationally.

Evidence IDs · P026

Investor implication

Overweight industrials as a structural diversifier from concentrated AI beta; express metals via industrial metals exposure with agriculture as the weather-driven satellite.

Trade expression

Long industrials/industrial metals; agriculture as El Niño hedge; hold gold strategically.

Invalidation

A capex slowdown in AI infrastructure or evidence that metals demand is decoupling from the AI build-out.

Verdict

Bearish USDSGD toward 1.25 by Sep 2027 on Fed repricing; AUD faces a live RBA hike risk after the July CPI upside surprise.

Analysis

UBS's USDSGD thesis is dollar-driven, not SGD-driven: 'Further USDSGD downside is likely to be driven by USD weakness rather than by SGD strength, as the SGD NEER remains near the strong end of the MAS policy band'. The catalyst is a reassessment of Fed expectations — UBS expects incoming data to support 'an extended Fed pause rather than the 1-2 hikes currently priced by markets, causing stretched long-USD positioning to unwind'. This is the same Fed-pricing divergence from the rates theme, expressed in Asia FX. The AUD is the opposite setup: Westpac reports the July Australian CPI upside surprise has markets 'now fully pricing in a rate hike before the end of the year, up from only a 50% chance at the start of this week' — a hawkish R repricing that supports AUD rates but complicates the equity outlook. The two trades are complementary: short USDSGD is a Fed-pause expression; AUD is a hawkish-RBA expression — both consistent with a Fed that underdelivers relative to December-hike pricing.

Desk views

UBS CIO: bearish USDSGD to 1.28 (Dec 26) then 1.25 (Sep 27); downside is USD-weakness-driven; MAS band limits SGD appreciation scope.

Evidence IDs · P024

Westpac: July CPI surprise moved RBA pricing to fully priced for a hike before year-end — a hawkish domestic repricing supporting AUD.

Evidence IDs · P028

Investor implication

Express the Fed-pause view through Asian FX (short USDSGD) rather than outright dollar index; AUD offers a hawkish-central-bank offset.

Trade expression

Short USDSGD targeting 1.28 then 1.26; hold AUD rates exposure into RBA hike risk.

Invalidation

A hawkish Warsh speech validating December-hike pricing would unwind the USDSGD bear case.