Tariffs, debt and the dollar: is the United States rebuilding power — or spending what it still has?
This essay proposes that the current external economic policy of the United States is not a sequence of improvisations, but the partial execution of a published strategy — one that carries an internal contradiction known since 1960.
The argument has four parts. First, the plan exists and is public: Stephen Miran's paper of November 2024, written before its author went on to chair the Council of Economic Advisers and, from September 2025 to May 2026, serve as a Federal Reserve governor. Second, the plan attempts to resolve the Triffin dilemma through coordinated currency devaluation plus trade coercion — and the devaluation is the piece that popular readings of the plan usually miss. Third, there are macroeconomic constraints that no financial engineering can bypass: the savings-investment identity and the debt dynamics equation. Fourth, the August 2026 data allow a partial scoreboard, and it is asymmetric: the coercive instruments work; the economic instruments don't.
That asymmetry has a specific consequence, and it gives the conclusion its name.
On August 21, 2026, minutes before the deadline, Canadian Prime Minister Mark Carney suspended trade negotiations with Washington, called the negotiators back to Ottawa, and promised "dollar for dollar" retaliation. The next day, 50% U.S. tariffs took effect on a list of Canadian products that includes wine, cement and hockey sticks.
None of that is the important part of the episode. The important part is one sentence in the statement:
"We have recognised from the beginning that America has changed, and that we will not return to our old relationship."
Carney did not say Trump changed. He said America changed.
The difference between those two claims is the difference between a political cycle and a regime change. Governments negotiate expecting political cycles; companies and states reorganize in the face of regime changes. By choosing the second formulation, Ottawa publicly communicated that it had stopped waiting.
Consider what this means in practice, from the point of view of whoever makes the decision that actually matters.
A multinational must choose where to build a factory that will operate for 25 years. Ontario or Michigan. Today, Canadian production enters the U.S. market. But a new variable has entered the spreadsheet: what is the probability that, in 2029, 2031 or 2034, this specific product shows up in some annex of some proclamation?
The company does not need to believe that Canada will be permanently excluded. It only needs to conclude that producing inside the United States eliminates a risk that producing outside does not.
The effect does not show up in trade statistics. No factory closed, no layoffs, no headline. Just a plant in Michigan that could have been in Ontario, and that no one will ever count as a Canadian loss.
If U.S. tariff policy has one central economic objective, it probably is not revenue. It is the relocation of capital. And the mechanism of that relocation is not the tariff in force — it is the expectation of the next one.
In November 2024, right after the election, Stephen Miran published A User's Guide to Restructuring the Global Trading System. Miran was then a senior strategist at Hudson Bay Capital; he became chair of the Council of Economic Advisers in March 2025 and, from September 2025 to May 2026, served as a Federal Reserve governor — resigning to make room for Kevin Warsh.
The paper's argument, in five moves:
This does not mean that every decision of the administration follows a script. It means something more interesting: the hypothesis of using trade, currency, debt and security as components of a single architecture of power is not internet speculation — it is a document signed by the author who was then appointed to execute it.
And the implementation is verifiable. In January 2026, the Council on Foreign Relations assessed how closely the paper matches reality: tariffs reshaping trade relations, allies raising defense spending, cheaper energy from increased supply, a dollar down roughly 9% on a trade-weighted basis during 2025, and explicit efforts to contain the rise in long-term rates. The CFR rated three remaining proposals as likely for 2026: direct currency intervention, a Treasury–Fed partnership to cap yields, and taxation of foreign capital inflows.
On January 23, 2026, the Federal Reserve Bank of New York — under Treasury instruction — consulted currency desks about the dollar-yen rate. Historically, that kind of "rate check" signals imminent intervention. The dollar weakened immediately.
On August 1, 2026, the threat became action: coordinated currency intervention with Japan to support the yen. The plan's currency leg is no longer hypothetical.
The plan is leaving the paper.
Robert Triffin formulated the problem in 1960: the issuer of the world's reserve currency must supply the rest of the world with assets denominated in that currency, which requires running external deficits; but permanent external deficits end up eroding confidence in the currency itself.
Gourinchas and Rey updated the diagnosis for the contemporary era, shifting the axis from the "exorbitant privilege" — the right to issue the currency everyone wants — to the "exorbitant duty": the obligation to absorb the world's excess savings and to serve as the system's insurer of last resort.
In its 2026 version, the contradiction presents itself like this:
Here is the point that popular readings of the plan almost always get wrong: they read the strategy as a simultaneous expansion of dollar dominance and of the American industrial base. Miran says the opposite. For him, inelastic global demand for the dollar is the cause of the industrial problem — not the prize to be maximized.
That is why the paper needs the devaluation. Without it, the rest does not close. Tariffs alone tend to appreciate the currency of the country imposing them — because they reduce demand for foreign currency — undoing part of the very protectionist effect they sought.
Any version of the hypothesis that omits the currency devaluation is asking reality to deliver three mutually incompatible objectives at once. Miran's version includes it. It is the piece that makes the plan coherent — and simultaneously the piece that makes it risky, because it requires foreign holders to accept exchange-rate losses on assets they already hold, without running for the exit.
The plan does not resolve Triffin. It tries to manage Triffin with coercion.
There is a simplistic version of the critique that says: attracting foreign investment widens the trade deficit by accounting identity, therefore the objective is self-contradictory.
That formulation is imprecise. The identity holds for net flows, not for any specific gross inflow; flows run in both directions, and the current account includes services and income, not just goods. A foreign factory built in the U.S. can substitute imports or generate exports in the future.
But the real constraint exists, is deeper, and is harder to get around. It comes from the savings identity:
Reindustrializing means raising I. If S does not rise by the same amount, the current account worsens by definition — and the country needs more external financing, not less.
Now look at the composition of S. It includes public saving, which in the United States of 2026 is deeply negative: a federal deficit of $1.9 trillion, 5.8% of GDP, according to the CBO.
A country running a fiscal deficit of that magnitude while simultaneously wishing for a domestic productive investment boom is, by pure arithmetic, increasing its dependence on foreign savings. Tariffs do not change S − I. They change the composition and the prices of flows, not the balance. Only three things change the balance: fiscal consolidation, higher private saving, or lower investment.
None of the three is on the table.
There is also a second-order effect that rarely enters the discussion. The Japanese factory in Ohio remits profits to Japan. That is a permanent debit on the current account's primary income line. Trading a trade deficit for an income deficit is a legitimate trade — you buy productive capacity, jobs and a supplier chain today, and pay with an income stream tomorrow. But it is a trade, not a solution. And it shifts the problem to a line of the balance sheet that no one tracks politically.
On August 19, 2026, two things happened on the same day.
The U.S. Treasury announced it would double — "at least double" — the maximum size of its long-end bond buyback operations, from $2 billion to at least $4 billion per operation, in the 10-to-20-year and 20-to-30-year sectors, between September 9 and November 4.
And the U.S. public debt crossed $40 trillion for the first time.
The 30-year bond was trading at 5.2%, near two-decade highs. It fell nine basis points on the announcement and gave the move back the following session.
Before discussing what the buyback does, we need the equation against which any fiscal maneuver must be tested:
Where r is the average real cost of the debt, g is the real growth rate, and the primary surplus is the fiscal result before interest.
There are exactly three levers: grow faster, pay less interest, run a primary surplus. There is no fourth. Every proposal to "solve the debt" must pass this test, and the question is always the same: does it move r, g, or s?
The American numbers for 2026, from the CBO's February baseline:
| Variable | 2026 | 2036 (projected) |
|---|---|---|
| Total deficit | $1.9 T — 5.8% of GDP | $3.1 T — 6.7% of GDP |
| Primary deficit | 2.6% of GDP | 2.1% of GDP |
| Net interest | $1.04 T — 3.3% of GDP | $2.1 T — 4.6% of GDP |
| Debt held by the public | 101% of GDP | 120% of GDP |
| Interest as % of federal revenue | 18.6% | 25.8% |
The primary balance is in deficit today and stays in deficit across the entire ten-year projection. Net interest surpassed in 2026 the post-war record set in 1991.
And the warning that closes the picture: the Committee for a Responsible Federal Budget estimates that the average interest rate paid on the debt may exceed the growth rate starting in fiscal year 2031. When r > g on a sustained basis and the primary balance is negative, the debt grows even without a single dollar of new spending. That is the technical definition of a spiral, and it is arithmetic, not ideology.
There is a correct intuition behind the idea that the face value of the debt hides something important. It is just not about buybacks.
Robert Hall and Thomas Sargent built an entire literature on the distinction between the nominal and the market value of public debt. The implication, applied to the recent United States, is this:
The great devaluation has already happened.
Between 2020 and 2023, the combination of accumulated inflation and an abrupt rise in rates destroyed a substantial fraction of the real market value of long-dated U.S. debt. Thirty-year bonds issued with coupons of 1% to 2% lost between a third and half of their value. The holders paid the bill: pension funds, insurers, commercial banks — whose unrealized losses brought down regional institutions in 2023 — foreign central banks, and the Federal Reserve itself.
It was a wealth transfer from creditors to the sovereign debtor on a historic scale. It was not announced, not voted on, and never given the name of default. But it was the largest real reduction of the U.S. government's liabilities since the 1970s.
That is the real machine running in reverse. And it has already run.
What we are watching in 2026 is the consequence. The market did the math and started demanding the term premium back. The 30-year at 5.2% is the price of that memory. Buyers of long dollar duration learned, in a single cycle, what it means to carry uncompensated inflation risk — and they do not intend to repeat the experience for free.
A sovereign does not use the same trick twice in a row on the same audience. That is the technical reason why the "time bridge" — issuing short and expensive today while hoping to refinance long and cheap tomorrow — is harder now than it would have been in 2019. Not that the mechanism is impossible. It is that the premium demanded to cross the bridge rose precisely because the previous crossing left scars.
A Treasury issued in 2020 with a 1.25% coupon maturing in 2050 has a face value of $100 and may trade today at $66. If the Treasury issues $66 in short bills and uses the cash to buy back and cancel $100 of face, the nominal stock falls by $34.
This is not an accounting illusion. The face value genuinely falls, and since the American debt-to-GDP ratio is measured at face value, the indicator improves. That deserves to be stated clearly, because much of the criticism of this mechanism gets this point wrong.
First: the discount is not a discount. The bond is worth 66 precisely because it pays 1.25% in a 5% world. The price already incorporates the entire future cash flow discounted at current rates. Buying back at 66 by issuing 66 is neutral in present value by construction. There is no $34 of captured wealth — there is one obligation swapped for another of equivalent present value.
Second: the low coupon is an asset, not a liability. That 1.25% locked in until 2050 is the gift the market gave the U.S. Treasury in 2020. It is the exact mirror image of the losses that broke regional banks three years later. Buying back that bond and replacing it with bills at roughly 4% raises interest expense immediately. The Treasury is consuming its best balance-sheet asset to buy price movement at the long end of the curve.
Third, and decisive: the buyback touches none of the variables in the equation. It does not move r — in fact it raises current cost. It does not move g. It does not move the primary balance. It rearranges the composition of the liability.
Applied to the test from Section 5, the score is zero out of three.
If it is not debt reduction, what is it? Three things, all more interesting than the accounting reading.
Deliberate shortening of the average maturity. The buybacks are funded with bill issuance. The Treasury swaps long debt for short debt, cutting cost today at the price of higher rollover risk tomorrow. The CFR noted that the Treasury signaled a preference for short maturities in new issuance through the end of 2026. It is the sovereign version of financing a thirty-year mortgage with an overdraft renewed weekly: it works as long as the credit rolls over.
Fiscal dominance. The Treasury is performing a function that belonged to the central bank — managing the term structure of interest rates. Sargent and Wallace described the mechanism in 1981, in Some Unpleasant Monetarist Arithmetic: when the fiscal authority dominates, the monetary authority loses control of the price level in the long run. It is no coincidence that Miran explicitly proposes a Treasury–Fed partnership, invoking Operation Twist as precedent, nor that candidates to succeed Powell have publicly defended a new arrangement between the two institutions.
Political signaling. The most revealing detail of the August 19 operation: the Treasury had published its quarterly buyback calendar two weeks earlier and ran over it. The CFR rated the buybacks as more signal than substance. Mohamed El-Erian noted that the purchases are small both in absolute terms and relative to net issuance. And the chosen window — September 9 to November 4 — coincides with the final stretch of the midterm campaign.
The buyback is not a disguised deleveraging strategy. It is curve management with a political purpose, executed by an authority that has decided the long end is its problem.
The GENIUS Act, signed in 2025, created a federal framework for stablecoins with a requirement of liquid reserves in dollars and short Treasuries. The Treasury openly described the mechanism as broadening worldwide access to the dollar and reinforcing demand for American debt.
The mechanism is elegant. A person in Brazil, Nigeria or Argentina buys a thousand digital dollars believing they are simply holding dollars. Behind the token, the issuer holds T-bills. Millions of users scattered across the planet become, without knowing it, marginal financiers of the U.S. Treasury.
The numbers from August 2026:
Three qualifications, in ascending order of importance.
Scale. $141 billion against $40 trillion is 0.35%. Even the optimistic $2 trillion scenario would amount to a few percentage points of the total stock.
Substitution. A relevant share of the flow comes from money market funds and bank deposits that were already buying T-bills. The net addition to demand is smaller than the gross number suggests.
Curve mismatch — and this is the point that disarms the strategic argument. The American fiscal problem sits at the long end. Stablecoins buy exclusively the short end. The 30-year at 5.2% has nothing to do with Tether.
Worse: financing the state with liabilities redeemable in minutes, held by a global base with no prudential obligation and no deposit insurance, adds run risk to the sovereign financing structure. The BIS has published work on the effect of stablecoins on safe asset prices. The relevant question is not how much they buy in the expansion. It is what happens to the bill market in a disorderly redemption, and who is the lender of last resort in that scenario.
Combined with the deliberate shortening of the average maturity described in the previous section, the strategy produces an uncomfortable result: the Treasury is simultaneously increasing its dependence on short funding and the fragility of the base that provides that funding.
In March 2025, an executive order created the Strategic Bitcoin Reserve, directing that confiscated bitcoins not be sold and instructing Treasury and Commerce to develop budget neutral ways to acquire more.
The situation in mid-2026: approximately 328,372 BTC, about $25 billion. As of July 2026, no confirmed open-market purchase. Neither the BITCOIN Act — which proposed one million BTC — nor ARMA, which would codify the reserve with a twenty-year lockup, has become law.
$25 billion against $40 trillion is 0.06%.
The question "how much Bitcoin does the U.S. need to pay off the debt" is the wrong question, and the original text of this hypothesis was right to reframe it. A state does not need to own assets equal to its entire liability. It needs to own future revenues, productive capacity, assets, taxing credibility, natural resources, technology — and the ability to finance its liabilities at an acceptable cost.
Bitcoin, gold and critical minerals occupy small slices of the asset side. As long as the reserve depends on criminal forfeiture and an executive order not codified into law, it is political signaling and an optional position, not a strategic pillar.
Options have value. But option value does not pay coupons.
Here is the part of the hypothesis that survives scrutiny intact — and it is the most important one.
Henry Farrell and Abraham Newman named the phenomenon in 2019: weaponized interdependence. In networks with centralized topology, whoever controls the central nodes acquires two distinct powers — the panopticon effect, the ability to observe the flows, and the chokepoint effect, the ability to interrupt or make access more expensive.
The United States occupies the central nodes of almost every network that matters: invoicing currency, dollar clearing, the most liquid capital market on the planet, the richest consumer base, advanced semiconductors, cloud and artificial intelligence infrastructure, and the Western defense umbrella.
Power does not consist in preventing anyone from leaving. It consists in making leaving cost more than staying. That is lock-in, and it is real.
On February 20, 2026, the Supreme Court ruled 6–3 in Learning Resources, Inc. v. Trump that the International Emergency Economic Powers Act does not authorize the president to impose tariffs. The Court of International Trade ordered the return of approximately $165 billion to more than 330,000 importers.
It was the biggest legal defeat for American trade policy in decades. And tariff capacity did not shrink.
The administration immediately invoked Section 122 of the Trade Act of 1974 for a global 10% tariff, later raised to 15%. On July 20, 2026, three proclamations imposed 50% on Canadian goods using Section 338 of the Tariff Act of 1930 — a provision whose last invocation dates to the 1940s, which authorizes tariffs of up to 50% against a country that discriminates against American commerce, covers 554 tariff lines, applies even to goods that fully comply with USMCA rules, and carries no expiration date. Meanwhile, Section 232 — national security — has never been successfully challenged, and the steel tariffs have remained in force since 2018, spanning two administrations of opposite parties. Section 301 operates in parallel: Canadian products were already subject to a 10% forced-labor tariff since July 24, 2026.
The correct reading of this episode is not that the judiciary constrained the executive. It is the opposite: American tariff capacity is redundant by legislative design accumulated over a century. The constraint is not legal. It is political.
And that is exactly the kind of durability that the company from Section 1 — the one deciding between Ontario and Michigan for 25 years — must build into its spreadsheet.
In March 2026, Treasury International Capital data recorded net foreign selling of $138.4 billion in Treasuries in a single month: Japan $47.7 billion, China $41 billion, plus Luxembourg, Taiwan, Saudi Arabia, India, Canada and the Emirates. Swedish and Danish pension funds announced reductions or elimination of Treasuries from their portfolios. Central banks keep swapping dollars for gold.
Nobody is fleeing. The system has no ready substitute and the cost of exit remains prohibitive — the lock-in thesis is correct.
But prices form at the margin. And the margin is selling.
A hypothesis that produces no falsifiable predictions is not a hypothesis. These are the seven the theory implies, and the partial result as of August 2026.
1 — Growth in U.S. production and productive foreign investment, driven by trade policy.
FAILING.
Manufacturing construction spending: $172.7 billion annualized in June 2026, down 21.4% from $219.6 billion in June 2025. The peak was $235.6 billion in 2024, under the CHIPS Act and IRA — earlier legislation. Excluding electronics, the increase since the tariffs began was 5.6%, short of what would qualify as a boom. Roughly $1.6 trillion in announced commitments coexist with 82,000 fewer factory jobs.
Announcement is not poured concrete. And the lag between announcement and employment in a semiconductor fab is nine years, which means this prediction will only be fully judgeable in the 2030s.
2 — Relevant expansion of T-bill demand tied to stablecoins.
PARTIAL, STALLED.
$308 billion in August, below the $322.4 billion peak in May. Real and growing over the long horizon; irrelevant in scale and located on the wrong end of the curve.
3 — Increased buybacks and increasingly active management of off-the-run Treasuries.
CONFIRMED.
Doubled on August 19, with the Treasury running over its own quarterly calendar published two weeks earlier.
4 — Deliberate change in the maturity profile of new issuance.
FAILING — the direction was the opposite of what the original hypothesis assumed.
The average maturity is shortening, not lengthening. That raises rollover risk instead of locking in cost.
5 — A subsequent fall in the marginal cost of American financing.
FAILING.
Thirty-year at 5.2%, near two-decade highs, with a buyers' strike at the long end since June.
6 — Sustained improvement in the debt-to-GDP ratio or its perceived trajectory.
FAILING.
$40 trillion crossed on August 19, 2026. The CBO projects 101% → 120% of GDP by 2036; the CRFB estimates 131% under alternative scenarios. A primary deficit of 2.6% of GDP with no return to balance anywhere in the ten-year projection.
7 — Growing use of access to the American market as a foreign policy instrument.
CONFIRMED, without reservation.
This is the pillar that works.
Scoreboard: two confirmed, one ambiguous, four failing.
And the distribution is not random. The two confirmed are instruments of coercion and short-term management. The ones failing are precisely the real variables — productive investment, financing cost, fiscal trajectory, maturity profile — on which the virtuous circle proposed by the original hypothesis depended.
The plan is working where it involves imposing costs on others. It is not working where it involves building its own capacity.
No country in the world is more integrated with the American economy. Automotive supply chains cross the border multiple times before the final product. Energy, potash, aluminum and critical minerals flow in volumes with no immediate substitute. And there is a trade agreement in force, renegotiated by this very administration in its first term.
That is why Section 338 matters more than its $20 billion of coverage suggests: it applies to goods that fully comply with the USMCA.
The implicit message is that treaty compliance does not guarantee access. And if treaty compliance does not guarantee access for the most integrated partner, the inference for everyone else is direct.
Washington is communicating that access to the American market is not a contractual right. It is a strategic asset — granted, and revocable.
That settles the factory decision from Section 1 — but it settles it in both directions. It lowers the risk of producing inside the U.S. and raises the risk of producing outside. Simultaneously, however, it teaches every boardroom on the planet that American trade contracts have conditional value. Companies that learn this do not only move to Michigan: they also move toward redundancy, diversification and political hedging.
Lock-in produces dependence in the short run and incentives to diversify in the long run. The empirical question is which effect dominates, and on what horizon.
Fiscal space is power. That remains true, and it is the most valuable intuition in this entire hypothesis. A state with fiscal space can finance a war, rescue the banking system, subsidize semiconductors, bankroll an artificial intelligence race, absorb a recession — or simply wait six months longer than its adversary in a negotiation. In an international dispute, whoever can wait longer holds the structural advantage.
But fiscal space is not produced by accounting engineering. It is produced in three ways: growth, primary surplus, or inflation. Bond buybacks are not a fourth way. Stablecoins are not a fourth way. Bitcoin is not a fourth way.
And this is where the scoreboard's asymmetry stops being a statistical curiosity and becomes a mechanism.
When a strategy's economic legs fail and its coercive leg works, the state does not abandon the strategy. It increases the dose of the part that responds to command.
That is exactly the 2026 pattern. The factories are not coming back — but tariffs escalated from a global 10% to sectoral 50% under a 1930 statute that no president had dared use in decades. Yields did not fall — but the Treasury began intervening directly at the long end of the curve, running over its own calendar, in an election year. The debt did not stabilize — but the average maturity is shortening to push the bill forward.
This does not prove the strategy will fail. It proves that it is being rebalanced toward the instrument that is cheapest to operate and most expensive to sustain.
A state that can impose costs but cannot convert them into productive capacity or budget room becomes progressively dependent on imposing costs. It is the only lever that still responds. And every pull consumes a little of the confidence that made the lever effective.
For years, players sat down at the table believing the American was weakening. The debt was growing. The deficits were widening. There was talk of de-dollarization. China and the BRICS were expanding. Long Treasuries were losing value. The conclusion seemed obvious: at some point he would run out of chips.
Then comes the discomfort. The chips are denominated in his currency. The credit runs through his infrastructure. The most liquid assets are issued by him. Companies want access to his consumers. The compute that trains the models is his. And a good part of the world's capital is already invested inside him.
A player may conclude the system is unfavorable and decide to leave. Fine. But leaving requires selling assets, swapping reserves, building alternative payment systems, reorganizing supply chains, finding substitute markets, and accepting losses in the transition.
Discovering the problem is easy. Leaving the table costs money too.
And here is what the original version of this hypothesis did not incorporate, and what the 2026 scoreboard forces us to incorporate: owning the table and winning are different things.
The American player owns the casino. That is established and not in dispute. But in August 2026 that same player pays 5.2% to finance thirty years, has crossed $40 trillion in liabilities, runs a primary deficit of 2.6% of GDP with no reversal plan, returned $165 billion for having used the wrong statute, watches factory construction spending fall 21% a year while announcing the greatest reindustrialization wave in history, saw Japan and China sell $88 billion of its debt in a single month, and is shortening the average maturity of its own liabilities to avoid paying the long end's price.
Owning the table guarantees the right to keep playing. It does not hand out chips.
That is why the question that defines the next decade is probably not who loses more from the 50% tariff, nor whether the world will discover that the United States remains at the center of the system. It does, and that is not seriously in dispute.
The August 2026 scoreboard does not answer definitively. But it leans.
And there is a worse discovery at a poker table than realizing your opponent is bluffing. It is realizing that you calculated his cards correctly, calculated correctly that he owns the casino — and still got it wrong. Because the missing question was not who owns the house.
It was how long the house has been spending more than it takes in.
© 2026 Marco Barreto. All rights reserved.