Debt, Debt and Debt

On August 18th, the Treasury’s daily statement printed a number with fourteen digits in front of the decimal: $40,047,425,768,420.

We crossed $38 trillion in the fall of 2025. $39 trillion in March. $40 trillion five months later. The first trillion took roughly two hundred years. The last one took a hundred and fifty days. And the debt limit set by last year’s reconciliation bill sits at $41.1 trillion — about six months away at the current run rate.

Four weeks earlier, Alphabet — a company that has printed money for two decades without interruption — reported its first negative free cash flow quarter since the 2004 IPO. Not because the business broke. Because it spent $44.9 billion on capex in ninety days.

And in June, American households pushed revolving credit to $1.351 trillion — roughly 24% above the pre-pandemic peak, and within about a billion dollars of its all-time high — while the savings rate sat around 4%.

Three balance sheets. Three extremes. Three completely different diseases.

That last part is what this letter is about, because “debt” gets discussed in this country as if it were one problem with one prognosis. It isn’t. The government’s debt is worse than consensus believes. Corporate debt is less bad than consensus believes. Household debt is neither — it’s split, which is harder to trade than either.

And all three of them are being priced into the same organ at the same moment: the long end of the bond market, in the weakest ten weeks of the four-year political cycle, with two shooting wars running and a trade fight that just opened a second front against our closest ally.

Periodically I will provide our thoughts on how we view the global markets and the macro backdrop.

Part I — The Government

A Math Problem with Delusional Politicians on Both Sides

Where we actually are

• Gross federal debt: $40.05 trillion, roughly $117,000 per American, growing about $7 billion a day.
• FY2026 deficit: $1.9 trillion, 5.8% of GDP, with unemployment at 4.1%. That combination is historically abnormal. You are supposed to run deficits like this in a war or a depression, not at full employment. (CBO has since revised FY2026 to roughly $2.1 trillion after the Supreme Court struck down a set of tariffs and forced the refunds.)
• Net interest: $970 billion in FY2025, crossing $1 trillion this year. Interest is now the third-largest line in the federal budget.

The growth argument, and why it fails

The official answer, in both parties, reduces to one sentence: we’ll grow our way out of it. Hope is not a strategy. WTF!

Let me show you why that isn’t a plan. It’s an absurd wish, and the arithmetic is not close.

Debt-to-GDP is governed by two things: the gap between what you pay on the debt (r) and how fast the economy grows in nominal terms (g), and the primary deficit — what you overspend before interest.

• Interest paid divided by debt held by the public gives an effective rate of roughly 3.2%.
• New Treasury paper is being issued at 4.3% to 4.75%. The 10-year sits near 4.65%, having backed off a 20-month high of 4.75% on August 21st; the 30-year is above 5.20%, its highest in nearly twenty years.
• CBO has real growth at 2.4% this year decaying to 1.8% thereafter. Add a ~2% deflator: nominal growth around 4%.
• The primary deficit is 2.6% of GDP.

Read those together. The effective rate is below nominal growth today only because a large slice of the debt still carries ZIRP-era coupons. As that paper matures into 4.5% paper, the average cost climbs toward the market rate. r converges up to g. It does not converge down. That process is mechanical and it is already underway.

When r equals g, the primary deficit passes straight through to the debt ratio, dollar for dollar. Call it 2.6 points of debt-to-GDP added every year, forever, with no recession, no new war, no crisis.

To genuinely grow out of it — stabilize the ratio through the denominator while holding the primary deficit where it is — you’d need nominal growth to beat the effective rate by roughly 2.6 percentage points on a ~100% debt load. Roughly 6.5% nominal against a 4% cost of funds. That’s about 4.5% real growth, sustained, for a decade.

The United States has never sustained that over ten years in the postwar era. CBO’s own commentary notes that even 3% has been historically rare and hasn’t been sustained in forty years. Q2 real GDP came in at 1.5%.

So the growth thesis isn’t optimistic. It’s innumerate. It requires an economic miracle roughly two and a half times the size of the one being promised — and it requires that miracle while AI capex is simultaneously being sold to us as the thing that will deliver it. Each party is equally responsible for this debacle.

Where the baseline goes, don’t shoot the Messenger

	FY2026	FY2036
Deficit	$1.9T (5.8% GDP)	$3.1T (6.7% GDP)
Net interest	$1.0T (3.3% GDP)	$2.1T (4.6% GDP)
Interest / federal revenue	18.6%	25.8%
Debt held by public / GDP	101%	120%

CBO Budget and Economic Outlook, February 2026 baseline.

Ten-year cumulative deficits: $23.1 trillion over 2026–2035 — $24.4 trillion if you slide the window to 2027–2036. Ten-year cumulative interest: $16.2 trillion. Extend the same baseline twenty more years and public debt reaches 175% of GDP. CRFB’s projection is the one that stays with me: interest overtakes Social Security as the single largest federal program around 2047.

One in four tax dollars buying nothing. Not a road, not a carrier, not a benefit check. Rent on the past. Think about this!

Gold is the most interesting data point of the year

Gold set a record of $5,597 on January 29th, then had its worst quarter since 2013 — down about 14% in Q2 as the Iran conflict pushed oil and inflation expectations higher and the market repriced the Fed hawkish under Warsh. It broke below $4,000 on June 24th for the first time since November 2025 and closed the quarter at $4,008. It has since clawed back to roughly $4,650, a three-month high.

Now look at what central banks did during that drawdown. Per the World Gold Council, official-sector buying in Q2 hit 288.9 tonnes — up 62% year over year, the strongest second quarter in the data series. In a quarter gold fell 14%. Poland added 51 tonnes against a 700-tonne target, taking reserves to 632 tonnes. The PBoC added 33 tonnes, its biggest quarterly addition since late 2023, extending a streak running nearly two years. A record 45% of surveyed central banks expect to add more over the next twelve months. Ray Dalio believes one should hold 10–15% of a portfolio in gold as a strategic holding, he said just a couple of days ago.

The honest complication, because I’d rather put it in front of you than have you find it. First-half official buying was only 345 tonnes, the weakest half-year since 2022 — and the reason is not that the buying stopped. In July, Metals Focus issued an erratum revising Q1 official-sector purchases from 244 tonnes down to 57 tonnes, reclassifying the missing 187 tonnes as OTC and other demand. The metal was still bought. It can no longer be attributed to central banks. Separately, Russia and Turkey were net sellers covering budget shortfalls, Russia alone for 43.5 tonnes in H1.

That distinction matters more than the headline. My thesis is that a large, price-insensitive buyer is absorbing gold into weakness. That thesis is intact — the Q2 print is real and the OTC bid did not evaporate, it got relabelled. What I can no longer say with the same confidence is that the buyer is sovereign. If you see a letter this quarter quoting 533 tonnes for the first half, the author added the pre-revision Q1 to the post-revision Q2. I nearly did the same.

Part II — The Corporations

A Timing Problem Challenged as a Solvency Problem

The story the market told itself this summer: Big Tech went from self-funding to debt-funding, free cash flow collapsed, and we’re watching the setup for a credit event.

The numbers behind it are real, and I won’t soften them:

• 2026 hyperscaler capex estimates run from roughly $700 billion (J.P. Morgan) to $860 billion (BofA). Goldman’s number is $750 billion this year, nearing $1.2 trillion in 2027.
• Hyperscaler bond issuance went from about $108 billion globally in 2025 to roughly $220 billion so far in 2026. Goldman estimates about a third of capex is debt-financed this year, rising toward 35% in 2027. Total global AI-related debt issuance is tracking near $570 billion for the year.
• Free cash flow cracked in Q2. Alphabet: −$5.9 billion, first negative quarter since the 2004 IPO, on $44.9 billion of capex, with full-year guidance lifted to $195–205 billion. Amazon: trailing-twelve-month FCF −$7.6 billion, with 2026 capex just raised again to about $220 billion. Meta: +$784 million, barely. Microsoft: +$19.6 billion — the only one generating real cash.
• Spreads noticed. Median 2–4 year paper from Amazon, Alphabet, Meta and Oracle widened to about 40bp over Treasuries from 30bp last year; 20-year-plus paper is out near 118bp from 108.5bp. Of 91 hyperscaler bonds issued in 2026 with comparable pricing, 78 were trading wider than issue by late July.
• Demand is thinning. Order books covered hyperscaler deals nearly five times over in February. By July, under two. Amazon had to pay up 18–21bp on the long end of a surprise $25 billion deal.
• And a large share isn’t even on balance sheet — Moody’s has flagged roughly $662 billion of data-center lease obligations sitting off-balance-sheet under GAAP. Amazon’s own 10-Q discloses $650 billion of total contractual commitments, about $286 billion of which appears nowhere on the balance sheet.

That’s the bear case, stated fairly. Here’s why I think it’s being over-extrapolated.

The number nobody is looking at imo

Everyone is watching free cash flow. Free cash flow is a difference. It tells you about the timing of one line item, not about the health of the business.

Look at the two components separately. Across the group, operating cash flow is compounding at roughly 23% a year. Cash capex is compounding at roughly 70%. Those curves crossed this quarter. That’s the entire story of “negative FCF.” The engine didn’t weaken. The spending accelerated past it.

Amazon is the cleanest illustration in the whole complex: TTM free cash flow swung to −$7.6 billion — while TTM operating cash flow rose 33% to $161.4 billion. The company attributed the swing entirely to capex, not to any deterioration underneath. Alphabet is the same shape: operating cash flow up 41% year over year to $39.1 billion, capex exactly double at $44.9 billion.

That distinction is everything. A company that stops generating cash is in trouble. A company generating record cash and choosing to spend all of it plus borrowed money is making a capital allocation decision — and capital allocation decisions are reversible. We own and really like $AMZN.

The 2028 math matters here

Run it forward. If operating cash flow compounds at even 20% — below the fitted 23% — group cash generation is roughly 1.4x today’s level by 2028 and 1.7x by 2029. Capex, meanwhile, cannot compound at 70% for long; the physical world won’t permit it. Transformer lead times are past two years. Grid interconnects, turbines, substations, skilled electricians — the binding constraints are increasingly industrial, not financial.

The moment capex growth merely flattens — not falls, flattens — free cash flow doesn’t recover gradually. It snaps, because two years of 20%+ compounding on the inflow side meets a flat outflow. That’s the spigot. And they control the valve: Microsoft has guided to remaining FCF-positive through fiscal 2027 while its capex ran up roughly 70% year over year. That’s an existence proof that the throttle works.

Microsoft, Alphabet, Amazon and Meta all entered this cycle under 1x debt-to-EBITDA. This is not 1999. The borrowers then were pre-revenue. Moody’s puts it plainly — most of this investment is carried by companies with enormously profitable existing businesses that aren’t going anywhere.

Two things I’ll concede. Group leverage has gone from roughly 0.9x to 1.8x in two quarters — still low, but no longer the fortress number, and now above the entire energy sector. And Alphabet didn’t only borrow: it raised about $85 billion of equity in June, its first share sale in more than two decades, on top of nearly $100 billion of debt this year. A company that taps equity at these levels is telling you something about how it sees the funding gap. Neither fact breaks the thesis. Both belong in it.

Part III — The Household

Records on Both Ends of the K

Good debt, bad debt

Before the data, the taxonomy — because “consumer debt” in aggregate is close to meaningless.

Good debt	Fixed-rate, amortizing, attached to an appreciating or income-producing asset, priced below your expected nominal income growth. A 30-year mortgage at 3% taken out in 2021 is the best financial instrument available to an American household in my lifetime. Inflation repays it on the borrower’s behalf. The homeowner is effectively short a bond at a negative real rate.
Bad debt	The mirror: floating or revolving, non-amortizing, attached to a depreciating asset, priced above nominal income growth. A revolving balance at 21%. An 84-month auto loan on a vehicle that loses half its value in five years — a contract mathematically guaranteed to keep the borrower underwater for most of its life. BNPL stacked on consumables.
Gray debt	Student loans, where the asset is human capital: excellent when the credential earns a real premium, catastrophic when it doesn’t, and non-dischargeable either way.

The single most important fact about the household balance sheet is composition: roughly 70% of the $18.8 trillion is mortgage debt — about 72% counting HELOCs — most of it fixed and locked below today’s market, which is now 6.78% on a 30-year conforming loan. That’s why the aggregate keeps looking better than the anecdotes.

The aggregate: better than the mood

The New York Fed’s Q2 report, out August 11th, is not the report a bear wants:

• Total household debt fell $13 billion to $18.8 trillion — though the Fed’s own researchers flag that the decline is largely an artifact of a “servicer transfer gap,” delayed reporting when a mortgage moves between servicers. Don’t lean on it.
• Mortgages −$74B to $13.1T. Credit cards +$21B to $1.26T. Autos +$28B to $1.71T. Student loans −$7B to $1.65T. HELOCs +$13B to $459B, $142 billion above the Q1 2022 low and still expanding.
• Aggregate delinquency improved to 4.7% of balances, from 4.8%. Early-delinquency transitions ticked down for cards (8.7% to 8.6%) and mortgages (3.9% to 3.8%).
• Debt service is about 11.2% of disposable income. Pre-GFC peak: 15.8%. COVID trough: 9.1%.
• Household debt is about 79% of disposable personal income — excluding two distorted pandemic quarters, the lowest in twenty-four years, against 116%+ in 2007–08.
• Bankruptcy filings ran about 137,000 in the quarter — below the pre-pandemic pace.

And two pieces of primary-source work I consider the most valuable data in this letter, because both cut against the scary headline:

The New York Fed’s own Liberty Street analysis found the rising stock delinquency rate on credit cards is largely an artifact — a pool of stale, charged-off debts lenders have been leaving on credit reports longer than they used to, rather than a fundamental worsening in the incidence of delinquency.

The Philadelphia Fed, in April, examined the record subprime auto number and concluded the increase is driven primarily by loans remaining delinquent across multiple quarters and by redefaulters — not by a growing inflow of newly delinquent borrowers. First-time subprime delinquency entry has been broadly stable since late 2022.

Both say the same thing: the headline is measuring accumulated damage, not accelerating damage. That’s a materially different economic condition, and almost nobody writing about the consumer has adjusted for it.

The other end of the K

Now the part the aggregate is hiding. And I want to be straight: the same NY Fed report I just used to make the bull case contains the following.

• Transitions into early delinquency rose for mortgages and autos. A greater share of borrowers went 30-plus days late on a mortgage than in any quarter since 2015.
• More went seriously delinquent — 90-plus days — on a car loan than in any quarter since 2010.

Layer the rest on top. Subprime 60-plus-day auto delinquency hit 6.9% in January per Fitch — a 32-year high, above the Great Recession peak. TransUnion and Equifax both describe the credit market in explicitly K-shaped terms: record bankcard origination driven at both ends simultaneously, super-prime and subprime, with subprime borrowers leaning harder on high-cost credit. Revolving balances at $1.351 trillion. Savings rate around 4%, down from 5.2% at the start of 2025.

Meanwhile household net worth set a record above $180 trillion — and the gains were overwhelmingly financial assets, which is to say they accrued to households that already owned assets. Michigan consumer sentiment printed 49.8 in April — the lowest in the series going back to 1978, below the June 2022 record of 50.0 and below anything recorded in the Great Recession or the COVID collapse. The August preliminary came in at 51, down 7.6% on the month. The Conference Board’s confidence index fell to a seven-month low of 89.4 in August.

Hold those next to each other: record aggregate net worth, sentiment worse than the financial crisis. Not a contradiction — a description of two economies.

The synthesis: the American consumer in aggregate is in defensibly good shape, and the aggregate is a composition effect. The top two quintiles hold the assets, hold the 3% mortgages, and are carrying consumption. The bottom quintile holds the revolving balances and the underwater autos and has been in a private recession for three years.

What breaks the good half? Not debt. Employment. That balance sheet is bulletproof as long as the paycheck arrives — and July payrolls came in at −23,000. Read the internals before you panic or dismiss it: the headline was driven by −53,000 in government, including −50,000 in local-government education, while private payrolls actually rose 30,000. The number that should worry you is not the headline — it’s that May and June were revised down by a combined 103,000, participation fell to 61.4% (lowest since February 2021), and average hourly earnings growth slid to 3.2%, a five-year low. That’s the number to watch, not the delinquency print. Household debt problems are almost always employment problems wearing a costume. Albeit a lagging indicator, testing this patient is a must.

Part IV — The Backtest

Why This Calendar Is the Worst One We Own

Now to the part that should shape sizing between here and November 3rd.

What the backtesting shows us

1. Aug–Oct is the worst three-month window of the year. BofA’s seasonality work puts it as the weakest stretch on average since 1928. Since 1990, both August and September have carried negative average monthly returns.

2. Midterm years are the weakest seat in the four-year cycle. Lowest average returns of the presidential cycle across the last three decades. Capital Group’s long series has midterm years averaging 4.7% since 1931 against 9.5% for all other years — half the return.

3. The drawdowns are the real story. The average intra-year peak-to-trough decline in a midterm year runs roughly 17–19%, versus a normal-year average closer to 13%. In 11 of the last 16 midterm years, the S&P fell 15% or more at some point. Longview’s dataset across 25 midterm cycles since 1926 puts the average pre-election drawdown at −19.4%, and it typically bottoms in exactly this Aug–October window as election uncertainty peaks.

4. And then it resolves, with unusual reliability. The S&P has been higher twelve months after every midterm election since 1950. From the midterm-year trough, the average subsequent twelve months has run roughly +31%. Post-midterm averages: +5.7% at three months, +10.5% at six, +13.6% at twelve, with only two negative twelve-month periods in the entire series — 1930 and 1938, both inside the Great Depression. Roughly 87% of midterm years close positive despite the violence in the middle.

A caveat on my own evidence: those four points come from four different vendors with four different sample windows — 1928, 1931, 1926, 1950. Stacked together they look more precise than they are. Treat the direction as real and the decimals as decoration.

What is odd is we really haven’t experienced a BIG correction

What makes this year uncomfortable to me is that the S&P’s deepest drawdown this year was about 9% peak-to-trough in March — less than half the midterm average. The index set a record close of 7,798.99 on August 13th, touching 7,816.70 intraday, sold off hard the week of the 20th as long yields spiked, and sits around 7,680 — about 1.6% off the high, with 27 record closes and a 13% gain year to date. Seven-plus brokerages now carry 8,000 year-end targets.

Earnings did the work — but read the fine print. Blended Q2 S&P 500 EPS growth was 50.4%, the best since Q2 2021, and 86% of reporters beat, the highest hit rate since 2021. Now strip out two companies. Excluding Alphabet and Amazon, growth falls to 32% and the aggregate earnings surprise falls from 29.2% to 10.9%. And what drove those two? Alphabet booked a $98 billion unrealized gain on equity securities; Amazon booked $53.4 billion, primarily on its Anthropic stake.

Roughly a third of the best earnings quarter in five years is a mark-to-market on private AI companies. That is not a cash flow. It is a valuation opinion, booked as income.

So we enter the historically worst ten weeks of the cycle within two percent of the record, with the seasonal correction unpaid, on an earnings story concentrated in the same handful of names now issuing $220 billion of debt — and partly marked to the value of the very companies that debt is being raised to serve. Nvidia reported Wednesday night; how the tape treats that print tells you how much of this is still faith.

I want to be careful here, because seasonality is a probability distribution, not a schedule. Plenty of midterm years took their pain in the first half and rallied through the fall. But the combination — no toll paid, record index level, hawkish Fed, term premium expanding, and the specific catalysts in Part V — is a genuinely poor risk-reward setup into October. Food for thought.

My read: the base case is a 5–10% drawdown that finds its low somewhere between late September and the week after the election, followed by the seasonal resolution the record describes. We need these WARS to END

Part V — The Wobbly Mess

Five Fronts at Once

Which brings us to why I think this is a genuinely precarious moment rather than a routine seasonal one. Count the live fronts.

1. The Iran war and the Strait. Approaching six months with no military or diplomatic resolution. The June Versailles agreement set a 60-day deadline that expired August 17th with the parties further apart than when they signed. Brent has traded a violent range this year — Dated Brent above $140 at the March panic, back to the $87–90 area now — and it moves on headlines out of Hormuz, which remains far less active than pre-war.

In fairness to the tape, the last ten days have gone the other way. Washington answered with sanctions rather than fresh military threats, Iran resumed talks with Oman on managing the Strait, crude fell for three straight sessions, and long yields came in with it. That is de-escalation, and I am not going to pretend otherwise to protect a thesis. It is also exactly the kind of headline-driven relief that reverses on a single tanker.

This is the single most important variable in the letter, and here’s why: oil is the transmission mechanism from geopolitics to your portfolio’s discount rate. J.P. Morgan’s framework is that each sustained 10% rise in crude takes 15–20bp off GDP and 2–5% off S&P earnings. But the second-order effect is worse — an oil shock raises inflation expectations, which pushes the Fed hawkish into a softening labor market, which is the definition of a policy trap. That’s the mechanism that turns a normal 10% seasonal correction into a 20% one.

2. Ukraine–Russia. Now longer than the Soviet fight against Nazi Germany, with a frozen-but-lethal front, drone campaigns against Russian export terminals and refineries, and Russian strikes on Ukrainian energy infrastructure. It matters here for one reason: it’s a second constraint on global energy supply, layered on the first.

3. The trade war with China — quiet but structural. Beijing’s expanded rare-earth export controls sit suspended only until November 10th, 2026 — one week after the election. China holds roughly 70% of global REE production and, per the IEA, about 94% of permanent magnet production. The IEA puts $6.5 trillion of downstream production outside China at risk under full reimplementation. That’s not an abstraction for a defense and space book. It’s an input to F-35s, drones, EVs, and every actuator and motor in the supply chain.

Two things I’d add for precision. It isn’t binary — Beijing can extend, selectively reinstate, or fully reimpose, and the licensing machinery for a partial reinstatement already exists. And the April 2025 regime covering seven heavy rare earths, dysprosium and yttrium among them, was never suspended and still bites today. The November date is a hard, dated catalyst on the calendar. I’ll stop short of saying nobody is watching it — the IEA, CSIS and half the compliance industry have been publishing on it for months. What I will say is that almost nobody is positioned for it.

4. The trade fight with Canada — the new front. Talks collapsed Friday night the 21st and 50% Section 338 tariffs hit $20 billion of Canadian goods at midnight. Carney walked away from what he called uneconomic and unfair terms, said on the 22nd that Canada is “at war” — his word — and committed to match the tariffs dollar for dollar starting September 8th. Washington’s account differs: Greer says Canada declined to finalize terms already agreed. Either way, September 8th is another dated escalation inside the window. Beijing’s state press immediately and gleefully praised it as a “Chinese-style counterattack,” which tells you exactly how it’s being read internationally. Canada holds uranium, potash, nickel, cobalt, graphite, copper and REE potential. Fighting the largest non-China critical-minerals holder while trying to build non-China supply chains is a strategic contradiction, and markets will eventually price contradictions.

5. Sanctions escalation — and this one is about Iran, not China. On August 24th Bessent launched “Operation Economic Outcast,” which he called an economic D-Day: sectoral determinations across five of Iran’s lifelines — digital assets, technology, gold, aviation and shipping — plus OFAC designations on more than 60 individuals, vessels and entities, and a broad expansion of secondary-sanction exposure for anyone transacting with Tehran. Trump is personally calling foreign leaders with compliance deadlines that have not been made public.

Here is the tell, and it is a better one than the headline. U.S. officials say expanded secondary sanctions are expected to be the main course of action against Iran until at least after the midterms, when a new military campaign could come back on the table. Read that as written. The administration has told you it intends to fight this war with the Treasury until November 3rd and is reserving the option to fight it with the Pentagon after. That is a dated risk sitting one week outside the window I am telling you to buy into.

Put it together. A hawkish Fed holding at 3.50–3.75% with three members voting to hike and a chair in Kevin Warsh, who has staked his credibility on “there’s only a target and it’s 2%.” A market that flipped from pricing cuts in January to pricing a real chance of a hike — CME futures put September at roughly 38%. Core PCE stuck at 3.3% year over year, headline at 3.7%. Payrolls negative. Oil hostage to a strait. Rare-earth controls expiring the week after the election. A synchronized global bond rout — 30-year Treasuries above 5.20% and at 20-year highs, French OATs at 2008 levels, Bunds at 15-year highs, long gilts near 6%, JGBs off a three-decade high — and an IIF global debt stock at a record $348 trillion.

And the Treasury doubling buyback operations for longer-dated bonds, with reporting that Bessent could tap close to a trillion from the General Account to fund them. Read that carefully. It’s plumbing, not policy — but it’s also an admission that the long end needs a buyer of last resort at these levels. When the issuer starts supporting its own secondary market, that’s information. It worked, in the narrow sense: the 10-year fell from 4.75% to 4.62% on the news and the dollar hit a three-month low.

Just this week Stanley Druckenmiller — for whom Bessent worked twice — said the quiet part out loud: the buybacks undermine the credibility of the Treasury market and waste the opportunity for meaningful debt reform. He’s right, and it’s worth noticing that the objection came from the man who trained the Secretary.

Part VI — Positioning

What I’m Actually Doing

I run a concentrated micro- and small-cap book in defense, space and adjacent technology. So let me be direct about what changes and what doesn’t.

The bad news first, because it’s mine. My working line has been that a 10-year above 4.6% is bearish for a low-float, beta-sensitive micro-cap book. We are through it — 4.65% now with 4.75% behind us, and 82% of surveyed strategists flagging upside risk to their own forecasts. A pre-revenue space name is the longest-duration equity on earth: a bundle of cash flows sitting five to ten years out. Nothing is more sensitive to the discount rate. Parts I, IV and V together argue for smaller positions and tighter catalyst windows in the speculative sleeve, not larger ones. I’d rather say that plainly than pretend the macro cooperates with my book.

The good news, and it’s structural. The same fiscal arithmetic that crushes duration multiples is protecting my sector’s revenue line — and the mechanism gets missed. When interest and entitlements consume a rising budget share, discretionary spending gets crushed. But not uniformly.

The FY2027 request asks roughly $1.5 trillion for defense — up 42% against FY2026’s enacted level, a $445 billion increase — built as $1.15 trillion of discretionary plus $350 billion routed through reconciliation, while non-defense discretionary is cut about 10%. That structure is the tell: routing defense money through mandatory reconciliation rather than annual appropriations is how you insulate a priority from the caps the debt trajectory will eventually force. Procurement and RDT&E together total $756.8 billion — larger than the entire defense budget of a few years ago. Munitions procurement alone goes from $26.8 billion to a requested $76.3 billion. The Space Force is the fastest-growing service line in the request at +77%.

Now the part that cuts against me, and it’s sharper than I’d like. A request is not an appropriation — but the gap is bigger than that phrase suggests. None of the congressional spending bills account for the reconciliation money at all. The HASC, SASC and House Appropriations markups all track the $1.15 trillion discretionary line and simply omit the $350 billion, which functionally reads as a $350 billion cut to the request. House appropriators went further and trimmed procurement by another $9.2 billion. The Defense-Wide accounts carrying the industrial-base money — $54.6 billion for autonomous systems, $40.6 billion for industrial base sustainment — live almost entirely inside the piece Congress hasn’t funded. Continuing-resolution plans suggest full-year Pentagon funding gets punted past the start of the fiscal year. And the administration’s own out-year profile has national-defense funding falling about 16% after the FY2027 spike.

So the honest version of my thesis is narrower than the headline: the front-loaded modernization surge is real and it is where I want to be, but the reconciliation leg — the exact structural feature I called the tell — is also the leg Congress has not yet agreed to. House appropriators have already funded several space programs below request, and the Space Force cancelled GPS OCX in April. Programs die. The direction still survives a fiscal squeeze better than almost any other line in the federal budget. The timing is a coin flip on a reconciliation bill that doesn’t exist yet.

Four conclusions

1. Carry less risk into October than I did into August.

   Not a cash call — a sizing call. NO MARGIN. The seasonal toll is unpaid and the catalysts are dated. I want dry powder for the window between late September and mid-November, because the same record that describes the drawdown also describes the +31% average off the trough. The plan is to be a buyer into that, not a seller during it.

2. Favor the funded over the promised.

   In a rising-term-premium regime the market pays for near-dated contracted cash flow and punishes the story. Signed programs, appropriated line items, delivery against a PO — those beat TAM slides, and the gap should widen from here. That’s my LOI-to-PO discipline applied to the macro. Note the discipline applies to my own thesis: $350 billion of reconciliation money is a promise, not a funded line.

3. Hard assets earn a permanent allocation, not a trade.

   The bid under gold is sovereign, structural, and price-insensitive. That’s a different animal from a momentum bid, and Q2 demonstrated it in the cleanest possible terms — 289 tonnes bought into a 14% decline.

4. Watch November 10th.

   The rare-earth suspension expiry is the most specific, most dateable, least-priced catalyst on the board for anything touching magnets, actuators or drone supply chains. It lands one week after the election, when attention is elsewhere.

The Close

A Claim on Future Income

Debt is not a moral failing and it isn’t a bomb. It’s a claim on future income, and the only real question about any borrower — sovereign, corporate or household — is whether the income shows up.

For the corporates, it will. The cash engine is intact and compounding at 23%. What looks like distress is a spending decision they can unmake, and the honest risk sits downstream of them rather than at them.

For the top half of American households, it will. Locked-in fixed-rate mortgages, record net worth and an 11% debt service ratio are not the ingredients of a credit crisis — provided the paycheck arrives, which is now the only variable that matters.

For the federal government, it will not. Not at 4.5% real growth, not for a decade, not with a 2.6% primary deficit and a borrowing cost converging up toward the market rate. The growth thesis fails on arithmetic before it ever reaches politics.

That’s the whole letter in one line: the corporate and household balance sheets have a timing problem. The sovereign balance sheet has a REAL problem. The repricing has already started, in the term premium, in the tonnage central banks are quietly moving into vaults, and in a global bond market that spent this month telling five governments at once that the price of long money has changed.

Between here and November we have the worst calendar in the four-year cycle, a year without a true correction, a war holding the world’s most important shipping lane hostage, a second war grinding on, a trade fight on two fronts, a rare-earth deadline nobody has circled, and a $41.1 trillion debt limit coming into view behind all of it.

We are in some of the most interesting times and remain bullish on the 4th Industrial Revolution; however, I’m cautious on the quarter. Those are different statements and confusing them is how people lose money in both directions.

Watching the 10-year closely. Need this war in Iran to end asap!

Thank you for reading,