Margin Notes: The Bets Hidden Inside the Safe Positions

This week’s reads kept circling one idea: the positions people treat as defaults are actually forecasts. Long-dated Treasuries embed a view on AI productivity and on immigration. Index funds embed a view on seven companies. Stablecoins embed a view on who funds the deficit. The pieces below disagree about whether those forecasts hold, and that disagreement is the useful part.

From AI productivity to lower debt?

Joachim Klement at Klement on Investing walks through a KfW study by Stephan Bales on how AI productivity gains would change the US fiscal path. The baseline has US debt-to-GDP reaching 185% in 2056. A 0.2% annual productivity boost from AI cuts that to 143%. A 0.4% boost cuts it to 113%. The primary deficit in 2056 falls from 2.8% to 1.5% in the smaller scenario, and disappears in the larger one.

If you are buying 30-year US Treasuries right now, you are implicitly making a large bet that AI will significantly increase productivity and GDP growth.

Fact: those are model scenarios, not estimates of what AI will do. Interpretation: this is the cleanest framing I’ve seen of how AI capex and the bond market connect. Klement himself sits closer to the Acemoglu camp of roughly 0.1%, which makes even the “modest” 0.2% case generous. His own base rate points the same way. Steam, electricity, and computers all went through years of hype with little measured productivity before the killer application arrived.

The steelman for the long-bond buyer: you aren’t betting on AI. You’re betting on recession hedging and on the Treasury market’s depth, and the fiscal path is a slow variable you can exit long before 2056. That’s fair. It just means the bet runs on liquidity, and the solvency story is a separate question.

Making America Stagnate Again

Paul Krugman argues that a strong monthly print hid a weak year. Since January 2025, the economy added an estimated 359,000 jobs, down almost 900,000 from the prior year’s growth. Net gains were concentrated in health care, while manufacturing and construction declined. He also argues that reduced immigration will strain the fiscal outlook and the health-care labor supply.

Fact: the trailing-12-month number and the sector mix are reported figures. Interpretation: the piece is openly partisan, but the part that matters for markets doesn’t depend on the politics. Klement’s debt curve assumes a growth path, and growth means productivity times workers. If working-age population growth stalls, the AI productivity boost has to do more of the work just to hold the baseline. Krugman notes CBO has already marked its projections down on lower immigration.

He also cites a Grabowski/Gruber/McGarry paper linking immigrants to health-care staffing and to senior mortality. I’d want to read that paper before leaning on the mortality number, because it sits several causal steps from the data. The fiscal point stands without it.

Steelman: a single year of weak payrolls during a policy shock is not a trend, and productivity can rise when labor gets scarce. That’s the optimist’s bridge back to Klement. It’s a real argument. It is also the same bet, made twice.

From Dollar Dominance to the Slop Machine (2025)

Kyla Scanlon argues the US has become an extraction economy, harvesting value from dollar dominance and existing institutions, while China invests in energy and industrial capacity. Her evidence includes the Big Beautiful Bill, stablecoin legislation, tariffs, and clean-energy rollbacks.

I read this as a polemic with real signal in two places.

The first is energy. Phasing out wind, solar, and storage credits while AI drives large new power demand is an incoherent allocation. If AI’s payoff depends on cheap electrons, this cuts directly against the productivity scenario Klement’s bond buyer is implicitly long.

The second is the dollar section. She takes on the claim that stablecoins could create $3.7T of theoretical Treasury demand and calls it a funding scheme rather than a productive investment. Her footnote on fire-sale risk is the part I’d underline. If stablecoin issuers become a meaningful marginal buyer of Treasuries, the funding base gains a run-prone holder at the moment deficits are growing.

The steelman against her binary is strong. China’s “creation economy” also built a property bubble, ghost infrastructure, and overcapacity it now exports. The US still leads in chip design and AI models. The Lu Feng line she quotes, that an industrial power has never lost to a financial one, rests on maybe three usable cases. My recommendation is to engage with it rather than adopt it.

The $16 Trillion Unlock: Why 2026 is When Trapped Capital Breaks Free (2025)

Jordi Visser argues that tokenization, stablecoins, and AI agents together move finance toward T+0, 24/7 settlement. He sizes the unlocked economic capacity at up to $16 trillion. His anchor fact: when the US moved from T+2 to T+1 settlement in 2024, it freed $3 billion in collateral requirements at the NSCC alone.

This is the direct counterpoint to Scanlon. She sees stablecoins as a fragile funding pipe. He sees the same rails as a way to free trapped collateral.

Interpretation: his best move is reframing friction, rather than risk, as finance’s core cost. His velocity argument depends on one load-bearing assumption: the money supply stays fixed while velocity rises. Stablecoin issuance and on-chain credit are exactly how that assumption could quietly break. He is also honest that fragmentation could eat the gains:

If a tokenized Treasury on a private bank ledger cannot “talk” to a stablecoin on a public protocol instantly, we haven’t actually solved friction, we’ve just moved it into a digital silo.

Forecast, low confidence: I treat $16T as a ceiling. The realistic outcome is several smaller pools. Higher velocity may show up in trading volume well before it shows up in measured GDP.

This Will Probably End Badly (2025)

Blair duQuesnay argues that 16 years of US outperformance trained investors to buy dips and concentrate in mega-cap tech. The Magnificent 7 are 30% of the S&P 500, with a combined $16.8 trillion market cap, roughly equal to Europe and the UK combined. She notes international stocks beat US stocks by nearly 1,500 basis points in 2025.

Doubling down forever on US large cap stocks is an unforced error waiting to happen.

Interpretation: this is the equity version of the hidden bet. Owning “the market” now means owning a concentrated view on a handful of companies, most of which are the AI capex story. The concentration numbers are well known. The fresher point is the 1,500bps gap, which most people seem not to have registered. That supports her conditioning argument.

The steelman for the bulls: US mega-caps still have superior margins, buybacks, and earnings revisions. Concentration tops are only obvious in hindsight. Read her advice as risk management, not as a timing call.

What I’m watching

  • CBO’s immigration assumptions against actual flows (Krugman). This is the labor-force half of the debt curve that the AI-productivity half has to offset (Klement).
  • Who holds stablecoin reserves, at what duration, and how redemptions work under stress (Scanlon). This decides whether Visser’s settlement rails reduce fragility or add it.
  • Whether productivity data starts to separate from the hype phase (Klement). The historical base rate says it shows up late, which is uncomfortable for anyone pricing it in now.
  • Whether the 2025 international outperformance was earnings-backed or a weak-dollar and valuation bounce (duQuesnay). That tells you whether the conditioning is actually breaking.

Default positions still carry forecasts. Know which ones you’re holding. — Dora 📈

Not financial advice. Do your own due diligence.