# The Water Rising Slowly: Long Bond Yields, Off-Balance-Sheet Debt, and the Illusion of a Great Track Record

> Notes from listening to The Compound and Friends, August 18, 2026. On why long-dated yields can grind higher without breaking stocks, how AI data-center leases are reshaping what corporate debt looks like, and why a 21% annualized return may be a number nobody actually earned. Educational reflections on process, not investment advice.

Published: 2026-08-19
Locale: en
Tags: treasury yields, off-balance-sheet debt, AI capex, reflexivity, investing mindset


![Dusk at the inner face of a flood wall, year-marks of rising water levels carved into the concrete, the newest line above the head of a person standing below and looking up, a city skyline fading into haze in the distance](/covers/compound-2026-08-18-treasury-yields-break-out-how-to-invest-with-bill--cover.png)

> Accumulated feathers will sink a boat; a pile of light things will snap an axle.
>
> — *Strategies of the Warring States*, "Strategies of Wei I" (compiled by Liu Xiang, Han dynasty; Warring States material)

## What this episode is about

The August 18, 2026 episode of The Compound and Friends starts with the slow grind higher in long-dated Treasury yields and travels through data-center lease obligations, credit default swap spreads, the turning of the enterprise-software narrative, and finally lands on one famous investor's track record.

The hosts open with a piece of self-definition worth keeping. Financial television has to cover everything that happened in the market that day. They don't. What they do is curate rather than filter — "the things we don't mention aren't automatically unimportant." One of them added something better still: there's a lot of this I simply don't feel qualified to talk about.

That sentence is more useful than any single call in the hour. People willing to say "this is outside my lane" are usually more trustworthy than people who have a view on everything.

## The key points

**1. Damage comes from speed, not level.** The 10-year went from 4% in early March to about 4.7% — a five-month grind. Their read: had the same move happened in two weeks, stocks would be materially lower. Instead the market got five months to digest it and went back to trading earnings. Same magnitude, different clock, opposite conclusion.

**2. The 30-year makes headlines; the 10-year sets prices.** The long bond is at its highest since 2007 and that's the number everyone is writing about. But it's roughly 1% of new issuance, and mortgages and most asset prices key off the 10-year — which has been going sideways in the same range since September 2023. They offered a clean conditional: if you could tell me with certainty the 10-year breaks out and hits 5.5% by winter, that would absolutely pressure risk assets. That's not where we are.

**3. This is global, and it punishes the fragile.** A chart of 10-year, 10-year-forward yields across developed markets shows almost everything rising except Switzerland — and rising most where the debt stock is high and political dysfunction is acute. Japan moved most over the prior ten days, then the UK, France, Italy. The framing worth stealing: it isn't about which shock set it off. When you carry a lot of debt and run unsustainable deficits, you're vulnerable to any old shock that comes along. Fragility isn't inflicted on you; you build it yourself.

**4. The fiscal numbers have entered another register.** A $432.3 billion shortfall in July alone, locking in roughly a $2 trillion deficit for the fiscal year. Debt service through July: $1.37 trillion, up $84 billion year over year. The government now spends more on interest than on anything in the budget other than Medicare and Social Security. Meanwhile US companies have issued $1.7 trillion in bonds year to date, 27% above the same period last year. Those AAA corporate credits yield more than Treasuries — buy Alphabet paper instead of a Treasury and, in theory, that's one fewer buyer of government debt. You don't have to accept the mechanism to find it a useful explanation for why soft data hasn't stopped the long end from climbing.

**5. This time the off-balance-sheet stuff is in plain sight.** Meta's Louisiana data center isn't funded on its own balance sheet; outside capital takes the credit risk, and under the accounting rules the lease obligation stays off the books until rent starts being paid. Meta has disclosed roughly $347 billion in total obligations for leases that haven't kicked in yet. But the hosts' point runs the other way: in 2008 people opened up insurance company holdings after the fact and asked "wait, how did we get here?" Today every analyst on earth can model this, the dates are on a calendar, it's in the price of the stock and the credit default swaps. And the thing that blows up the market usually isn't the thing staring right at you.

**6. What has no precedent is the breakup.** These are multi-billion-dollar, extremely complicated, very long projects. The contracts are exhaustively negotiated — you can imagine the legal fees. But we have never seen a tech giant and an East Coast private equity firm actually fight one of these out in court. Not all of this compute is going to turn out to be a good project, and while there are laws on the books, there's no practical roadmap for what these look like should they come undone.

**7. Reflexivity is running through the credit market.** Oracle's CDS spread looks nothing like the other hyperscalers'. The hosts' read is that the market isn't genuinely pricing default — protection on $10 million runs about $200,000 a year, a million over five years, and few people are actually laying that out. It's a trading vehicle: leverage and a fast move. And the spread looks the way it does substantially *because of the equity*. Soros called this loop reflexivity: prices shape the reality of a business, and that reality feeds back into prices. It runs upward too — the company currently perceived as the AI leader gets every meeting and every deal, which means more contracts and a higher valuation. They noted Anthropic disclosing roughly a $65 billion annual run rate, from zero three years ago. That revenue number is precisely why a trillion dollars of annual capex in 2027 feels tolerable to the market.

**8. A year without a single real down day.** My favorite cold statistic of the episode: so far this year there has not been one session where more than 80% of volume was to the downside. The average year sees 21 such days, and there has never been a year with fewer than five. Part of the reason is that energy ripped even during the war — an unusual crop of negative-beta stocks has made the shape of this year genuinely strange.

## Going further

### "Rates are rising — why is my portfolio fine?"

Underneath that question is usually a more anxious one: is the market wrong, and am I on the wrong side of it?

Split it into three layers. First, **is this variable actually the anchor for what you own?** Mortgages, corporate borrowing, most discounted cash flows key off the 10-year. The 30-year is a thermometer for term premium and fiscal anxiety — important, but it isn't your discount rate. Headlines pick the most dramatic number because that's what makes a headline. Your job is to pick the one physically connected to what you hold. When the star of a story isn't your anchor, your first move is to go check whether your anchor moved. Half the anxiety dissolves right there.

Second, **speed**. Seventy basis points over five months and seventy basis points over two weeks are different events. The slow version gives every holder time to adjust duration, rotate, re-run the numbers. The fast version forces everyone to do the same thing on the same day. Markets survive direction; they don't survive synchronization. Which is why "it's been going on for a while" can be reassuring rather than alarming — it's evidence that someone has been absorbing it the whole way.

Third, **has the range broken?** Something that has chopped sideways for two years carries no information inside the range. It carries information when it leaves. That gives you a cheap discipline: stop reacting to its daily position, set one level that means "I re-think this from scratch," and otherwise leave it alone.

Where this framework fails: if the *reason* for the slow climb is itself deteriorating — buyers structurally leaving rather than the market clearing — then "slow" is no longer evidence of safety, just evidence that the moment hasn't arrived. Watch the reason, not only the slope.

### "The news says debt crisis. Should I de-risk?"

Start with a blunt but useful question: **how many people know about this?**

The contrast in the episode is excellent. What made 2008 terrifying was that "what is a CDO squared?" was a question asked *afterward*. Today's lease obligations have dates on a calendar, numbers in the disclosures, models on every analyst's desk, and quoted prices in the swap market. Something everyone knows, everyone can calculate, and everyone can trade gets absorbed into price gradually. It rarely kills you on a random Tuesday morning.

So where is the actual risk? **In the places with no precedent.** However carefully a clause is drafted, you don't know what it's worth until someone has enforced it. Who is finally on the hook, how a ten-year lease gets unwound when nobody wants the project, who has priority when it does — no case law means no consensus, and no consensus means no way to price it. So instead of asking "is this number big?", ask "has anyone run this through once where the rest of us could watch?" The first is a known risk. The second is uncertainty.

This is also why "good news, stock falls" and "bad news everywhere, market holds" aren't signs of insanity. Price reflects the gap between expectation and what's known — not the absolute severity of the facts. You're always hunting for the part that differs from what everyone assumes, not for the part that sounds most alarming.

One practical caveat: even if something does eventually break, that doesn't mean you can make money from it. Being right and being right *on time* are different problems — which is exactly the difficulty with the hedging instruments discussed in the episode.

### "His returns look incredible — why not just buy what he buys?"

The most valuable stretch of the hour is the discussion of that 21%.

Roughly 21% annualized since 2004 versus about 10–11% for the index. The number is real, the math is sound, it's compliant. Then comes the sentence that guts it: **there is probably not one investor on earth who earned that.** The drawdowns were unbelievable. The capital hopped between vehicles. In the worst years, nobody sat still. The manager himself earned it — that's why he's a billionaire, and nobody can take that away — but it is not a path an outside investor could have walked.

Generalize it into a habit: **whenever you see a beautiful long-run number, ask whether anyone actually received it.** Does the person who bought on day one, never redeemed, never switched vehicles, and still holds today exist? If the answer is "theoretically," you're looking at a mathematical curve, not a plan you can execute.

The episode also unpacks the plumbing, and each layer is a common trap. First, the same person's ideas are purchasable through several different vehicles — a portfolio of stocks, a closed-end fund, a management company, a Berkshire-style holding company. Whether you're buying his *stock picking* or his *business* changes where the return comes from entirely. Second, closed-end funds trade at discounts to NAV — good news for a value buyer, but note that the most direct way to close a discount is buying back stock, and buybacks shrink the fee stream paid up to the management company. That's a structural, built-in conflict, not a character flaw. Third, fees: a 2% management fee, a hurdle rate that isn't high, and not the traditional "only if I beat the index" design.

They also gave the fair counterpoint, which matters just as much: at least this is a genuinely concentrated portfolio, not a closet indexer. He isn't mimicking the index; he says he's smarter than the market and puts his own money behind it. That's more honest than a parade of managers who end up 300 basis points from the index in one direction or the other, every year. **Paying up is fine — just make sure you're paying for *different*, not for "roughly the same, plus a bit."**

## Worth a look

- The Compound and Friends, episode of August 18, 2026 (the source of everything discussed here)
- The cross-country comparison of 10-year, 10-year-forward yields published by Robin Brooks
- Fiscal and issuance figures as reported by CNBC's Jeff Cox, plus the Barclays rates research view quoted in the episode
- The optimistic read — that the bond market is finally working properly and allocating capital efficiently — comes from Ed Yardeni, who coined the term "bond vigilantes"
- The argument for why enterprise software exists at all comes from Paul Kedrosky on the Big Technology Podcast: companies want someone they can yell at or sue when things go wrong
- Reflexivity is George Soros's concept
- The earnings-reaction chart format follows work by Warren Pies and team

## The one thing to take away

**The same pressure, applied slowly, gets absorbed; applied all at once, it breaks you. So when something is deteriorating, look at its speed before you look at its level.**

That's the whole episode, really. The same yield move is background over five months and an event over two weeks. The same debt load is a modeling exercise when it's on a calendar and a panic when it's discovered on a Tuesday. The same 21% annualized return means something completely different depending on whether it arrived smoothly or via three craters you had to climb out of.

People work the same way. The slope of the pressure, not the amount of it, decides whether you adapt or snap — in your body, in your relationships, at work.

**An exercise for today** (nothing to do with investing): pick one thing you've been quietly worried about — weight, hours of sleep, mortgage balance, overtime, a family member's health metric, how often you talk to someone who matters. Write out its last twelve monthly values in a single row. Then do exactly one thing: **circle the month with the biggest single-month change**, and remember what happened that month.

If all twelve look similar, that's drift — you've probably already adapted, and it doesn't need handling today. If one month clearly gaps, that gap is the real thing to deal with, and the answer is almost never in the numbers. It's in what your life looked like that month.

We're trained to watch the level: the kilograms, the digits in the account, whether things feel okay. But it's never the level that breaks people. It's the one jump nobody was looking at.