Contrarian Codex — Newsletter #123: A Rough Patch, What's Next?

The rollover wall — $8 trillion repricing higher

• Roughly $8tn of Treasuries roll off over the next 12 months. The average coupon on that stack is ~3.3% and the one-year bill it rerolls into sits closer to 4%, so refinancing the lot bolts ~$50bn of fresh annual interest onto a budget already carrying a ~$2tn deficit.
• That $50bn compounds: every quarter more zero-rate-decade cheap paper matures and reprices, so the weighted-average rate on the whole federal stack grinds higher for years even if the Fed never touches policy again. Interest expense already runs ~$1.2tn, ~23% of federal receipts (chart credited to Lukas Ekwueme).

Why the bond market can't freely price 8% — capping the long end

• "Volcker could run 20% money-market rates because the prior decade had already inflated post-war debt from ~100% of GDP down to ~30%. Inflate first, hike second. We are standing at the wrong end of that sequence." At today's >100% ratio, the cure feeds the disease straight through the interest channel.
• Treasuries at 8% with $8tn rerolling annually is "a market that seizes," dragging the whole collateral base down — an outcome Washington cannot choose. So you cap the long end, openly or through the back door, and let the currency absorb the strain. Term premium feeds straight back into rollover cost, and once that loop turns the Fed gets dragged back to the desk to keep the UST market functioning.

The release valve — repression and Waller's inflation "range"

• Hold nominal rates under inflation, lean on some form of yield-curve control, and let the dollar do the adjusting — "financial repression is the courteous name for it." For risk assets that setup is a nominal tailwind: capital is pushed out of cash and government paper into equities, gold, hard assets and real cash flows.
• "Right on cue," Waller floated making the inflation objective a range rather than a fixed 2%. A point target turns every 3-handle print into a visible failure; widen the band and a 3% overshoot becomes "a polite miss." Warsh muttering about frameworks means it isn't one governor freelancing — the goalposts are being fitted with wheels before the political moment to formalize them arrives.

Fiscal dominance — the hike is the stimulus

• Under fiscal dominance the size of the debt decides monetary policy, not the other way around. The 1970s lesson (higher rates kill inflation) held because the marginal dollar was privately borrowed and answered to the price of money.
• Now the marginal dollar comes off the fiscal side — deficit spending — which doesn't answer to the funds rate. Worse, hiking into a >$30tn stack pays hundreds of billions of interest income into private hands: "the hike is the stimulus." So the tightening reads as a phase, not a destination; fighting inflation with hikes inside this regime just reroutes it into the dollar.

The industrial capex handoff — from chip/EV fabs to AI compute & the grid

• Manufacturing construction spending is down ~22% YoY — but that's the rear-view mirror, the tail of the CHIPS/battery-plant boom that crested in 2024, still running at >2x the pre-2021 baseline. CEO capex intentions sit near an 18-month high (~37% planning to lift spending) and the ISM manufacturing gauge is back above 53 for six straight months.
• The read is a handoff, not the end of the cycle: the new leg points at AI compute, the electrical grid, data centers and reshored defense — which run far more copper, electrical steel and transformer per dollar than a chip fab. Early-cycle, forward-looking industrial capex slides a floor under commodity demand right as the disinflation crowd uncorks the champagne.

Soft payrolls into a stagflation-lite box

• June payrolls landed at 57,000 vs ~115,000 consensus, with April/May revised down a combined 74,000. Unemployment fell to 4.2% — but only because ~three-quarters of a million people left the labor force; participation sank to 61.5%, a five-year low. "A lower unemployment number built on people giving up is not a certificate of things going well."
• Layer that onto >3% inflation and 3.5% wages and you land in stagflation-lite: a cooling jobs market is the growth-side excuse to keep real rates pinned below inflation, arriving just as Waller hands them the inflation-side excuse. White-collar fingerprints (information payrolls down, big tech trimming to fund compute) tie back to the capex handoff.

Seasonality — the midterm-year drawdown math

• The S&P closed Q2 up ~15%, the 5th-best Q2 since 1950 and the strongest midterm-year Q2 ever. Nine prior double-digit Q2s finished the back half higher 8 of 9 times, averaging +11.7% — which off current levels lands near 8,400.
• Midterm peak-to-trough drawdowns since 1950 average 17.3%, usually bottoming by mid-August; this year took ~9% early (Q1 Hormuz/oil spike). Either the pain was pulled forward or it's deferred into Aug–Oct. Consensus 2026/27 EPS growth of ~24%/17% on a 22x forward multiple is "the whole trade"; sell-side year-end target ~7,850 sits ~7% below what seasonality implies — "somebody is wrong here."

Geopolitics — Hormuz throughput is the number he watches

• Strikes have widened to critical infrastructure (Bushehr perimeter, railway bridges, Chabahar/Bandar Abbas/Jask), ~170 US strikes in two days, with Jordan now pulled in alongside Bahrain, Kuwait and Qatar.
• Tracking shows ~13 vessels through Hormuz in 24 hours (vs ~110/day pre-war, 25–40/day last week) — and with transponders off and GPS spoofing back, the true figure is unknowable. War-risk cover runs 2–6% of hull value per transit. Oil hasn't reacted much yet, "but something resembling a cooling down needs to arrive soon, or price will react."

Uranium — sentiment capitulates while the term price prints a record

• Mart's sentiment gauge fell ~6 points to 19 (pessimism) over two weeks while URNM lost just ~1.5% — a big sentiment move on a price that barely budged tells you people are "giving up," not selling. Emotional capitulation shows up in inboxes while price chops sideways. "If you are not yet invested in the sector or looking to scale in more, I would argue now is a good time to start."
• The TradeTech long-term indicator printed $97/lb, a nominal all-time high ($2 above the 2007 peak, though inflation-adjusted still below it). It's a grind, not a melt-up: just 21 transactions / ~32m lbs YTD in the term market, sellers now disciplined ("patience pays"), and the buy-side tell is utilities circling back to offers they rejected 6–12 months ago. Spot ran 26 transactions in the mid-$80s. Either term drops (confirming equity weakness) or equities rise to meet term — he thinks the latter is far more likely.

The AP1000 program and $17.5bn of long-lead financing

• The DOE's conditional commitment of up to $17.5bn (Office of Energy Dominance Financing) funds long-lead items — heavy forgings, pressure vessels, turbines — for up to five sites of two Westinghouse AP1000 units each (the "ten reactors by 2030" goal). Westinghouse has LOIs with seven partners chasing five slots; each partner must post $500m equity first.
• Mart is enthusiastic but won't oversell it: a conditional commitment isn't a build contract, ten AP1000s could run toward $200bn all-in, and Vogtle came in late and over budget. Still, every unit built needs fuel for 60–80 years, and "utilities do not start reactors on spot purchases."

SMRs — the fuel filter and why the BWRX-300 leads

• Mart's first screen for which SMRs run in the early 2030s is brutally simple: does it run on fuel you can buy today (low-enriched uranium, <5% U-235) or on HALEU (5–20%), which "barely exists as a commercial product outside Russia and China." Most hyperscaler-backed marquee designs — Kairos (Google), X-energy (Amazon), TerraPower/Oklo (Meta, NVIDIA) — are Gen-IV designs needing HALEU and carrying first-of-a-kind fuel risk.
• GE Vernova Hitachi's BWRX-300 comes out ahead: a ~300MW boiling-water reactor on conventional LEU, already under construction at OPG's Darlington (first Western SMR, ~end of decade), with a pipeline (TVA Clinch River, Orlen Synthos in Poland, Fortum, Samsung C&T). GE Vernova's gas turbines are sold out to 2029 and it's in daily contact with hyperscalers — "the design furthest along in construction has drawn less hyperscaler headline flow," a gap Mart expects to close in its favor. NuScale's cancelled Utah CFPP is the cautionary first-of-a-kind cost blowup. (Rolls-Royce SMR — Qatar SWF / Constellation / CEZ backed — "remains the one that got away for the portfolio.")

Who owns the demand — hyperscalers reach upstream toward the pounds

• Three buyers converge on the same firm-power problem: utilities (the AP1000 program is a utility program with federal financing bolted on), hyperscalers (announced data-center nuclear deals now >9GW across the four biggest tech names — Three Mile Island restart, Susquehanna, Meta's long-dated commitments, Google/Kairos), and sovereign wealth / infrastructure capital (Qatar inside Rolls-Royce SMR).
• The read-through Mart finds most interesting: hyperscalers don't stop at buying power — they move upstream and finance the fuel cycle to lock in supply. NexGen has said it's in talks with data-center providers to help finance Rook I, and Leigh Curyer draws the automaker-lithium analogy. The market treats a credible SMR ramp as a tail scenario "when the buyers with the deepest pockets in the world are behaving as though it is the base case."

SMR range, Canada's strategy and Sizewell B — the demand that's contracted years early

• The 2035 installed-SMR range (~10–25GW, ~1–3% of world nuclear) isn't market-moving by itself; the dispersion by 2050 is — IEA base ~40GW, up to ~120GW with policy support, ~190GW at cost parity, NEA sketches to ~375GW. "The difference between SMRs being a footnote and a second nuclear industry." Mart's read: back-loaded and lumpy, inflecting early-to-mid 2030s. A big-oil/miner consortium (Exxon, Shell, Chevron, Conoco, Rio Tinto, Freeport, Nucor) exploring nuclear process heat would have been unthinkable a few years ago.
• Canada's late-June Nuclear Energy Strategy is a blueprint for up to 10 large reactors over 15 years (two building by 2035) plus the four BWRX-300s at Darlington — but no money attached until an April 2027 financing policy. Sizewell B's life extension to 2055 adds 20 years and ~9–10m lbs of demand. All of it contracts fuel years ahead of first power.

Oil — the surplus is a mirage; supply loss is not an inventory draw

• Jeffrey Currie's point: the crude stranded behind the Strait has been pushed to market over 3–4 weeks, and that release (not a loosening balance) rolled the front of the curve over. The selloff "has run too far." But Mart says that's only half the story.
• The "world lost ~1.2–1.4bn barrels" math (12–14 mb/d lost for ~100 days) doesn't hold: the world entered the war with 8.2bn barrels, its highest cushion since early 2021. Kepler Cheuvreux shows the IEA had the market in deficit only 5.6/3.3 mb/d in Mar/Apr with observed inventories down only ~250m barrels. The shortfall got absorbed — demand destruction (~3–4 mb/d, 1.6 in China), unwinding ~3 mb/d of oversupply, Gulf reroutes, and Kepler's "dark transits" (~90m barrels of non-Iranian crude hugging the Oman coast). This is why he didn't go 50/50 oil in the portfolio like some analysts.

Oil — thin buffers, a starved product complex, and China's swing

• None of that makes the market loose: buffers are thin and thinning (OECD cover toward the fewest days since the early 2000s), and the real tightness migrated into refined product — middle-distillate cracks at record highs. When a refiner buys soft crude and sells diesel into a starved market at a fat crack, it runs harder and pulls crude back out of the surplus — "how you get violent snapbacks." Contrarian on both sides: bulls oversold the inventory story; anyone reading the drop as comfort is about to make the mistake in reverse. The IEA's first 2027 look shows a real overhang building.
• China back-of-napkin: ~3 mb/d coming out of storage at the current pace, ~250–300m barrels drawn since late February — on its own ~half the global two-month draw. A restock would be ~1 mb/d of extra pull, a "soft floor" like the US SPR refill of 2023–24. Separately, China lifting refined-fuel export limits and letting Zhejiang Petrochemical (Rongsheng) resume shipments reads as confidence returning early: "bullish crude, bearish cracks" — the factories must buy more oil to make the fuel they're finally allowed to export.

Precious metals — central banks a floor, not a fuse

• Gold/silver back above $4,000 / $60 but volatile — he wouldn't be surprised by one more dip, and remains very bullish, having started buying. A net 30% of central banks (OMFIF) plan to add over 1–2 years and 82% now hold physical gold (up from 71%); the top motive is protection against a weaponizable financial system (51%, +11 points since 2024).
• The central-bank bid is "a slow clock — a floor under the market rather than a fuse under the price": strategic, funded from reserve rotation, accumulating into weakness. Rates and the dollar set the marginal price day to day and both point the wrong way near term — but 61% of managers see $5,000–6,000 by mid-2027, and some reports suggest official buying runs ~15x the reported figure.

Copper, cannabis & tin — noise, not new information

• Copper chopped with the complex ($6.00–6.15/lb COMEX) on a firmer dollar, not its own balance. The one substantive item: BHP's first environmental approval for the Escondida expansion (~$1.3bn early works, part of a $10.7–14.7bn Chilean program). Mart cautions against reading it as new supply — grades are falling and the capital mostly holds output flat; Chilean production fell 5.8% YoY in Q1.
• Cannabis: the DEA rescheduling hearing (deadline 15 July) is oddly structured — all seven designated parties oppose rescheduling, no reform side admitted; only adult-use/unlicensed marijuana's Schedule-I status is in play, and 280E still bites MSOs. Tin fell to ~$50,550/t on an AI-infrastructure sentiment repricing, not a physical change (Indonesian licences, Man Maw still constrained). "None of that is new information."