Subscribe

Good Saturday all! (Kinda I guess)

In the words of a 90s favourite, Cheryl Crow…. There goes the neighborhood!

Let's give that song some context….

On the evening of 9 November 1989, an East German official named Gunter Schabowski held a press conference he had not properly prepared for.

He had been handed a note about new travel regulations shortly beforehand. He had not been briefed on it. Somewhere near the end, an Italian journalist asked when the new rules would take effect. Schabowski shuffled his papers, looked at them, and said, as far as he knew, immediately, without delay.

That was not the policy. The regulations were meant to take effect the following day, through an orderly application process, administered by officials who would have gone home for the night. Schabowski had no authority to say what he said. He was reading a note.

Within hours the crossings at Bornholmer Strasse were overwhelmed by people holding him to it. The Berlin Wall did not fall because of a decision. It fell because a man who was not supposed to be giving guidance gave some by accident, and the market took him at his word.

I thought about that on Friday, watching the most effective piece of forward guidance in a decade delivered by a man whose entire public identity is built on refusing to provide forward guidance.

Kevin Warsh spoke at Jackson Hole. Two year Treasury yields jumped 12 basis points to 4.35%, the biggest move at that event since at least 2010. September rate hike odds went from about 34% to roughly 60% in the span of one speech. Barclays flipped from forecasting no hikes this year to a quarter point in September and another in December. The dollar rallied, gold fell, Bitcoin fell, and thirty year yields barely moved.

The head of fixed income at DWS Americas put it best. Call it what you want, he said, but this is the exact forward guidance the market was looking for at the July meeting.

Warsh, in the same speech, defended his refusal to give forward guidance, argued it can mislead households and businesses outside a crisis, and said markets need to form their own views on the economy.

They did form their own view. They formed it in about ninety seconds, off his sentences, in the direction he pointed.

The genuinely delicious part is what he actually said, because there is an academic paper about this published the day before.

Two economists updated a 2025 study on when Fed communication works. Their finding: guidance is most effective when the central bank talks about financial conditions rather than the direction of interest rates. Rate projections are contaminated, because a projected rate blends the Fed's view of demand with its view of the risk premium, and nobody can tell which is which. Financial conditions distil pure macroeconomic judgement.

Warsh's market-moving line was this: he would be hard-pressed to describe broad financial conditions as restrictive.

He gave the single most effective form of forward guidance identified in the literature, on the day after the literature identified it, while insisting he does not do forward guidance. If you had told me a year ago that the way to fix Fed communication was to appoint a man who hates communicating, I would have laughed. He may have accidentally solved it.

The number underneath the speech is the one nobody is putting on the front page. Bloomberg's financial conditions gauge hit a record loose level the day before Jackson Hole. Equities near all-time highs, VIX and MOVE both near cyclical lows, inflation at 3.7% and above target for five consecutive years.

Try to say the words "monetary policy is restrictive" while looking at that. Warsh could not, so he said so, and the entire front end repriced.

The closest precedent is 1993 to 1995, when financial conditions loosened, momentum strengthened, and Greenspan tightened pre-emptively by 300 basis points over fifteen months to lean on inflation before easy conditions fed into prices. That episode is remembered as the bond massacre for a reason, and it is worth remembering that it began with a Fed nobody expected to move.

Now, my ledger, and it is mixed with one genuine loss.

The rates volatility position I flagged on Tuesday with the MOVE index at the twentieth percentile has worked, and it worked for the reason I gave. I argued that volatility was too cheap for the amount of event risk in the calendar and that a chairman who declines to narrate his reaction function structurally raises realised volatility around meetings. Friday produced the largest two-year move at Jackson Hole in sixteen years, and options going into it were pricing a muted reaction, with SPY implied volatility in the third percentile of its one-year range. Nobody paid for that move. Stay long. Add. The MOVE index is back down in the eleventh percentile on a five-year lookback, which after that speech is close to comedy, and Warsh has told you plainly he intends to keep not explaining himself, which makes every single meeting between here and March a live event.

The thirty-year breakeven trade from Tuesday is wrong and I am taking it off. My argument was that drawing down the Treasury's cash account pushes reserves into the system and lifts inflation compensation. The mechanism is real. What I failed to name as an invalidation was a credibility restoration, and that is exactly what happened. Ten-year real yields rose about three basis points to 2.37% while nominals rose two and breakevens fell about one to 2.32%. Real yields leading is the market saying it believes him. That is the opposite of my thesis, cleanly and immediately, and the correct response is to close it rather than to explain why the market has misunderstood my excellent idea.

Gold is the same story wearing different clothes. I am long from $4,415 on the eighteenth and it fell on Friday. The debasement trade was premised on the erosion of Fed credibility. Warsh spent an hour restoring it. In a world with abundant real yield across sovereign and corporate markets, a metal that yields nothing has a harder argument to make. I am not selling into the move, but I am tightening the stop to the 100-day, and I would rather be honest that the reason I own this is now weaker than it was on Thursday.

The Canadian dollar short, which I cut in half on Wednesday while explaining why I was wrong, has started working. Second-quarter GDP came in at 3.3% annualised against 3.4% expected, the July flash estimate was flat, and the loonie hit session lows. Note the awkward detail in that report, which is that the prior quarter was revised up, confirming Canada never actually had a technical recession and was more resilient to tariffs than anybody feared. So the trade is working on momentum loss rather than on the tariff thesis I originally wrote, which is a distinction worth keeping straight in your own head even when the profit and loss does not care.

The trade I like most today is the ugliest part of the curve.

Warsh has stripped out forward guidance, which means the five to seven year sector has lost its anchor. That part of the curve prices the terminal rate, the full cumulative tightening path over two to three years, and it can only do that against some reference point for where policy ends up. There is no longer a reference point. The front end knows exactly what it is pricing, because a September hike is a binary. The long end has the Bessent buyback bid underneath it as a technical floor. The belly has neither. It carries maximum rate-path uncertainty with no support.

Watch what already happened. The 5s30s spread peaked around 93 basis points on 17 August as the market priced fiscal risk into the long end. It has retraced that entire move in under two weeks. Warsh undid in one speech what the bond market spent a month building.

Buy the wings and sell the belly. Long two-year and thirty-year Treasuries against a short five-year, sized on duration rather than notional. You are not betting on direction. You are betting that the sector with no anchor and no floor cheapens against the two sectors that have one each. The invalidation is Warsh reversing and providing an explicit terminal rate view, which he has now told you twice he will not do, and the risk is that the Fed talks tough and repeatedly declines to act, in which case the front end eventually stops listening and this unwinds from the wrong end.

Elsewhere, a genuinely magnificent piece of arithmetic. SpaceX and Anthropic have each recently suggested their ultimate addressable market could approach $30 trillion, which is roughly the size of the entire US economy. For scale, one strategist charted the cumulative combined revenues of Apple and Microsoft from 1990 to today. Adjusted into current dollars, the two most successful technology companies of the modern era, over a period beginning two years before the invention of the world wide web, do not reach $10 trillion between them.

Not annually. Cumulatively. Everything they have ever sold, added up, in today's money, is a third of what two private companies have described as their eventual market.

Hyperscaler credit is quietly telling you what the professionals think. Alphabet, Amazon, Meta, Microsoft and Oracle are trading more than a standard deviation wide of their 52-week average spread, and they are dragging the communications sector wider with them even after you strip their own bonds out of the index. They are borrowing more than creditors want to lend and they do not appear to mind paying up as long as the size gets done, which is a sentence that has preceded every credit cycle turn I have watched since 1999.

One last positioning note that made me laugh out loud. The CFTC report showed speculative accounts sold 59,000 ultra bond futures last week. That is the largest weekly net sale in the history of the contract, and it came directly after the Treasury Secretary announced he would be doubling his buybacks of long-dated bonds.

The government announced it was buying. Record numbers of people sold it the paper.

Schabowski, incidentally, spent the rest of his life being asked about that press conference. He never claimed it was deliberate. He said he was tired, he had not read the note properly, and he answered the question in front of him.

Warsh has a whole speech about why he will not tell you what he thinks, delivered inside a speech that told everybody exactly what he thinks. He will do it again next month, and the month after, and each time the market will insist it was not guidance.

Immediately, without delay.

Crypto!

In my small circle of degens everyone is arguing about whether quantum computers will steal your Bitcoin. The real question is what happens to the four million coins everybody has already written off.

Yesterday, in Bitcoin block 964,199, a researcher moved ten thousand satoshis.

At current prices that is about eight dollars. Call it a sandwich, without the drink…. Especially in the Singapore CBD lately ..

The transaction cost somewhere between $150 and $200 to produce, took hours of computation, and could not travel across the Bitcoin network, because ordinary nodes classify it as nonstandard and refuse to pass it along. So it was hand delivered to a single mining pool through a private submission service, where it was included in a block as a favour.

The security overhead came to roughly two thousand percent of the value secured. The world's premier permissionless, censorship-resistant, trustless monetary network processed its first quantum-resistant payment by asking one specific company nicely.

I want to be completely clear that I think this is a genuine achievement and the people involved are not idiots. Avihu Levy at StarkWare has demonstrated that Bitcoin's existing consensus rules can accommodate quantum-resistant spending with no protocol change at all, which is a real result. He has described it himself as a last resort rather than a fix. The engineering is not the funny part.

The funny part is what it reveals about the timetable.

Google's researchers published an estimate in March that a sufficiently capable quantum computer could derive a Bitcoin private key nine to twelve minutes after the corresponding public key becomes visible. Which, if you have ever sent a Bitcoin transaction, is a number that should make you sit up, because your public key becomes visible the moment you broadcast. That confirmation window, the ten minutes you spend refreshing a block explorer, is the exposure. An attacker with that machine does not need to break Bitcoin. He needs to be faster than a block.

Everybody in this debate has been arguing about when that machine arrives. That is the wrong argument, and it is why the entire quantum conversation in crypto has been strategically useless for about four years.

Here is the right one.

Somewhere between three and four million bitcoin are sitting in addresses that will never move again. The earliest coins were paid to raw public keys rather than hashed addresses, which means their public keys have been sitting in plain view since 2009. Add the addresses that got reused, the wallets on the hard drive in the landfill, the seed phrases written on paper that went into a house fire, and the roughly one million coins attributed to Satoshi Nakamoto that have not moved in seventeen years.

Every credible model of Bitcoin's scarcity treats those coins as destroyed. Twenty-one million minus the lost pile. That subtraction is doing enormous work in every valuation framework I have ever been shown by someone at a dinner party.

Now consider that the coins most exposed to a quantum attack are precisely the coins nobody can move. A soft fork can protect you, if you are alive and paying attention and willing to shift your holdings to a new output type. BIP-360 proposes exactly that, introducing a new output format and removing the vulnerable spend path. Fine. Wonderful. Every active holder can migrate.

The dead cannot migrate. The landfill cannot migrate. Satoshi has been notably quiet on the subject.

So the first practical consequence of a working quantum computer is not that your coins get stolen. It is that several million coins that the entire market has written off as permanently destroyed become spendable by whoever owns the machine.

Bitcoin's scarcity has always rested on an assumption that lost is the same as burned. Quantum computing is the only known technology that turns lost back into found.

It is not a theft device. It is an un-losing machine, and it points at the most concentrated, longest-dormant, most psychologically load-bearing supply in the asset.

There is a further wrinkle that makes the coordination problem worse than the cryptography problem, and it is the part that should worry anyone modelling this seriously. Bitcoin cannot move quickly. A soft fork requires miner signalling, node adoption, wallet support and exchange integration, and the last contentious upgrade fight took years and produced a permanent schism. Taproot, which is uncontroversial and useful, took roughly four years from proposal to activation and is still not universally adopted.

So the sequence is not a computer arrives and Bitcoin patches. The sequence is a computer arrives, and Bitcoin begins a multi-year governance argument about how to patch, in public, while the vulnerable coins sit in plain view with their public keys exposed, and every participant in that argument holds a position in the asset being argued about.

The engineering has a solution. The politics has a solution. They are not on the same calendar.

Which brings me to a metal.

In the 1850s, aluminium was the most valuable metal on earth. Not because it was rare, it is the third most abundant element in the crust, more common than iron. It was expensive because nobody could get it out of the ore. The extraction was brutal, the yields were miserable, and the result was a metal that cost more than gold. Napoleon III reportedly served his most important guests on aluminium and everyone else on the gold plates, which is the single greatest flex in the history of tableware.

In 1884, the United States capped the Washington Monument with a hundred-ounce aluminium pyramid, chosen because the metal was as precious as silver and the monument deserved something extraordinary at the top.

In 1886, two chemists working independently on two continents cracked electrolytic smelting within months of each other.

Within a few years aluminium was cheap. Within a few decades it was foil. The capstone is still up there, incidentally, doing an excellent job of not corroding, worth about the price of lunch.

Aluminium's value was never a property of aluminium. It was a property of a computational problem nobody had solved yet. When somebody solved it, the value did not decline. It relocated, permanently, to a different thing entirely.

Three to four million lost bitcoin are currently locked behind a mathematics problem nobody has solved yet.

Now, positions, and I am going to start with the uncomfortable one.

This is not tradeable today. There is no quantum computer capable of this, the timeline estimates run from five years to never, and anyone selling you a quantum-panic Bitcoin trade this week is selling you a story. If I dressed this up as an actionable idea for tomorrow morning I would be doing the thing this letter exists to make fun of.

What is tradeable is the pricing of the tail, and the tail is genuinely mispriced, because the market has been pricing the wrong risk. The options market treats quantum as a security story, meaning idiosyncratic theft, low probability, bounded loss. It is a supply story. Those have completely different distributions. A theft event hurts a holder. A supply event reprices the asset.

So: long-dated out-of-the-money Bitcoin puts, twelve months and beyond, sized as convexity rather than a view, funded from a position you already hold rather than as a fresh directional bet. You are not predicting a quantum computer. You are observing that the strike structure prices a world in which the lost coins stay lost, and buying the version where they do not. The invalidation is a clean, activated soft fork with genuine migration underway, because at that point active holders are protected and the argument narrows to the dormant supply alone.

The second idea is the one nobody will write up, and it is my favourite thing in this story.

That transaction was nonstandard. The network would not relay it. It reached a block only because it went directly to a mining pool through a private submission channel, and MARA runs exactly such a service. If quantum-resistant transactions proliferate before a soft fork standardises them, and Levy's own framing suggests a long gap between the two, then every one of those transactions has to be hand-delivered. Direct submission stops being a curiosity and becomes a toll booth on a class of transactions that cannot use the public road.

That is a revenue line hiding inside a science experiment. Miners are currently valued almost entirely on hashrate and energy costs. Nobody is valuing the private mempool business, because until this week it was mostly used for large transfers wanting discretion. Own the miners with established direct-submission infrastructure rather than the pure hashrate plays. The thesis breaks if BIP-360 or something like it activates quickly and these transactions become standard, at which point the road reopens and the toll booth closes.

On the spot market, and here I have to be honest that the tape is not helping the bulls. Bitcoin has failed to reclaim $80,000. ETF inflows have slowed to $232 million. Yesterday I noted that the futures roll spread into September was the widest in three months and that the May peak of $82,807 was back in view, which was a reasonable read of the plumbing and has not worked. The roll spread told me fast money was positioned for a further leg. The absorption question is what actually matters and the absorption is not there.

That is two days running where I have been wrong about Bitcoin in two opposite directions, which is a good argument for me writing about bonds.

There is one more thing in this week's crypto news that deserves a mention purely for its comic timing. A lender has launched Bitcoin-backed mortgages. So on the same week that researchers demonstrated the collateral has a live, published, credible cryptographic tail risk requiring a network-wide protocol change, somebody started accepting it as security against a house. I have watched a great many things get accepted as collateral in twenty-seven years, and the pattern is unfailing. The collateral is always safe, right up until the moment somebody needs to sell it, and then it turns out the reason it was cheap was written down somewhere all along.

The genuine takeaway, if you want one you can use next week on something that is not Bitcoin, is this. Whenever you see an asset whose value depends on a quantity being permanently unavailable, ask what makes it unavailable. If the answer is a law, laws change. If the answer is geography, transport improves. If the answer is a computational problem, then you do not own a scarce asset. You own a bet on a research programme, and you are short the people trying to solve it.

Napoleon III's dinner guests genuinely believed they were eating off the most precious metal in the world. They were right, for about thirty years. The capstone on the Washington Monument was chosen because it was worth more than silver.

It is still up there. Nobody has climbed up to get it.

On Monday, Citadel Securities publicly described the Treasury's bond buyback programme as financial repression.

On Tuesday, Citadel Securities' head of macro strategy published a note saying the asymmetry is now skewed toward lower long-end yields and that he had turned bullish.

The same firm. Two days apart. Denounce the manipulation on Monday, position for it on Tuesday. And I want to be clear that this is not hypocrisy, it is professionalism. Complaining is free and being short into a trillion dollar buyer is not.

The entire market has arrived at the same place. Nobody thinks this works. Everybody is buying it anyway.

Which brings me to George Cassiday, who wore a green hat.

During Prohibition, Cassiday ran liquor to the United States Congress out of an office in the House Office Building. He had a room. He had regular clients. He operated for five years before anyone bothered to arrest him, and when he finally wrote it all up for the Washington Post in 1930 he estimated he had served roughly four out of five members of Congress. These were the men who had voted the country dry, buying gin from a man with a desk down the hall.

Nobody in that building thought Prohibition worked either. It just would have been extremely inconvenient to say so out loud while the vote was being counted.

That is where the Treasury market is this morning. Everyone has filed their objection for the record and everyone is long.

The evidence that the Bessent put is working is not in the yield, it is in the plumbing. Since the announcement, Treasuries have outperformed equivalent-maturity swaps, narrowing the thirty-year spread between the two to its tightest since February. The ten year swap spread has compressed by three basis points to around 38. The call-to-put skew on long bond futures has run up sharply while the front end has not moved at all, which tells you exactly where people think the intervention lands. One derivatives broker described the current fear in the market as the fear that long rates might plunge, which is a sentence that would have gotten you laughed at three weeks ago.

Then Stanley Druckenmiller wrote an op-ed saying the whole thing is a mistake.

I want to sit with that for a second, because the comedy here is genuinely first-rate. Bessent learned his trade at Soros Fund Management under Druckenmiller. The firm's signature achievement, the one that put both of them in the textbooks, was betting against the Bank of England in 1992 when it tried to defend a level it could not defend. Bessent is now the official defending the level. And his old boss has gone to print to explain, in public, using the reasoning he presumably taught Bessent in the first place, that the position is wrong.

That is not a policy debate. That is a man watching his former student attempt a trade he personally spent a career on the other side of, and choosing to say so in a newspaper.

Druckenmiller's argument is the uncomfortable one. An economy at essentially full employment, running deficits at 6% of GDP, with a ten-year yield at or below nominal growth, is an economy where policy is already accommodative. Buybacks do not change any of that. They change the price at which you find out.

Now, my own ledger, and this one is not flattering.

Yesterday I said do not be short the long end into 9 September. That call is working. Ten-year yields fell seven basis points on Tuesday to 4.63%, thirty-year is at 5.17%, down from 5.34% last week. I would love to tell you the buyback thesis did that.

It did not. Oil did that. WTI is down roughly 5% on the week and Brent is below $87 after Iran and Oman met to discuss a temporary joint maritime corridor through the Strait of Hormuz. Cheaper crude eased the inflation pressure, softer consumer confidence at 89.4 helped, and the buyback signal stopped the bleeding at the margin. The ten-year is sitting exactly where it has spent most of the past month. Right call, wrong engine, and if I do not say that here then the next time I am right for the right reason you have no way of telling the difference.

The Canadian dollar short from Sunday is worse. I sold it at 1.3775 arguing the asymmetry was overwhelming. Carney has since matched Trump's tariffs with 15% to 50% levies on about $20 billion of American goods, doubling steel to 50%, and the loonie went up and Canadian bonds got bid. The market's read, which I now think is correct, is that the affected goods amount to roughly 0.6% of total US imports and about 0.3% of US goods consumption, and about 5% of Canada's shipments south. Painful for Canada, invisible to America, and not large enough to move a currency. Canadian high-grade credit is actually trading at below-average spreads on a leverage-adjusted basis while US credit sits in line with its two-year average.

I was right about the asymmetry and wrong about whether this particular fight is big enough to express it. Cut the position by half, keep the 1.3650 stop on the rest. The energy line is the one neither side will cross, and until somebody does, this is a tariff dispute about cheese being priced like a tariff dispute about cheese.

Here is the trade I actually want you to take away today, and it is the strangest setup I have seen in the oil market in years.

For four decades, closing the Strait of Hormuz was the oil market's doomsday scenario. The thing you bought calls against. The tail.

The tail has now happened. And the options market has responded by refusing to pay for upside.

The call-to-put open interest ratio in Brent has averaged 1.27 since 2021. During the Russian invasion of Ukraine it peaked at 1.72. It recently hit a ten-year low of about 0.7 during the Iran war, which is a substantially larger physical supply disruption. Put open interest is piled across the $65 to $90 area. Upside exposure sits in thin clusters at $100, $110 and above.

The market is not underpricing this because it is complacent. It is underpricing it because it has run out of imagination. Hormuz was the last item on the list of things to be frightened of, and once you have used your worst-case scenario, there is nothing left to escalate into, so nobody knows what to pay for the next tail. Add an administration whose entire posture has raised the cost of holding bullish oil exposure, and open interest has drained out of flat price altogether.

Buy the upside. Brent calls or call spreads in the $100 to $110 strike cluster, three to six months out, sized as convexity rather than a directional view, because those strikes are the cheapest they have been relative to realised risk since before the war started. The invalidation is a signed and implemented Hormuz transit corridor with Omani participation actually moving cargoes, not a foreign ministry statement that the two sides met and agreed reopening would be nice.

Which is what happened yesterday, and which knocked two percent off crude.

Stay long the diesel cracks, which I flagged on the 19th and which have now survived their first real test. The flush this week was severe. Open interest collapsed, implied volatility retreated, call selling accelerated, spec accounts dumped risk. And the cracks held. That is the tell. When positioning gets destroyed and the spread does not follow, the tightness is physical rather than speculative, and Russian diesel exports have not recovered. Add on weakness rather than cutting into it.

Nvidia reports after the close tonight. Analysts expect revenue to have nearly doubled to $92 billion for the quarter, which is more than any of its competitors generate in a full year. One strategist summarised the setup with a line I would put on a wall: we are anticipating good news here, but so is everybody.

The ratio of semiconductors to equal-weight technology has now underperformed for more than two months and has broken below its uptrend from 2025. The stock snapped a seven-day losing streak into the print. Do not own this for the earnings. The number will be enormous and the reaction will be about 2027, and 2027 is where analysts have spent the past month quietly trimming.

Korea is quietly doing something that should not be happening. The won is the best-performing major currency quarter-to-date while Samsung and SK Hynix stay soft, which breaks the emerging-market rule that equities and currency fall together. That decoupling is a structural positive for Korean assets and it is also the reason my SK Hynix position, still open and still underwater, is not a currency problem on top of an equity one. The Bank of Korea decides on Thursday with a slim majority expecting a second consecutive hike, and won strength has already done some of that tightening for them. If they hike and soften the guidance, the Korean curve steepens, which is the cleaner expression than the equity right now.

I also spent yesterday mocking the argument that Bitcoin should be bought because its relative strength index was at the ninety-ninth percentile of the past decade. It then went up. It is $78,400, the futures roll spread into September is the widest in three months, and the May peak of $82,807 is back in view. Mocking a signal does not make it wrong, and I would rather record that than pretend I said something cleverer.

Elsewhere, OpenAI says its in-house chip, which is called Jalapeno, outperformed Nvidia's current lineup in testing. SpaceX will build a $100 billion Starship launch site on the Louisiana coast. Jack Ma has been buying Alibaba stock. And the Treasury's new Iran sanctions campaign has been formally named Operation Economic Outcast, which sounds less like a national security programme than something that happens to you in year nine.

China Daily published an editorial arguing that the United States is trapped in a crisis of its own making, citing gasoline above $4 a gallon and national debt through $40 trillion, and suggesting Washington is trying to look stronger by threatening a foreign adversary. When the state media of a one-party system is running better macro commentary than half the sell-side, something has gone sideways in the discourse.

Bessent, for his part, told reporters he does not want to blow up the global financial system. I believe him. I would simply note that this is not usually a sentence that requires saying.

Three things land this week. Nvidia tonight, PCE on Wednesday, and Warsh's first Jackson Hole speech as chairman on Friday, from a man whose stated governing philosophy is talking less. The market is holding its breath for a speech by someone who has spent a year explaining he would rather not give one.

Congress voted dry and kept a bootlegger in the building. The market has called this repression and gone long. Everybody knows. Everybody is in.

The only question worth asking is who gets caught holding when somebody finally decides to enforce the rule.

Good morning all,

Another Tuesday or Chewsday as my Cockney mates may say.

Here we are.. Speculative positioning in ten year Treasury futures currently sits at the second percentile of its historical range. Not the twentieth. The second. Which means that on a scale of one to a hundred measuring how many people are betting against US government bonds, we are at two, and the only reason it is not one is that somebody somewhere has not updated their file.

The US Treasury is sitting on a cash account of nearly one trillion dollars, rebuilt from about $250 billion in June of last year. That ratio of cash to bank reserves sits near the ninetieth percentile of the past decade once you strip out the covid distortion.

So we have a record crowded short base on one side, and on the other, a trillion dollars of dry powder controlled by a man who spent his formative professional years at Soros Fund Management learning precisely how to detect, isolate and financially disassemble a crowded position.

On Monday, CNBC reported the Treasury is considering using that cash pile to fund bond buybacks. Thirty-year yields fell immediately.

That is not debt management. That is a raid.

I called this futile on Thursday and I want to correct myself, because the arithmetic has changed and pretending otherwise would be dishonest. My argument was that $32 billion a quarter against a $31.5 trillion market is one tenth of one percent, roughly the market impact of a strongly worded letter. That remains true of the size. It is not true of the mechanism.

Funding buybacks by issuing Treasury bills is a two-sided trade, because you remove long duration and add short supply at the same time, and the market nets them out. Funding buybacks from a cash account you already have removes the offset entirely. Duration leaves the market first. The bills that eventually rebuild the cash pile arrive later, after the long end has already tightened. That gap, however brief, is the whole trade, and if you are short thirty-year bonds inside that window you are not participating in the bond market, you are the raw material.

Which brings me to professional wrestling.

The thing that makes wrestling work is kayfabe, the shared agreement that the contest is real. Everyone in the building knows the promoter decides the outcome. Everyone also agrees, for the duration of the evening, to behave as though he does not, because the alternative is watching two men in swimwear hug for forty minutes. The system functions perfectly right up until the promoter stops pretending, walks into his own ring, and starts throwing people over the ropes himself. At which point the audience does not applaud. It asks for its money back.

The Treasury market has run on kayfabe since about 1979. The government issues, the market prices, the government pretends it has no view on the price, and the market pretends to believe it. Both parties know the government would prefer lower yields. Neither says so.

Last week the promoter climbed into the ring.

Citadel Securities used the phrase financial repression to describe this in public, which from a firm that makes its living inside Treasury market microstructure is roughly the equivalent of your own dealer describing you as a person of interest.

So position accordingly, and by that I do not mean sell bonds.

Do not be short the long end into 9 September. I say that as someone who has been structurally bearish duration for a month and who has just watched a positioning statistic that would embarrass a meme stock collide with a trillion dollars of committed capital. The 4.75% level in tens is now visibly the pain threshold at which somebody in Washington starts making phone calls. Getting run over by a sovereign with a cash account is a bad way to be right.

The trade is the breakeven, and this is the cleanest idea I have seen in weeks.

Thirty-year inflation compensation, the gap between nominal yields and inflation-protected yields, has traded in a band between 2% and 2.4% since 2023. It is currently 2.255%, up about ten basis points since late July. That band is the Federal Reserve's credibility, expressed as a number.

Here is the mechanism nobody is pricing. Drawing down the Treasury's cash account does not just buy bonds. It pushes reserves into the banking system, because that money has to land somewhere. So the operation suppresses the supply component of real yields while simultaneously adding liquidity, and liquidity is the exact channel through which inflation compensation rises. The policy is designed to lower one number and structurally lifts the other.

Buy thirty-year breakevens. Long TIPS against nominals, target 2.50%, which is where a technical supply operation gets rebranded in the market's head as an inflationary one, and where a genuinely different class of seller arrives. Above that, the next level is 2.70%, the April 2022 peak reached during the worst inflation in forty years. Invalidation is a close back below 2.15%, which would tell you the credibility band is holding and I am early. Horizon runs through the September operations.

Gold, since I keep a ledger, is the position I recommended at $4,415 on the 18th while looking rather stupid at $4,339 two days later. It is now $4,648. Global money supply is above $123 trillion and gold's twelve-month correlation to it has repaired to 0.47 from roughly 0.2 earlier this year. The stop never triggered and the thesis is now doing exactly what it said on the tin, which I mention mainly because I have spent two weeks reporting the losing half of this ledger and would like one paragraph of dignity.

Elsewhere in things that are technically happening, Nvidia has fallen for seven consecutive sessions, its longest losing streak since 2022, into earnings on Wednesday where analysts expect 99% year-on-year profit growth.

Ninety-nine percent. Almost double. And the stock's price relative to expected profits is merely in line with the S&P 500, which means the market has looked at a company doubling its earnings and concluded it will grow about as fast as a supermarket chain from next year onward.

The reason is the funniest supply chain development of the year. Nvidia has been raising prices roughly 15% to 17%. The memory makers who sell Nvidia the components have been raising theirs by more. The most powerful company in the world, sitting on a 90% share of a trillion-dollar market and a 75% gross margin, is being quietly squeezed by its own suppliers, and there is nothing it can do about it, because you cannot conjure a memory fab out of enthusiasm. It takes years and billions and the queue is already full.

The memory stocks fell anyway. Sandisk, Seagate, Micron and Western Digital all dropped more than 5% on Monday while their pricing power was actively improving, and Micron is forecast to generate close to $50 billion of free cash flow this fiscal year. My SK Hynix position remains open and underwater and I am not adding here, because a memory stock that falls on good news is telling you the marginal holder is a momentum fund rather than a fundamental one, and momentum funds do not read cash flow statements.

Bitcoin has run 25% in a week to test $80,000, with a fourteen-day relative strength index at 85, which is the ninety-ninth percentile of the past decade. The published guidance on this is that historically, after readings this extreme, prices have gone up. So the recommendation is to buy an asset because it is the most overbought it has been in ten years. I want that framed. Every risk manager I have ever worked with is currently screaming into a pillow.

The oil options market has arranged itself into something more actionable. Dealer positioning in the US Oil Fund has slipped from stabilising into fragile, with light gamma between spot and the large upside strikes, meaning hedging flows will now amplify moves rather than dampen them. Energy equities are the mirror image, with dealer gamma positive across maturities and structural protection further out. Same underlying commodity, opposite plumbing. If you want exposure to a Middle East disruption, own crude volatility and not energy equity volatility, because a shock transmits straight into the barrel and gets absorbed by the equity.

Which matters, because shipping a supertanker of oil through the Strait of Hormuz now costs about $20 million, and Asian refiners are set to nearly double their purchases of American crude in September to more than 40 million barrels. American oil is being loaded onto ships bound for Asia while American drivers pay record prices at the pump. That is not a conspiracy, it is just what a global market in a fungible commodity looks like, and it is going to be explained badly by everybody for the next six weeks.

The Americas trade split is now clean enough to trade directly. Mexico's president expects a deal with Washington, second-quarter GDP grew 1.4%, Banxico is comfortably parked at 6.5%, and the peso has the best carry in the region. Canada announces retaliatory tariffs today, its front-end yields are already nine basis points lower as growth gets repriced, and the President and the Premier of Ontario spent the weekend exchanging the words flunky and loser. Long peso against the Canadian dollar. The rate differential pays you to wait and the political trajectories are diverging in real time.

Two more, quickly, because they are the sort of thing that ends up mattering in a year. Taiwanese prosecutors have indicted a senior Nvidia manager over an alleged scheme to smuggle advanced AI chips into China via Japan and Indonesia. And Chinese state-linked hackers are reportedly using DeepSeek to conduct attacks, which means the model China gave away for free is now being used to attack the country that spent $744 billion trying to keep it scarce.

Saudi Arabia is building a €6 billion manga theme park outside Paris, championed by Emmanuel Macron, because he and the Crown Prince discovered a shared enthusiasm for Japanese comics. The proposed fee for a new H-1B visa sponsorship is $103,000. SpaceX will launch AI satellites powered by Nvidia chips in late 2027, which means the data centre capex debate is about to acquire an orbital component. And the global average sea surface temperature hit a record 21.1 degrees on 22 August.

The promoter is in the ring. The shorts are at the second percentile. There is a trillion dollars of cash pointed at them and a date circled on the calendar.

Just remember what happens to kayfabe once the audience works out that the fix is real. They do not stop watching. They stop paying.

So this one is not academic for me. I know people who work in those plants. I have made the arguments in this letter at dinner tables where they were not welcome. And I have never once enjoyed being proved right about it.

Let's be clear, today is Not about sovereignty. Not the Arctic. Not the rules based international order our Prime Minister pronounced dead from a stage in Davos in January, which was a bold thing to announce about an order that had been paying our bills since 1945. The 50% tariffs, the counter tariffs, the war footing, the whole magnificent bonfire came down to whether an F-350 counts as a car.

Both sides had a deal. Auto tariffs were coming down from 25% to 15%. Canada asked, late on Friday night, on a phone call, that the relief extend to medium and heavy duty vehicles, because Ford's new plant in suburban Toronto builds F-350s and larger. Washington called it a new demand. Ottawa said it obviously was not. And somewhere around midnight, two countries that exchanged nearly $900 billion of goods and services last year decided they would rather have a trade war than a conversation about trucks.

By Saturday, Mark Carney was in front of the cameras saying, and I want to quote this exactly, "You're at war when you get attacked. We got attacked."

This from the man who spent his first year in office rolling back Justin Trudeau's retaliatory tariffs, axing the digital services tax that annoyed Trump, and re-cutting the Detroit-Windsor bridge deal in America's favour. Three unilateral concessions. Nothing measurable came back for any of them. That is not a negotiation, that is a tip jar.

And then it ended at midnight on a phone call because nobody had established, in writing, weeks earlier, whether a truck is a car.

Consider what we were sold. Two central bank governorships. Threadneedle Street through Brexit. A man whose entire professional identity rests on precision, forward guidance and never being caught out by a definitional ambiguity, hired at a premium because he was the adult who could handle Trump. The pitch was competence. What we got was a product classification dispute discovered at midnight.

I have watched Canadian trade policy for my entire adult life and I have never seen anyone go from elbows up to elbows down to elbows up again this quickly. It is less a negotiating strategy than a form of interpretive dance.

In fairness to the man, the arithmetic he is working with is genuinely grim.

The US imposed 50% tariffs on roughly $20 billion of Canadian goods on Saturday under Section 338 of the Tariff Act of 1930, a provision unused for ninety-six years that requires only a finding that a country is discriminating against American commerce. The stated justification is our treatment of American dairy, alcohol and cars. Bloomberg Economics ran the numbers. It lifts the average effective US tariff on Canada by 2.5 percentage points, to 8.4%.

It lifts the overall average effective US tariff rate, across all countries, by three tenths of one percentage point.

That is the whole relationship in one statistic. Canada is just under 5% of American imports by value. Of the goods targeted, more than 80% go to the United States, and for many the figure is close to 100%. They can find plywood somewhere else by Thursday. We cannot find anyone else to sell it to at all.

Pierre Trudeau put it better than any economist ever has, in Washington, in 1969. Living next to you, he told the assembled American press, is in some ways like sleeping with an elephant. No matter how friendly and even-tempered the beast, one is affected by every twitch and grunt.

He was right, which is the infuriating part, because he then spent a decade proving that diagnosing the elephant and managing it are entirely different skills.

In 1972 his government launched the Third Option, a formal national strategy to reduce Canadian dependence on the United States by diversifying toward Europe and Japan. White papers. Speeches. A contractual link to the European Community, signed with great ceremony in 1976.

By the time it was quietly abandoned, the American share of Canadian exports had gone up.

Not sideways. Up. It kept climbing for another twenty-five years, through the 1988 Free Trade Agreement and NAFTA in 1994, until four of every five dollars we earned abroad came from one country. The Third Option achieved the precise opposite of its stated goal and cost a fortune in conference catering.

This is not a Canadian failing, it is a law of physics that keeps getting mistaken for a policy choice. In November 1806 Napoleon issued the Berlin Decree, closing the entire European continent to British trade, on the theory that a nation of shopkeepers could be strangled by paperwork. The most ambitious boycott ever attempted, enforced by the largest army in the world. It failed. Smuggling exploded, because demand does not read legislation. The countries doing the boycotting suffered more than the one being boycotted. Russia gave up in 1810, which is a substantial part of why Napoleon marched on Moscow two years later, which is a substantial part of why there is no French Empire.

Geography is not a negotiating position. It is the table everyone else is already sitting at.

I mention this because in January our current Prime Minister told Davos that if you are not at the table, you are on the menu. He has since signed a limited trade deal with China and pledged to quadruple defence spending over a decade. It is the Third Option with better tailoring.

Diversification is a decade-long infrastructure problem wearing a press release. You do not diversify away from the elephant by announcing that you have stopped liking the elephant. You do it by building pipelines, ports, rail and terminals that point somewhere other than south, and Canada has spent thirty years demonstrating a magnificent inability to build any of those things on time, on budget, or at all. More on that shortly. It is the most expensive own goal in our modern economic history.

Which brings me to softwood lumber, the dispute that explains everything and that nobody outside Canada has heard of.

We have been fighting the Americans over softwood since 1982, in numbered rounds, like a boxing franchise nobody asked for. Lumber I. Lumber II. Lumber III. Lumber IV. Canada took the case to panel after panel under the very agreements we signed so disputes like this could be settled by rules rather than power. Canada won. Repeatedly. On the merits.

The Americans kept collecting the duties anyway.

By 2006 they had taken more than $5 billion. The settlement returned about $4 billion, meaning the United States kept roughly a billion dollars that a series of tribunals had ruled it was not entitled to. Canada signed the paper. Everyone shook hands. The dispute resumed almost immediately and is still running today.

My favourite detail is the Byrd Amendment, passed by Congress in 2000, which took the duties collected from foreign exporters and handed the money to the American companies that had filed the complaint. Canadian sawmills paid. The money went to the American sawmills that wanted them gone. The World Trade Organization ruled it illegal and Congress repealed it in 2006, with a two-year grace period during which it carried on operating, because naturally it did.

There is a lesson in there that every Canadian government since Mulroney has refused to learn, and I will say it flatly. In a dispute between a country with leverage and a country with a legal argument, the legal argument is a hobby.

Jean Chretien campaigned in 1993 on renegotiating NAFTA and scrapping the GST. He won, obtained two side letters of decorative value, signed the agreement and kept the tax. Justin Trudeau announced in 2018 that Canada would not be pushed around, was called dishonest and weak by the American president after a G7 summit, and conceded on dairy anyway.

Campaign against the deal, negotiate, concede on dairy, declare victory, schedule the next round. It is less a policy than a liturgy.

Which is why the current strategy interests me, because for the first time in my career somebody in Ottawa has noticed where the actual leverage is, and it is not in the tribunals.

Carney said it out loud on Saturday. Canada supplies 99% of American natural gas imports, 85% of electricity imports and 60% of crude oil imports, and he did not think they wanted us to stop sending it. Doug Ford, emerging from several days of uncharacteristic silence, went further and proposed targeting products from politically significant states, naming Florida, Texas, Iowa and Wisconsin. That is the most sophisticated thing Doug Ford has ever said and I would like it noted that I am as surprised as anyone.

The Leger polling says Canadians agree. Fifty-six percent want no more concessions. Seventy-four percent support a special tax on electricity sold to the US, seventy percent support taxing oil and gas exports, and sixty-four percent want American liquor banned outright, which tells you how seriously we take this and how much we like Ontario wine.

And forty-seven percent think there should be a federal election if a deal gets signed, rising to sixty-seven percent among Conservative voters. That number is not about trade. That is a country asking to see the receipts.

Here is my problem with all of it, and it is the problem I have had with every Canadian retaliation since 1982.

Counter-tariffs on $20 billion of American steel, dairy, appliances, farm equipment, electronics and pulp are a tax on Canadians. That is what a tariff is. We import half our goods from the United States. When Ottawa taxes American agricultural equipment, the person who pays is a farmer in Saskatchewan who needs a combine and has precisely one place to buy one. Carney conceded this himself in the same breath, saying the measures would raise costs and reduce choice for Canadians. He said it reluctantly. The combine costs the same either way.

Energy export taxes are different. Energy export taxes are a genuine weapon, because 99% of anything is not a supply chain, it is a hostage situation.

The trouble is that the hostage lives in Alberta, and there is a reason Alberta is not returning Ottawa's calls.

In October 1980, Pierre Trudeau's government introduced the National Energy Program, which set a made-in-Canada oil price well below the world price, taxed petroleum revenues, and redirected the difference eastward. Estimates of what it moved out of Alberta run from $50 billion to well past $100 billion, depending on who is counting and how furious they are. Peter Lougheed cut provincial production in response. The bumper stickers suggested eastern Canadians might freeze in the dark, and that was not fringe sentiment, it was the provincial mood.

Forty-six years later, a federal Liberal government is asking Alberta to accept a federal tax on its energy exports for the good of Confederation. You can see the difficulty.

That would be awkward enough if the intervening decades had gone well. The Trudeau government that preceded this one cancelled Energy East in 2017, passed Bill C-69 and the C-48 tanker ban, watched Kinder Morgan walk away from Trans Mountain, then bought the pipeline itself for $4.5 billion because no private operator would touch a project inside the regulatory regime that same government had just built. Final cost, roughly $34 billion. Twelve years. The Panama Canal took ten, and they had malaria.

So: fifty-six percent of Canadians want a hard line, only 46% of Albertans do, and Alberta holds the gun. A decade was spent making it structurally impossible to sell our energy to anyone except the country now tariffing us, and the plan is to threaten that country with the energy we can only sell to them.

So the country with the leverage cannot politically use it, and the tools it can use hurt itself first. That is not a strategy. That is a hostage situation in which the hostage taker has agreed, in advance, to shoot himself in the leg on a published schedule.

Now, positions, because none of the above is a reason to lose money.

The Canadian dollar is around 1.3775, near a three-month high, having strengthened roughly 2.5% over the past month on weak US data and firm crude. A currency at a three-month high heading into a trade war its country cannot win on arithmetic is a gift. Sell it here, target a retest of the 1.4248 mid-June high, stop above 1.3650 on a weekly close, which would tell you the market has decided the tariffs are noise and I am wrong about the asymmetry. The Bank of Canada warned last year that a prolonged tariff war could cut Canadian output by nearly 3% over two years and wipe out growth entirely, and that warning described a smaller fight than this one. Horizon three to six months, spanning the September 8 counter-tariff date and the USMCA reviews Trump declined to renew in July.

The second is potash, and it is a position to build rather than a trade to put on. Sixty-three percent of Canadians favour taxing potash exports, and Canada supplies roughly a third of world production into an American market with no meaningful domestic alternative. Any serious move toward export taxes is a supply shock to global fertiliser prices arriving with almost no warning. Own the producers rather than the commodity, because an equity gives you a business underneath the option. This is a position for the six months in which somebody in Ottawa finally works out that dairy quotas are not leverage and potash is.

The third is the crude differential. If energy export taxes move from polling question to policy, Western Canadian Select widens against WTI immediately and violently, because those barrels have nowhere else to go on the timescale that matters. Long WTI against short WCS is the cleanest expression of Canadian political dysfunction available in a listed instrument, and I say that with affection for the country and none whatsoever for the people who built the constraint.

Ledger, since I keep one. The rates volatility position from earlier in the week is working. Gold, recommended Tuesday at $4,415, has climbed back to a three-month high with the stop never hit, so that one has gone from embarrassing to fine, which in this business counts as brilliant if you say it confidently enough. SK Hynix remains open and underwater.

The elephant has not changed since 1969. It still twitches. It still grunts. What has changed is that we have had fifty-seven years to build a second bed, three national strategies announcing our intention to do so, one program that took a hundred billion dollars out of the only province that could have funded it, and a Prime Minister with two central bank governorships explaining from Davos that the mattress is philosophically dead.

We are still in the bed. The elephant just rolled over about pickup trucks.

Good morning indeed..

Last night was a typical night in Trumps never ending market of surprises… Overnight, here in Singapore, the US Treasury offered to buy back $2 billion of its own old bonds. It received roughly $20 billion of offers.

This has been reported, in several places, as evidence of enthusiasm for the buyback programme.

I want to be precise about what a tent imes oversubscribed buyback actually measures. It measures how many people want to sell. The Treasury put its hand up and said it would purchase two billion dollars of long-dated government debt, and ten times that much government debt sprinted across the room shouting take me. That is not a vote of confidence in the programme. That is a fire exit with a queue.

And on Wednesday, the Treasury announced it would at least double the size of these operations, bringing quarterly purchases to roughly $32 billion, beginning 9 September.

So the government has now announced, three weeks in advance, in writing, that it will be a large and committed buyer of a specific and publicly identifiable set of bonds. If you have ever wondered what it looks like when the world's biggest borrower volunteers to be the world's most predictable counterparty, you are watching it.

Here is the everyday version. You are at a party. You stand on a chair and announce that in three weeks you will be buying used furniture, at good prices, in quantity, and you name the exact furniture you want. Then you sit down and enjoy the rest of the party.

Two things happen next. Every stick of used furniture in the neighbourhood begins moving toward your house. And nobody sells any of it to anybody else in the meantime, because why would they, when the guy on the chair has pre-committed to a bid.

Thirty-year yields fell as much as 10 basis points to 5.18% on the news, away from Monday's highest level since 2007. The dollar dropped to its weakest in three months. The market, briefly, was delighted.

The comparison everybody reached for this week was King Canute ordering the tide to stop, and I would like to file a small objection, because that story gets told backwards roughly nine times out of ten. Canute knew perfectly well the tide would not stop. He had his throne carried to the water's edge specifically to demonstrate to his flattering courtiers that royal power had limits. He was doing performance art about humility. He got soaked on purpose, and it worked.

If you want a monarch who genuinely believed he could win an argument with a body of water, you want Xerxes.

In 480 BC, Xerxes built two pontoon bridges across the Hellespont so his army could cross into Europe. A storm destroyed them. Xerxes, being an emperor, responded in the manner emperors respond, which is to say he had the engineers beheaded and then turned his attention to the water itself. He ordered the strait given three hundred lashes with a whip. He had a pair of fetters thrown into it, so it would understand it was a prisoner. Some accounts say he had it branded with hot irons. And he had his men shout at the sea while they did it, a formal denunciation, informing the Hellespont that it was a treacherous and briny river and that King Xerxes would cross it whether it liked it or not.

He did cross it, incidentally. They built new bridges. The whipping contributed nothing.

The Treasury is going to buy $32 billion of bonds a quarter out of a market that contains $31.5 trillion of them. That is roughly one tenth of one percent. Last week, the same Treasury paid out about $85 billion in interest to bondholders, the largest single sum in Bloomberg's records. So the buyback programme, at its newly doubled and much-heralded size, will purchase over three months a little more than a third of what the government pays out in coupons in a week.

Three hundred lashes. Very stern. Very well publicised. The tide is not consulting the schedule.

Nobody has explained how the buybacks get paid for, but the Treasury funds this sort of thing by issuing Treasury bills, which mature inside a year. So the plan is to retire long-dated debt by issuing short-dated debt. Deutsche Bank's head of FX research called it what it is, an Operation Twist, and added the phrase soft-form financial repression, which is a wonderfully polite way of describing a government leaning on the shape of its own yield curve because it does not care for the answer the curve is giving.

The household version is simpler. You have a thirty-year mortgage at a rate you dislike. Your solution is to pay it down using your credit card. Your monthly interest cost may well drop. You have not reduced what you owe by a single dollar. You have made the entire liability reprice every four weeks instead of every three decades, and you have done it because you have an important social event in November and you would like the numbers to look better before it.

The important social event is on 3 November, and the polls are not encouraging. Mortgage rates are back around 7%, which is a genuine problem for an administration that campaigned on making borrowing cheaper. A 10-year auction last week drew the highest financing cost at that tenor since 2007, and the 30-year sale a day later cleared at the greatest yield since 2001. The deficit for 2026, with two months of the fiscal year to run, stands at $1.8 trillion, 5% wider than last year, propelled by Social Security, Medicare, Medicaid and, with beautiful circularity, interest on the debt.

The part I enjoy most is the paperwork.

Bessent's predecessor did a version of this in 2023, tilting issuance to keep yields down ahead of an election. Bessent, then a private citizen with opinions, criticised it. The administration's former chief economist co-wrote a paper in July 2024 attacking exactly this behaviour, gave it the name activist Treasury issuance, and warned, and I want you to appreciate the prescience here, that once one political party starts using it to stimulate the economy into election season, it may be used repeatedly by all future administrations.

The paper was correct. It took about two years and one change of desk. The author subsequently joined the administration.

There is also the small matter of the Federal Reserve chairman, who spent last month enthusing about markets being liberated from forward guidance, saying investors were learning to play the ball and not the referee. Three weeks later the Treasury Secretary announced he would be buying the ball. Between a central bank that has stopped explaining itself and a Treasury that has started intervening, the referee has left the pitch and is now trading the game from a laptop in the car park.

None of this is stupid, and this is the part that matters. Bessent spent decades at a hedge fund, where he became famous for enormous, correct, high-conviction bets against the pound and the yen. One portfolio manager described Wednesday's move as the lift everything on the screen trick, the old technique of hitting every dealer simultaneously to force the market to move. That is exactly what it is. He is doing the thing that worked brilliantly for twenty years, with a bigger balance sheet.

The difficulty is that when Soros broke the Bank of England, he was on the side of the fundamentals and the central bank was defending a level. Bessent is now the one defending the level. He has taken the losing side of his own signature trade, at scale, using public money, with an audience.

He is also fighting the wrong ocean. Japanese 10-year yields are at roughly their highest in thirty years, inflation has returned to Japan after a quarter century away, and the Bank of Japan is expected to resume raising rates by October. Japanese government bonds compete directly with Treasuries for the same global savings, and that repricing is an earthquake happening under a market fifty times the size of the intervention. Alphabet sold $25 billion of bonds this month with maturities out to forty years. Amazon and SpaceX have been issuing long-dated paper too. There is a global queue for long-duration money and the Treasury has joined it holding a whip.

So, positions.

The cleanest trade on the screen is one the Treasury has helpfully published for you. Buybacks target off-the-run securities, the older, less liquid bonds that trade at a discount to current benchmarks precisely because nobody wants them. The government has now announced it will show up on 9 September and want them, at double the previous size, in the 10-year to 30-year sector. Buy the off-the-runs and sell the on-the-runs against them. You are not forecasting anything. You are standing between a discount and a pre-announced, publicly scheduled, government-sized bid for the specific bonds carrying that discount. The trade is invalidated if the Treasury quietly reduces operation sizes or shifts the tenor buckets, which is why you read the tentative calendar rather than the headlines, and the position closes as the operations run through early November.

The second is to fade the rally in the long end rather than chase it. Wednesday bought roughly 10 basis points at the 30-year. It did not buy a single fundamental. The deficit is unchanged, Japan is unchanged, the AI issuance calendar is unchanged, and the war is unchanged. Sell the 30-year into strength below 5.20%, look for a retest of the 5.40% to 5.44% zone that marks the 2007 peaks, and accept that you are invalidated by two things and only two things, a genuine resolution of the Iran conflict or an economy that slows properly. A press release is not a fundamental.

My rates volatility call from yesterday, for the ledger, worked, and I want to be careful about why. I said buy September straddles on ten-year futures with the MOVE index at the twentieth percentile because volatility was too cheap for the amount of event risk in the calendar. A surprise Treasury announcement moving the long bond 10 basis points in a session is precisely the kind of thing that was not priced. The position is up and the reasoning held. That is one for three this week, and the other two, gold and SK Hynix, remain open and losing with their stops intact. I would rather show you the whole card than the good hand.

The dollar is the quiet one to watch. It fell to a three-month low on Wednesday, which is what happens when the market reads soft-form financial repression correctly on the first attempt. If the plan is to hold long yields down while the deficit runs at $1.8 trillion and the Fed declines to explain itself, the adjustment does not vanish. It moves to the exchange rate. That is not a trade for this week. It is the reason your unhedged dollar assets deserve a conversation this quarter.

Xerxes did get his army across the Hellespont. He lost the war anyway, at Salamis, in the water.

The tide has never once lost one of these arguments.

Finally the heat in the lion city has subsided… It's funny how it is coming at the same time the market is cooling off a bit… coincidence? Maybe…

Anyways, two of the largest private companies in the world are racing each other to sell shares to the public, and according to the reporting, the reason they are in a hurry is that they want the deals done before the shine comes off artificial intelligence.

That is not a short seller talking. That is the explanation, in print, for why the deals are being rushed.

Anthropic's listing is expected within weeks. OpenAI is racing it to market. Both are reportedly selling cheaper models now to compete with Chinese rivals, which will compress margins. And the stated logic for listing this month, in a thin August tape, is to get the money in while people still feel good about the story.

Every company that ever listed wanted to do it into a strong market. That is not new and it is not sinister. Saying so out loud is new. The polite version normally appears in a prospectus as "market conditions may not persist." The impolite version got printed instead, and I have read offer documents for twenty-seven years without once seeing it make the news cycle before the roadshow.

Somewhere a very expensive syndicate desk read that headline over breakfast and put its coffee down slowly.

The market is not waiting for the prospectus. Overnight the semiconductor complex was taken out and shot. The SOX index fell 5%, Korea's Kospi dropped more than 5% in early trade after a 5% intraday slide on Tuesday, and Samsung Electronics and SK Hynix both fell around 7%. The US-listed Korea ETF fell more than 8% on Tuesday, its fifth single session loss of over 8% since early June.

Here is the detail that turns a market event into a farce. NYSE volumes are running roughly 50% below their twenty-day average. This was not the world's assembled capital repricing the global chip industry. This was August. Half the people who would normally argue with a seven percent move are on a boat. The market was unattended, somebody leaned on the door, and the whole thing swung open.

What the sellers did next is the funny part. They bought the bonds of the same companies. LQD, the investment-grade corporate bond fund, massively outperformed the plummeting chip index. Same firms, same buildings, same graphics cards, same cash flows. The sophisticated way to express enthusiasm about artificial intelligence is now to lend it money and hope it survives.

My own ledger, before I tell you what to do with yours. I flagged SK Hynix on Friday at 1,609,000 won and said accumulate in thirds, precisely because the 52-week range ran from 245,000 to 2,987,000. If you took the first third you are underwater, which is what that instruction is designed to feel like. Chinese memory capacity has not arrived and DRAM contract prices have not collapsed, so nothing that breaks the thesis has happened. August happened. Gold, recommended yesterday at $4,415, is now around $4,339, and the stop I named was a decisive close below the 100-day, which has not occurred. The position is losing and the thesis is not broken. Confusing those two is how people talk themselves out of trades the day before they work.

The actual story is in the bond market, which almost nobody is watching because the chip screens are prettier.

The 30-year Treasury yield touched 5.34% on Tuesday, the highest in nearly two decades. The 30-year inflation-protected yield closed Monday at its highest since 2008, and that is the number that matters, because it strips out inflation and tells you what capital genuinely costs. Ten-year yields sit at 4.75%, the Japanese 10-year is at its highest since 1996, and the US national debt passes $40 trillion this week. The same week, the Treasury will try to sell $16 billion of 20-year bonds on Wednesday and $8 billion of 30-year inflation-protected bonds on Thursday.

Into what, exactly. On Tuesday at least seven investment-grade issuers looked at the funding backdrop and decided not to announce their deals. Those are corporate treasurers paid to know when the window is open. Seven of them, same day, packed up and went home. That is a better signal than any yield print, because a yield is a number and seven abandoned bond deals is seven people with live mandates concluding today is not the day.

Very little stands between here and the 2007 peaks around 5.40% to 5.44%. Do not buy the dip in long duration until Thursday's auction clears. In a thin tape the auction is the clearing event, and until the paper is placed there is no reason to stand in front of it. The trade is more interesting on Friday than today.

Which brings me to the cleanest asymmetry on my screen, and it is not in equities.

The MOVE index measures how much volatility the bond market is pricing in. It sits around 75, the twentieth percentile of its five-year range, while ten-year yields press the top of a multi-year range, the debt crosses forty trillion, two large auctions land this week, Jackson Hole is next week, and the Federal Reserve is chaired by a man who has made declining to explain himself into a governing philosophy.

The bond market is at the top of its range and charging bargain prices for the possibility that something happens. In 2022, and again in late 2023, the journey toward 5% in tens came with violent volatility spikes as everyone repriced the Fed over and over. This time the climb has been orderly, polite, almost apologetic. Positioning explains why that will not hold: ten-year futures show the market extended short while consensus sentiment is aggressively bearish. Talking bearish and positioned bearish, at the top of a range, into supply, with insurance priced in the twentieth percentile.

Buy rates volatility. Straddles on ten-year Treasury futures, September expiry, sized as a position rather than a hedge, because you are not calling direction. A short squeeze that rallies bonds pays. A clean break through the range highs pays. The only outcome that hurts is another month of orderly sideways drift, which is exactly what is priced and exactly what August stops delivering the moment people come back. Invalidated by MOVE sustained above 95, at which point you are buying insurance at retail.

Now the best thing I have read all year.

Somebody ran a standard valuation model on the big AI names. Project cash flows five years out, assume modest growth forever after, discount it back, see what growth rate the share price requires to make mathematical sense. Ordinary work. Then they shocked the cost of capital by 50 basis points and looked at what broke. Results ranged from a 9% share price decline for Nvidia to 50% for Tesla.

Then the model met SpaceX, and the model gave up.

Not metaphorically. The formula requires a growth rate below the discount rate and positive free cash flow by the end of the projection window, and SpaceX analysts forecast negative free cash flow in every one of the next five years, so the mathematics produces nonsense. The published conclusion was that the company will either burn through its cash or rely on the kindness of strangers.

I have read a very large number of equity research notes and I have never before encountered a valuation model that surrendered.

The mechanism underneath the joke is the most important thing in this letter. Everyone is worrying about the debt these companies are taking on. Wrong worry. Most of them do not have much debt. They fund themselves from retained earnings and equity, so what kills the valuation is not the interest rate they pay, it is the discount rate applied to their future profits. Microsoft's cost of equity, on Bloomberg's calculation, has risen to 15.6% from under 10% a year ago, up 39 basis points since June alone, almost entirely because Treasury yields went up.

That is how a bond auction in Washington reaches into a technology portfolio in Singapore and takes a piece of it.

The trade is not to short any of this, because shorting expensive things that keep going up is how careers end. It is to stop paying for duration you are not compensated for. If your book is concentrated in the highest implied-growth names, the ones needing perpetual growth rates that would eventually make the company larger than the economy, you are running a leveraged position in the ten-year yield and calling it technology exposure. Rotate a slice into names whose cost of equity is not being repriced by the Treasury market. The gap between Nvidia's 9% and Tesla's 50% under an identical shock is the whole argument.

On oil I was directionally right yesterday and pointed at the wrong instrument, which I would rather say plainly than quietly correct. I said buy Brent call spreads because volatility sat in the third percentile. Brent implied volatility has since risen toward 52% with a pronounced call skew, and $110 and $125 calls have drawn heavy demand. It was cheap and now it is not. The better expression was always downstream.

Nymex heating oil and ICE gasoil cracks both hit fresh records on Monday. US diesel inventories are near multi-decade lows, European stocks near lows not seen since 2022. Middle East to China supertanker earnings have surged to roughly $500,000 a day and war risk insurance runs at high single-digit percentages of hull value, meaning one voyage now costs a real fraction of the ship. Brent at $91.30 makes this look contained. The tightness has migrated from the barrel to the refinery to the tanker, and diesel feeds straight into trucking, farming, construction and food distribution, which is to say into the inflation number that determines everything else here.

Own the crack, not the barrel. Refiners with heavy distillate yield, or gasoil crack spreads if you have the access. Spot crude is the least informative figure on the screen. The trade breaks on a Hormuz transit agreement or a meaningful build in US distillate inventories, and I want the inventory number rather than a headline about talks, because by the President's own account there are no active talks with Iran.

Two Iranian ballistic missiles fired toward the United Arab Emirates fell into the sea on Tuesday, the first known attack on the Gulf state since May. Crude moved 0.3%. Markets adapt to violence considerably faster than people do.

The United States and Canada, elsewhere, are hours from a midnight deadline after which 50% tariffs hit billions of dollars of Canadian goods under a provision of the Tariff Act of 1930 unused for ninety-six years. The sticking points are automotive rules, Canadian boycotts of American alcohol, and access for US dairy producers. The two largest trading partners in the hemisphere are going over a cliff about cars, booze and cheese, and the Canadian dollar has barely moved, which tells you the market does not believe the deadline.

The debt crosses forty trillion this week. The bond market is charging twentieth-percentile prices for the risk that this matters. And two of the most valuable private companies on earth are sprinting for the exit while the getting is good, and telling us that is why.

Everyone in this story is behaving perfectly rationally. That is the part that should worry you.

Good morning from the Lion City!

We truly are living through strange times… Yesterday the yield on the 30 year US Treasury touched 5.289%, the highest since June 2007. On the same day, a bond tied to a Microsoft linked data centre venture in Georgia drew north of $8 billion of orders for $3.9 billion of paper and is being upsized to mop up the excess.

Given a choice between lending to the United States government, an entity with a printing press, a navy and a two and a half century payment record, and lending to a shed in Georgia full of graphics cards that will be commercially obsolete before the coupon reaches middle age, roughly twice as much money chose the shed.

I want to be fair to the shed. It is presumably a lovely shed. Somebody has thought very hard about the cooling.

That is the tone of the whole week, so let us go through it properly, because there is real money hiding inside the comedy.

US investment-grade issuance hit $145.2 billion in August as of Monday, beating the record for the month, which was set in 2020, when the Federal Reserve was buying corporate bonds directly because everybody thought the world was ending. January, June and July also set records. Four of eight months this year have produced the largest corporate borrowing binges ever recorded, in an economy nobody is calling a crisis, at long-end yields last seen when Bear Stearns had a stock price.

The proceeds are going into artificial intelligence, and the circularity has stopped being subtle. Nvidia will spend up to $105 billion supporting a new data centre campus in Ohio, which will be leased by OpenAI, which buys chips from Nvidia. The supplier is now financing the customer's ability to buy the supplier's product. The industry term is vendor financing, and it was last this fashionable in 1999, when Lucent lent telecom operators the money to buy Lucent switches. That ended well for absolutely nobody, including the switches.

Groq, in the same market, on the same morning, raised $350 million at a $3.5 billion valuation, against a $6.9 billion peak in September 2025. A 49% down round for an AI chip company during the largest capital expenditure boom in the history of computing. Blackstone and Blue Owl are boosting bond sales out of their private credit funds. And Anthropic is tracking above $65 billion of annualised revenue ahead of an IPO, up sevenfold since December, with second-quarter revenue up at least fourteenfold to $11.5 billion and positive adjusted operating income.

Those Anthropic numbers are real and they are the best evidence anybody has that the capex is buying something. They also say nothing whatsoever about whether the shed in Georgia is money good in 2046, which is the question the bond market answered yesterday by declining to ask it.

The trade here is the curve, and it is not complicated. Investors are visibly rotating out of long-dated government paper and into higher yielding tech credit, which is a bear steepener with a corporate sponsor. Thirty-year yields will likely chop between 5.20%, the July payroll settlement, and Monday's 5.289% high into month-end, because it is August, the desks are half-staffed, and nobody launches a career-defining position the week before Jackson Hole. Own the 2s30s steepener through futures rather than cash, and understand that the invalidation is not a soft data print. Soft data used to rally duration. Friday gave you weaker retail sales alongside weaker Treasuries and weaker equities at the same time, which tells you the long end has stopped trading growth and started trading supply. The steepener breaks when issuance calendars thin out, not when payrolls miss.

Watch the Fed's balance sheet while you hold it. The ratio of bank reserves to commercial bank liabilities has reached the zone where the New York Fed's own reserve-demand elasticity metric goes nonlinear, which is the neighbourhood where money markets historically start behaving strangely. This is not a 2019 repo blowup in waiting, and the reserve distribution across the system is healthier than it was then. It is simply the first time in months that the balance sheet has been worth looking at, and Wednesday's FOMC minutes now carry more weight than usual under a communications regime that has become deliberately less communicative.

Politics is keeping pace with markets, in the sense that it has also stopped making sense.

The President told Fox he is in no hurry to end the war with Iran, will not extend the June memorandum that expired Monday, and will bomb Oman if it "gets in the way." Oman is the mediator. Oman has run the diplomatic channel for the entire Strait of Hormuz arrangement and is working with Tehran on a joint transit plan while Washington blockades Iranian ports. Threatening to bomb the referee is a bold approach to arbitration. He also confirmed that Kim Jong Un responded to his overtures and that the two leaders understand each other, which is either the most encouraging sentence of the year or the least, depending on your priors.

Brent sits at $91.04, up 2.8%. WTI at $84.91, up 3.0%. Saudi Arabia is reportedly offering oil ship-to-ship from Sohar in the Gulf of Oman, LNG cargoes appear to be transferring outside the waterway, and a Greek run Suezmax was attacked in the Black Sea after loading Russian-origin crude at the CPC terminal. Xi Jinping is skipping the UN General Assembly entirely and flying in for one day of meetings in Washington on 24 September, which is the diplomatic equivalent of attending the wedding but not the reception.

Which brings me to the single most mispriced thing on my screen.

One month Brent volatility spreads have collapsed to minus 22 vol points, from plus 20 when the Iran memorandum was signed, putting them in the third percentile of their three-year range. At the same time, at the money straddle breakevens and call skew still price meaningful upside tail risk. Implied is trailing realised. Open interest in crude futures has fallen sharply, thinning depth and making price action gappier. Market makers have grown comfortable selling straddles into a market they have decided is permanently range-bound, monetising gamma on both sides of a range.

That is a market pricing round trip volatility and refusing to price escape volatility, at the precise moment the truce has lapsed, the President is threatening the mediator, Saudi barrels are being shuffled around the Strait rather than through it, and tankers are being attacked in the Black Sea. Buy one to three month Brent call spreads struck 8% to 12% above spot, or own outright gamma if your mandate allows it, financed if necessary by selling downside you would be content to be assigned. It is invalidated by a signed, verifiable Hormuz transit agreement with Omani participation, and specifically not by another announcement that the parties are close to discussing the framework of an understanding. Horizon through late September, when liquidity returns and the CTA levels capping this market get tested for real.

Canadian producers, incidentally, are rethinking their oil hedges after finding them unprofitable, which is a sentence that should be tattooed on the inside of every corporate treasurer's eyelids. Hedges are supposed to be unprofitable when things go well. That is the entire product.

The commodity complex is having its own carnival. Cocoa rose 5.1% to $60.69. A near-term expiry is exacerbating a squeeze in a key coffee spread. LME copper finished lower after surging to a near-record high, aluminium sits at $3,266.50, platinum at $1,772.67, soybeans up 2.0%. And gold is up about 1% to $4,415.21, extending a one-month advance beyond 11%, the biggest 22-session move since early February.

Gold is the momentum trade of the moment and I will say the quiet part, which is that momentum is a perfectly respectable reason to own something provided you admit that is what you are doing. It has finally cleared the 100 day moving average after several failed attempts, with the 200 day roughly $90 above. Speculative longs are the highest this year. ETFs have added 38 metric tons since the July lows. South Korea's central bank bought gold related assets for the first time in thirteen years, Namibia intends to triple bullion's share of reserves by next year, and the PBOC added for a twenty-first straight month, the longest streak in the data.

Buy the 100-day break, hold toward the 200-day, and stop out on a decisive close back below the 100 day, which would tell you the breakout caught the last of the late buyers rather than the first of the new ones. The thing that kills this trade is not real yields, which have risen all month while gold rallied anyway. It is a dollar reversal, and DXY at 99.50 with two-year real yields still near 2% is a currency whose floor looks firmer than its recent price action suggests. Own the gold, respect the dollar, and do not be the last person in the room explaining why central bank buying makes valuation irrelevant.

Equities are where the plumbing is quietly interesting. The S&P closed at 7,745.06, down 0.5%, the Nasdaq Composite at 26,644.91, down 0.3%, the Dow at 53,459.78, down 0.5%. Roughly three in four S&P constituents fell, with more than three times as many names down 2% as up 2%, and only energy and tech positive. The Magnificent Seven were again a drag, which is a theme percolating under the surface rather than in the headlines.

Underneath that, the Nasdaq 100's breadth, measured as the share of members above their 200 day average, is around 73%, above the 90th percentile of its five-year range, at the same time the curve is bear-steepening hard. Those two conditions rarely coexist for long. The last comparable divergence was the first quarter of 2024, when roughly 45 basis points of 2s30s steepening coincided with a 25% compression in breadth. The index ground higher, but the rally narrowed sharply on the way.

Implied dispersion is collapsing alongside it, as the market rotates away from semis toward SaaS, metals and energy, which is the precise setup for a correlation shock. If you own the index and are relying on broad participation to hold it up, that participation is at a two-year high with a steepening curve leaning on it. The cheap hedge is not shorting the index. It is buying downside on the equal-weight version and letting the megacaps carry you, because in the scenario that hurts, the megacaps are the last thing to break and everything else goes first.

The deals desk is producing its own comedy. Shein is targeting up to $27 billion for its Hong Kong listing, raising about $2 billion, below its earlier target, which for a company built on shipping $6 dresses by air is a valuation discussion I would pay to watch. General Atlantic is working with JPMorgan, Goldman and Morgan Stanley on a listing as soon as this year. Brookfield has bid A$4.75 cash for Reliance Worldwide. Mark Walter has explored selling his Chelsea stake to Clearlake, while Sammons Financial bonds fell to their lowest since issue after a report probing the insurer's connections to Walter's Guggenheim Partners, a sequence I would describe as busy.

Andreessen Horowitz faces a Justice Department probe into whether its investment partners improperly serve on competitors' boards. L3Harris abruptly replaced CEO Christopher Kubasik after an internal investigation. Barclays named Adeel Khan and Mike Joo co-CEOs of its investment bank. Glencore acquired warrants for a potential minority stake in Radiant World months before moving to cut ties with the iron ore trader. Family offices from the Americas, Europe and the Middle East have accumulated at least $3.8 billion of SpaceX. And Jane Street posted its first monthly downturn in a decade, exposing a level of directional risk that rather undermines the polite fiction that market makers, proprietary traders and hedge funds are different species.

Emerging markets are sorting themselves into two piles. Nigeria's inflation cooled more than expected to 15.4% in July, strengthening the case for the central bank to resume cutting as soon as September. India's BJP reshuffled its national leadership before key state elections, making Commerce Minister Piyush Goyal treasurer, while the RBI weighs restricting nonbank lenders to term loans with fixed repayment schedules, a well-meaning directive that ignores how irregularly most Indians actually get paid. Canadian CPI ran hot in July, though core measures barely moved and swaps still price the Bank of Canada's next hike for January 2027. Casas Bahia filed for bankruptcy on high rates and tight credit.

Brazil is the pile I would stay away from. Economic activity fell in June by the most since May 2025. Restrictive real rates are doing exactly what restrictive real rates do, which is work. The real is the worst-performing Latin American currency against the dollar over one week and one month, and the largest US-listed Brazil ETF just suffered a record weekly outflow. There is an October election, an incumbent with a well-documented relationship with fiscal restraint, and a fiscal premium rebuilding into local assets in real time. Record outflows are a momentum signal, not a contrarian one. The classic error with Brazil is buying two months early because the valuation screen is so pretty. Stay short or stay out until the fiscal premium stops widening, and let somebody else pick the low.

A San Francisco Fed paper argues the medium-term real neutral rate is 1.5%, implying a nominal neutral of 4.5% or higher, which would mean policy is currently accommodative. R-star remains the Loch Ness Monster of macroeconomics. Everybody has heard of it, the photographs are always blurry, and an entire research industry depends on nobody ever getting a clear look.

Elsewhere in things that are technically happening: Alibaba launched an AI music model called HappyShrimp 1.0 that generates melodies, lyrics and vocals from a prompt describing an emotion. Vitol is supplying entire African nations with fuel under contracts largely awarded outside normal tender procedures. A week of rain fell on Manila in two hours. And American train robberies, which I had filed under "resolved, 1890s," have surged since the pandemic, with more than $200 million of goods stolen from US rail hubs in 2025. Somewhere a risk officer at a Class I railroad is writing a memo about Butch Cassidy and hating every second of it.

The through-line, if you want one, is that the market has stopped pricing the level of anything and started pricing the rate of change of everything. Ten-year yields near 4.7% and thirties above 5.3% are not troubling equities because they are not moving fast, and cross-asset volatility is on the floor. Through the 1990s the ten-year averaged north of 6.5% while the S&P compounded near 20% a year. High rates are survivable. Rapidly changing rates are not, and the current cohort of portfolio managers has never run a book in an era of substantial real yield.

Somewhere in Georgia, they are still building the shed.

Happy Friyay!

Its been a long week, but it's becoming a longer week for Chinese AI consumers.

Today we go all the way back to 1621, when the Dutch East India Company solved a pricing problem by killing roughly fifteen thousand people.

The problem was nutmeg. It grew on ten volcanic specks in the Banda Sea and nowhere else on earth, it was worth more than its weight in gold in Amsterdam, and the Bandanese had the irritating habit of selling it to whoever showed up with a boat. Jan Pieterszoon Coen, the VOC's governor general and a man with a spreadsheet where his conscience should have been, worked out that the monopoly was worth more than the population. He depopulated the islands and installed a system so thorough that for the next century the company burned surplus harvests in public, ordered growers to chop down healthy trees, and executed smugglers, all to keep a spice artificially scarce.

Read that list again slowly, because the middle item is the strange one. Burning your own harvest is at least legible as a business decision. Executing smugglers is monstrous but comprehensible. Ordering a farmer to walk out to a healthy, productive tree and cut it down, in a year when the crop was good, because the crop was good, is a man following a spreadsheet off a cliff and taking the orchard with him. Somewhere in Amsterdam a clerk was recording those felled trees as an asset.

It worked. For about a hundred and fifty years, it worked beautifully.

Then a French botanist called Pierre Poivre, whose name genuinely translates to Peter Pepper, stole some seedlings and planted them in Mauritius. The British grabbed the islands during the Napoleonic Wars and transplanted trees to Ceylon and Sumatra. And nutmeg became a thing that grows in a lot of places, and the greatest monopoly in commercial history discovered it had spent a century and a half in the wrong business.

I have been thinking about Peter Pepper all week, because on Thursday a company in Hangzhou raised its prices and half of Silicon Valley treated it as good news.

DeepSeek's new rate card for V4 Flash is 22 cents per million input tokens and 66 cents per million output tokens off-peak, doubling at peak hours. The old price was 14 and 28. So depending on when you call, this is a price rise of somewhere between 57% and 371%, which sounds dramatic until you check what the neighbours charge. Anthropic's frontier model is priced, per the same reporting, at $10 and $50 for the identical units. Moonshot's Kimi K3, considered a premium product inside China, runs $3 and $15.

The American model costs about seventy-five times the Chinese one at the output end. After the increase. That is the number to sit with.

The consensus read is that DeepSeek must be bleeding money and will eventually behave rationally. It is probably bleeding money. I have spent 27 years watching people mistake a strategy for a mistake, though, and the tell never changes: when the loss is this large, this deliberate and this well funded, stop asking whether they can afford it and start asking what they are buying.

Picture a street with five bakeries selling croissants at five dollars. A sixth opens and prices at seven cents. Everyone assumes the new guy cannot do arithmetic. Then someone checks the deeds and discovers he owns the wheat farm, the delivery vans, and the building the other five are renting. The seven cent croissant stops looking like an error and starts looking like an eviction notice written in pastry.

The deeds here are public. Combined 2026 AI capex for Alibaba, Tencent, Baidu, SenseTime and Kingsoft Cloud is projected at $42.8 billion. For Amazon, Alphabet, Microsoft and Meta it is $744.3 billion, rising to $993.9 billion in 2027. Seventeen dollars of American capex for every Chinese dollar, deployed to manufacture a product whose price the Chinese side is unilaterally setting at roughly one seventy fifth. Which is fine. That is a completely normal ratio to build a trillion-dollar industry on.

Nobody in this argument says the obvious thing plainly. The two sides are not selling the same object. American AI firms sell intelligence, metered per token, on a financial architecture assuming frontier access is a durable chargeable good. China is not selling intelligence at all. Beijing's "AI Plus" programme, launched by Li Qiang, targets deployment into electric vehicles, agriculture, robotics and factory equipment. The model is the road, and Beijing has decided roads should be free, because the money is in everything that drives on them.

The receipts are arriving faster than anyone expected. High tech exports rose 41% year on year over the first seven months against 19% for total shipments. Strip out the AI price surge and high-tech manufacturing still grew 13.3% in real terms in the first half, double the 5.4% pace of total industrial output. Bloomberg Economics now estimates high-tech and green industries could contribute close to 20% of Chinese GDP in 2026, overtaking property for the first time. Which deserves a moment of silence for every China bear who spent five years correctly identifying that the property market was a smoking crater and then incorrectly concluding nothing would ever grow in the ash.

Now the part that should actually keep American CFOs awake, and it has nothing to do with China's domestic market.

DeepSeek holds 18% of the Ethiopian market. It is dominant in Russia, where US models are embargoed anyway. It is spreading through sub-Saharan Africa and South Asia for a reason that requires no geopolitics whatsoever to explain. Somewhere in Lagos this quarter, a bank's head of technology walked into a room with two quotes.

Sixty six cents per million tokens. Or fifty dollars.

For the same job?

For the same job……

And that is the entire meeting. It lasts about nine seconds, nobody in the room has heard of anyone's constitutional AI framework, and the procurement officer who chose the cheap one will be promoted, correctly, for saving the bank ninety-eight percent on a line item. Whoever supplies the first generation of AI infrastructure to the developing world also sets its technical standards, its data formats, its integration patterns and its default assumptions about who you call when it breaks. Those things have a half-life measured in decades. Ask anyone who has tried to migrate a bank off a mainframe.

So the Americans compete furiously for the developed world's enterprise budget while the Chinese quietly take the other five billion people and the standards that come with them. It is what you do when you cannot win the premium market and have correctly worked out that the premium market is not where the century gets decided.

Open weights are seedlings. That is the whole thing, and I am not going to belabour it.

There are now close to a thousand large language models available in China, and DeepSeek's own weights are published under a permissive licence, meaning any competitor can download the model and undercut DeepSeek using DeepSeek's own product. That is a very firm ceiling on how far Liang Wenfeng can push this. Moonshot's answer to the same problem is the cleverest engineering in this story: buried in the Kimi K3 licence is a clause requiring any firm earning more than $20 million a year hosting the model to enter a commercial agreement. Alibaba is reportedly weighing the same for Qwen. It is a franchise fee dressed as an open-source licence, and it is the single most elegant thing anybody in this industry has done all year.

It will not survive contact with reality. Walk it through. A startup in Shenzhen crosses the twenty million threshold hosting Kimi K3. Moonshot's legal department notices, drafts a letter, and sends it. The startup reads the letter, agrees that it is a very well-argued letter, and spends the following weekend migrating to one of the roughly nine hundred and ninety-nine other models that do not come with a lawyer attached. Total switching cost: a long weekend and some regression testing. Moonshot now has a beautifully drafted clause, an unenforceable claim, and one fewer customer.

You cannot put a toll booth on a road you have already given away, and you certainly cannot put one on a road where nine hundred and ninety-nine parallel roads run to the same place.

Which brings me to where the money actually goes, and how you own it without underwriting a token price war.

If intelligence gets commoditised before anyone works out how to charge for it, value migrates to whatever stays scarce, and what stays scarce is memory. Not compute generally, and not Nvidia, which everyone already owns at a price assuming all of this. High bandwidth memory is the one component China's domestic champions genuinely cannot replicate yet, which is why total China-South Korea trade jumped 52% year on year in January to July against a 1.2% gain for all of 2025. SK Hynix reported a 76% operating margin last quarter on revenue up 257% year on year, started HBM4 mass production in Q2, and has ten long-term customer contracts signed. The stock closed at 1,609,000 won on 13 August, having fallen 35% during July and then rebounded roughly 30%, against a 52-week range of 245,000 to 2,987,000 won. That range is not a typo and it tells you exactly what you are buying: the correct instrument, in a violently wrong footed tape.

I would accumulate it in thirds rather than in one clip, precisely because of that range, and I would treat the June all-time high of 2,987,000 as the level that confirms the thesis rather than the level to chase. The trade breaks if Chinese domestic HBM capacity arrives materially earlier than expected, which would show up first as a collapse in DRAM contract pricing rather than in anyone's press release, so watch the monthly contract prints and not the headlines. Horizon is through the HBM4E ramp in 2027, which means eighteen months and a stomach.

The quieter trade is that American hyperscalers are beneficiaries of Chinese open models, not victims of them. Every enterprise that runs Kimi K3 rather than paying frontier prices still has to run it somewhere, and that somewhere is overwhelmingly AWS, Azure or Google Cloud, which bill for compute regardless of whose weights are burning it. Own the landlord, not either tenant. Expressed simply, that is Amazon and Microsoft held at benchmark weight or slightly above, funded by trimming whatever pure-play model exposure sits in the book, and it breaks if either firm starts writing large equity cheques into a model developer, at which point they have stopped being the landlord and become a tenant with extra steps. It is the least exciting idea in this letter and almost certainly the highest Sharpe.

Which leaves the obvious question of why I am not short anything here, given a thesis that says one side of a $744 billion race is structurally exposed. OpenAI and Anthropic both have listing ambitions, DeepSeek is preparing its own, and every one of those prospectuses will carry per-token revenue projections drafted before the price of the core product was settled. I am not shorting any of them, because being right about a valuation and solvent through a listing window are separate achievements and I have the scars to prove it. Read those documents the way you would read a nutmeg trader's business plan dated 1770, and then simply decline to buy.

The VOC burned harvests for a hundred and fifty years to defend a price. It executed people over seedlings. It built the most sophisticated commercial machine the world had ever seen on the assumption that scarcity was a thing you could enforce with sufficient violence and adequate bookkeeping. And in all that time, in all those board meetings in Amsterdam, it seems never to have occurred to anyone that if you already owned the ships, the ports, the warehouses and the trade routes, the correct move might have been to plant the trees everywhere yourself.

Somebody in Hangzhou appears to have read the footnotes.

Kevin Warsh wants the Fed to stop talking. Thirty years of bond market plumbing was built on the assumption it never would.

There used to be a job called Fed watcher. Before February 1994, the Federal Open Market Committee did not announce what it had decided about interest rates. It moved the target, told the trading desk in New York to go execute it in the repo market, and let everyone else figure it out by watching how much cash the Desk was draining or adding on a given morning. An entire profession existed to reverse engineer a number the Fed already knew and had simply declined to say out loud. Alan Greenspan called this constructive ambiguity and meant it as a compliment to himself.

New Fed Chair Kevin Warsh appears to want that job back. He has declined to commit to regular press conferences, stopped short of following his colleagues in publishing economic and policy forecasts, and told investors last year that the Fed should simply stop talking so much. Christine Lagarde found common cause with him on a panel in Sintra this summer. Andrew Bailey noted that forward guidance is much easier to install than to remove. There is a real argument buried in that discomfort, which is that central banks locked themselves into promises during the pandemic that the 2021 and 2022 inflation surge then made embarrassing to keep. That argument, whatever its merits, is not the one that should worry you as an allocator. The one that should worry you is simpler. Guidance was never a courtesy. It was a subsidy, and Warsh just announced he is ending it.

Here is the plumbing nobody at the podium ever explains. A central bank that tells you its reaction function is, whether it thinks of it this way or not, selling the market cheap volatility. If you already know the Fed hikes when core inflation clears three percent and cuts when unemployment clears five, you do not need to pay much for protection against the in-between scenarios, because there are not many in-between scenarios left to protect against. Forward guidance compresses the distribution of possible outcomes and the compression shows up, mechanically, as lower implied volatility across the entire rates complex. Every basis point of term premium embedded in a ten-year Treasury yield is partly compensation for exactly this uncertainty. Reduce the uncertainty about the reaction function and you reduce what investors demand to hold the long end. That is not commentary. It is the transmission mechanism.

The pandemic-era episode that Warsh keeps citing as his justification is worth taking apart on its own terms, because it is a genuinely different mechanism than the one he is proposing to fix. When the Fed promised in December 2008 to hold rates near zero for an extended period, and later tightened that into more specific calendar and threshold-based commitments, it was making a promise about its own future actions in order to pull down long-term borrowing costs while short rates were already pinned at the floor. That promise became a liability in 2021 when inflation ran hot and the Fed had boxed itself into a path it had explicitly told the public it would follow. The fix for an over-specific promise is a less specific promise, not silence. Warsh is treating the failure of one particular guidance commitment as proof that all commitments are the problem, which is a bit like blaming the existence of contracts because one contract had bad terms.

Take the subsidy away and the uncertainty does not vanish. It just stops being free. Somebody has to price it, and the only people who price uncertainty for a living are options desks. I would be building a small structural long in rates volatility right now, specifically through short-dated Treasury note options straddled around the September FOMC decision, sized as a persistent quarterly position rather than a single trade, on the view that realized volatility around Fed meetings should structurally rise as long as this chair keeps declining to narrate his own reaction function. You are not betting on a hike or a cut. You are betting that the market's confidence in guessing correctly, in either direction, goes down, and that confidence has historically had a price you can actually buy.

I have watched a version of this before, from the other direction. The 1994 Fed produced what bond traders of a certain age still call the bond massacre, a year in which the Fed raised rates six times without any advance signaling of the campaign's shape, and firms that had built their entire balance sheets on the assumption that duration risk had gotten boring got run over. Orange County's investment pool, levered on the same assumption, filed the largest municipal bankruptcy in the country's history that December. The lesson everyone drew afterward was that the Fed had been too opaque, and the transparency revolution that followed, culminating in Ben Bernanke's first post-meeting press conference in 2011, was built explicitly to prevent a repeat. Warsh is not proposing something new. He is proposing to walk back thirty years of institutional memory built specifically because the old way once broke a very large pension pool in Southern California. He may be right that the pandemic-era promises went too far. He is choosing an unusually specific historical scar to reopen in order to make that point.

The trade underneath the trade is the curve itself. A Fed that declines to explain its objective function is a Fed that forces the market to demand more compensation for holding duration into that ambiguity, which is another way of describing a steeper curve. Layer onto that an inflation backdrop where you have a Fed chair openly skeptical of talking about inflation and a set of hard July data points still running above target, and you get a term premium with two separate reasons to widen rather than one. I would own a 2s10s steepener here, expressed through Treasury futures rather than cash bonds for the margin efficiency, and treat it as a multi-month position rather than a trade around any single data print, because the thesis is about the shape of communication for the next several years of this chairmanship, not about next Wednesday's number. There is a second-order version of the same trade sitting in plain sight in bank equities. A steeper curve is a bank's oldest and least glamorous business model, borrow short, lend long, pocket the spread, and the regional and money-center banks that get penalized whenever the curve flattens are the same names that quietly re-rate when it doesn't. I would rather own a basket of the more rate-sensitive US regional banks against this thesis than try to trade the curve directly through futures margin every time volatility spikes, because it gets you the same directional exposure with an earnings tailwind attached instead of a mark-to-market that resets every afternoon.

Zoom out past Washington and the same asymmetry is opening up across the other major central banks, and this is the part that should actually matter to a reader in Singapore or London more than the domestic Fed story does. The Bank of Japan just did the opposite of what Warsh is doing. Its July meeting notes, released this past Monday, showed board members explicitly discussing whether the pace of hikes needs to run faster than the market currently expects, language unusually direct for an institution that spent a decade communicating almost entirely through vague hints. That vote was eight to one, with board member Hajime Takata dissenting in favor of an immediate move to 1.25 percent rather than the 1.0 percent hold the majority chose. Wholesale inflation data released in Tokyo this morning added more fuel underneath a September hike that the market was already leaning toward pricing. The ECB, in the same window, is workshopping an entirely new vocabulary word for its own version of guidance, what Lagarde is now calling framework guidance, an attempt to keep telling the market how it will react to news without promising what it will actually do. Three of the world's four most important central banks are experimenting with new, more precise ways to talk to markets in real time. The fourth just told everyone to stop asking.

That divergence is not a curiosity, it is a cross-border repricing already in motion. Global bond yields have pushed to multi-year highs across several major markets this month, and a widening gap in how transparently the world's rate-setters explain themselves is one entirely underappreciated reason why. A market that can read the Bank of Japan's internal debate almost in real time and cannot get equivalent visibility into the Federal Reserve's will structurally demand a different risk premium for each, all else equal. If you run cross-currency rate books or hedge dollar liabilities against a portfolio priced partly in yen or euros, the cost of that hedge is now embedding a communication discount that did not exist eighteen months ago, and it is going to keep widening for as long as one chairman keeps declining to hold a press conference the other three are competing to make more informative.

A skeptic will point out that markets survived worse blackouts than this one, and that is true, the pre-1994 Fed ran the largest economy on earth without a single press conference for eighty years. But that Fed also operated inside a bond market that was a fraction of today's size, without a swaps market, without an options complex pricing rate paths out to thirty years, without pension funds and insurers running liability-driven strategies that assume a knowable path for discount rates decades into the future. The infrastructure built on top of transparency cannot simply be unbuilt because one chairman finds forecasting distasteful. You do not tear out a load-bearing wall because you have decided you preferred the open floor plan of the original building.

None of this is a verdict on whether Warsh's monetary policy will be good or bad. Fedspeak, mumbled or otherwise, was never really about the words. It was collateral. Thirty years of central bankers built an entire pricing regime for global fixed income on the assumption that the world's most important rate setter would keep explaining itself, and every basis point of compressed term premium since 1994 was the interest paid on that collateral. Warsh just told the market the collateral call is coming and gave no date for it. Bond markets have never once handled that kind of silence gracefully, and I have no particular reason to think this one will either.

Good morning all!

A GPU has a useful life of about three to five years. The bond that paid for it matures in ten, twenty, sometimes thirty. Nobody has fixed that gap. Wall Street just found a way to sell it anyway.

Start with a number that should be on every allocator's desk and mostly isn't. Barclays ran the math on duration-times-spread, a measure that combines a bond's credit risk with its interest rate sensitivity to show how much of a portfolio's total risk any single issuer actually accounts for. Six technology companies, the five largest hyperscalers plus SpaceX, now make up 8.4% of that measure across the entire US investment grade corporate bond market. The six largest banks in America, the institutions three separate regulatory regimes have spent forty years building stress tests around, account for 7.3%. Six compute companies are now a bigger source of interest-rate and credit risk to the world's benchmark bond index than Bank of America, JPMorgan, Citigroup, Wells Fargo, Goldman Sachs and Morgan Stanley combined.

Sit with that. The banks get an annual stress test that assumes a severe global recession, a spike in unemployment and a collapse in commercial real estate, then publishes the result. The hyperscalers get an earnings call. Nobody runs a 2008-style stress scenario on what happens to a bond backed by racks of accelerators that a competitor's chip has made obsolete two years into a twenty year note.

This is not a leverage story, or not only one. JPMorgan estimates that $4.1 trillion of the projected $5.5 trillion in AI capital expenditure will be debt-financed. That is a familiar number by now and mostly discussed as a size problem. The size is not what should worry you. It is the tenor. A hyperscaler issuing 20 year senior notes to fund a data center is making a bet that whatever sits inside that building in year eighteen is still generating revenue against a coupon struck today. Their own accounting says otherwise. Depreciation schedules on GPU fleets across the industry run three to six years, because that is genuinely how fast a Blackwell-class chip gets outrun by whatever Nvidia, or somebody else, ships next. You are underwriting a thirty-year mortgage on a car.

I watched this exact mismatch once before, and it was not called artificial intelligence, it was called broadband. Between 1998 and 2002, telecom carriers issued something on the order of $500 billion of debt to lay fiber optic cable on the theory that internet traffic would keep doubling forever. WorldCom alone carried more than $30 billion of bonds into its 2002 bankruptcy, at the time the largest in US history. Global Crossing laid 100,000 miles of fiber and filed for Chapter 11 with $12.4 billion in debt the same year. The traffic thesis was not wrong. Internet usage did keep growing, for decades, in a straight line that would have made any of those bondholders whole eventually. It just did not grow fast enough to service coupons struck against equipment that both depreciated and, worse, kept getting cheaper to replace as the technology improved. The fiber itself mostly survived and got bought out of bankruptcy for cents on the dollar. The bonds did not survive. That is the part people forget when they reach for this analogy and then only use it to argue about whether the demand is real. The demand does not have to be fake for the financing structure to fail. It only has to arrive on a different clock than the one the coupon was written against. Print that on a card and staple it to every AI capex model in Singapore, London and Tokyo that still treats depreciation as a footnote instead of the whole argument.

Look at what is actually happening to the cash generating the collateral for all this borrowing before you decide the credit market is overreacting. Alphabet reported negative free cash flow last quarter for the first time since it went public more than two decades ago. Meta's free cash flow fell 91% year over year to $784 million, a company that a year earlier was generating cash the way a toll bridge does. These are not distressed companies. They are two of the most profitable enterprises in human history, choosing to spend every dollar of operating cash flow and then borrowing on top of it, on the theory that the equipment they are buying will still be earning in year fifteen of a note most of it will not physically exist for past year five. The credit market is starting to notice, which is the useful part of this story rather than the alarming part. AI-linked issuer spreads have widened in recent weeks, not to anything resembling distress, but enough that Barclays is showing it clearly on a chart and enough that State Street's fixed income desk is describing the spending cycle as entering a more mature, more skeptical phase. Spreads moving before headlines is not a new pattern. It is close to the only pattern credit markets reliably show. Widening here is not a crisis signal. It is buyers finally pricing the depreciation mismatch instead of pretending it does not exist, which is closer to the system working than to the system breaking.

There is a second mismatch running in parallel this week, on the other side of the world, and it rhymes with the first one more than it looks like it should. Japan and the United States intervened jointly to buy yen on 30 and 31 July, spending an estimated $53 billion and then $34 billion on consecutive days, the largest coordinated intervention since 1998 and possibly the largest single-day intervention on record. It worked for about a week. The yen ran from roughly 164 to the dollar back to 155, then spent Monday giving half of that back, closing near 159.3, because intervention buys you a level, not a trend. What actually moves a currency's trend is the interest-rate differential funding the carry trade against it, and Tokyo has not touched that differential yet. Traders are now pricing a 63% chance of a Bank of Japan hike by September and a move by October that is close to fully priced. Until one of those actually happens, every yen the Ministry of Finance spends defending 160 is a short-dated bond bought against a decision the market has already made about direction. Same mismatch, different balance sheet. Spend real money now to paper over a rate differential that is not going to close itself later.

None of this argues for panic, and panic is exactly the wrong instinct to bring to it, because the AI capex story and the earnings underneath it remain the best evidence anyone has that the demand is showing up. Second-quarter S&P 500 profit growth accelerated to 34%, from 25% in the first quarter, one of the strongest prints outside an actual recession recovery in the data set, and for the first time since ChatGPT's 2022 debut the growth is not entirely a Magnificent Seven story. Tech's contribution to aggregate earnings growth fell to about 55%, from roughly 90% a year ago, with seven of eleven S&P sectors now posting double-digit gains. That broadening is the single most bullish data point in this entire column, and it deserves to be treated as one instead of buried under the credit worry, because a market getting its earnings growth from more places is a market with more places for capital to hide when one of those places disappoints.

It is also, quietly, a market getting less honest about margins while getting more honest about revenue. Nearly 90% of S&P 500 companies reported higher sales this quarter. Barely half improved their margins. Selling more while keeping less of it is what a company does when input costs, tariffs or wage pressure are eating the difference, and it is a detail the earnings broadening headline lets you skip past if you are not looking for it.

So here is where I would actually put money, in three places that follow from the plumbing rather than from a view on whether AI itself works. Sell protection on nobody and buy a little instead. With investment-grade credit spreads on hyperscaler paper still historically tight even after the recent widening, the cheap trade is owning five-year credit default swap protection on the two or three most aggressively levered AI issuers funding the longest-dated notes, sized as portfolio insurance rather than a directional bet, because the premium for that protection has not yet caught up to a depreciation mismatch this obvious. On the equity side, the broadening story argues for actually following the earnings rather than the narrative, rotating a slice of concentrated mega-cap AI exposure into an equal-weighted S&P basket tilted toward the names posting margin expansion alongside revenue growth, since the Societe Generale data makes clear that half the index is growing sales into compressing margins and you do not want to own that half by accident through an index weighting. And on the yen, fade the intervention rather than the level. Short-dated yen call options funded against carry income look cheap precisely because the market is treating July's coordinated buying as a policy commitment instead of what it actually was, a single balance sheet transaction that expires the moment the rate differential it was papering over reasserts itself, which a 63% priced September hike suggests could be sooner than the spot price currently reflects.

Every one of these ideas rests on the same instinct, which is the only instinct that has ever reliably made money for me since 1999. Find the clock nobody has synchronized. A GPU depreciates faster than the bond that financed it. A currency intervention buys weeks against a rate decision measured in months. A stock market rewards revenue on a quarterly clock while margins quietly run on their own schedule underneath it. Every one of these mismatches looks fine for a long time, right up until somebody needs the money back on the clock they actually agreed to, instead of the one everybody was pretending was still running.

I am having a case of the… Munday's… but it's Tuesday….

Here is a tweet. "The best day of my life. I love you." A heart emoji. Sent via Twitter for iPhone last Saturday, by a 20 something hedge fund manager named Leopold Aschenbrenner, on the occasion of his wedding.

Here is the same week, more or less. Aschenbrenner's fund, Situational Awareness, grew to roughly $45 billion at the start of July. By the end of the month it had lost enough on artificial intelligence stocks, and a short against software names, that its prime brokers, Bank of America, Goldman Sachs and JPMorgan Chase, were marketing chunks of its book for sale ahead of Thursday's opening bell so the fund could meet margin calls in an orderly fashion. Most of what remained went to Citadel in one transaction. Aschenbrenner then wrote to investors that he had removed all leverage from the fund and called the episode expensive but invaluable.

I have been doing this since before the dotcom bust, which means I have now watched four separate generations of extremely smart young men discover the same fact about borrowed money at roughly the same velocity, and I want to be precise about what actually happened here, because the plumbing is more interesting than the wedding. I watched the same trade in different clothes during the dot-com margin calls, again in the commodity trader blow ups here in Singapore in 2015 and 2016 when a handful of firms discovered that a warehouse receipt is not the same thing as the metal it describes, and again during Volmageddon in February 2018 when short-vol funds that had been printing money for years lost it all in a single session because the instrument they were short had no mercy for how long they had been right before that. The names change. The mechanism does not.

Aschenbrenner did not lose because he was wrong. He wrote a widely read 2024 essay series called Situational Awareness, arguing that artificial intelligence would require an enormous buildout of compute, chips, memory and power, and he built a fund around that thesis with a few hundred million dollars less than two years ago. Through May it was up 270% after fees and more than 1,000% since inception. A fifth of the book sat in a single position, Anthropic, which is up more than 400% over the period. SK Hynix, on the memory side of the thesis, is up 220% over two years. Nvidia is up 83%. He called the trade of the decade and the decade agreed with him.

The problem was never the direction. The problem was the funding. Aschenbrenner's own thesis is denominated in years, by his own title, The Decade Ahead. Data centre bonds pay investors back over ten or fifteen years, because the underlying leases run that long. Hyperscalers raise equity for the buildout because equity never has to be repaid at all. Aschenbrenner funded a decade-long thesis with prime brokerage margin, which gets called on a Tuesday afternoon the moment the collateral moves against you. You can be right about 2034 and insolvent by lunch on the way there.

This is not a new mistake, it is the only mistake. Every blowup I have watched since the dot-com unwind wears a different costume over the same skeleton. A correct or nearly correct view about the future, funded with money that has the patience of a caffeinated day trader. Long-Term Capital Management held convergence trades that eventually converged almost exactly as modelled, and still needed a Fed-brokered rescue in September 1998, because the repo lines funding the position could not survive the interim. The trade worked. The financing did not wait around to find out.

The specific mechanics of last month's unwind are worth walking through, because they explain who was actually forced to sell and why the tape moved the way it did. Aschenbrenner's backers were mostly wealthy individuals and family offices concentrated in the San Francisco Bay Area, not the pensions and sovereign wealth funds that anchor more established multistrategy shops. That investor base matters because it is fast money dressed as patient money. Barclays declined to onboard Situational Awareness as a prime brokerage client in the weeks before the near miss, citing concentration in a single sector. Morgan Stanley passed at launch too, citing Aschenbrenner's inexperience, then reversed course this year and was preparing to onboard the fund as a client in the weeks the wheels came off. Read that sequence again. The bank that said no to the risk two years ago said yes just before the risk arrived, because 270% is a very persuasive marketing document. Bank of America, Goldman Sachs and JPMorgan kept extending leverage the whole way up, which is what prime brokers do until the day they stop, at which point they all stop within about seventy-two hours of each other and call it prudent risk management.

Here is the part that would have been unremarkable in 1998 and is genuinely new now. Within days of the fund telling investors it was not accepting new capital, Silicon Valley started calling to offer more of it. Elad Gil, a venture investor with a track record most allocators would kill for, publicly said he had asked to invest for the first time. A partner at Sequoia Capital told Bloomberg Television that Aschenbrenner would be "a fixture in Silicon Valley for a long time to come." A recruiter I would ordinarily dismiss put it better than any of them: "even a wounded lion is still a lion." Wall Street's read on the same event was the opposite. An AI-focused hedge fund with $20 billion of assets, run by people who trade against exactly this kind of concentrated, levered book, called it what it was, a forced seller finally meeting its forcing function.

That divide is not new either, and I have a name in mind that nobody under thirty-five will recognise but every allocator over fifty will.

Even the diversified professionals took a hit last month, which is worth noting before anyone frames this purely as a story about one overconfident 20-something. Millennium Management was down 2.1% in July. Point72 fell 3.3%. Altimeter Capital's AI-focused fund, which runs a genuinely concentrated book by design, tumbled 11%. A rival AI-only fund, VAR Advisors, staffed with veterans out of BlackRock and Hudson River Trading, still managed to post roughly 194% for the year through June, more than nineteen times the S&P 500's return over the same stretch, without needing a bailout call from Citadel. The difference between Altimeter's 11% July loss and Situational Awareness's forced full liquidation was not conviction. It was leverage and counterparty concentration, the two variables every one of these funds could disclose to their investors in a single paragraph and mostly don't. In September 2006, Amaranth Advisors lost roughly $6 billion in about a week on natural gas spreads that a single trader had built into the largest position most of the market had ever seen. Amaranth's founder, Nicholas Maounis, spent the years after doing what founders do after a blowup, which is nothing publicly interesting, until he quietly started Verition Fund Management with around $1 billion in 2019. Verition runs $11.8 billion today. It carries close to three times the number of risk managers Amaranth had per dollar of capital, and it caps how much any single portfolio manager can put on. The lesson Maounis apparently learned was not stop taking risk. It was stop letting one trade be the whole firm. That lesson took him roughly thirteen years and one closed fund to fully absorb, and Aschenbrenner has had about three weeks.

Zoom out past the two funds and the actual global story is the financing stack underneath the entire AI buildout, which is where Singapore, London and Tokyo readers should be paying closer attention than the hedge fund gossip suggests. New asset-backed issuance hit a record $335 billion last year, with collateralised loan obligations at $201 billion, and data-centre-backed bonds are now a standing category at industry conferences that used to be about auto loans and credit-card receivables. BlackRock estimates the buildout will cost roughly $3 trillion over five years. Some of that gets funded with fifteen-year asset-backed paper matched to fifteen-year hyperscaler leases, which is boring, appropriate, and exactly the kind of financing Aschenbrenner should have used to express his own thesis. Some of it gets funded with overnight prime brokerage margin against volatile public equities, which is what he actually used, and which is the same mismatch that froze the market in October 1987 for entirely different plumbing reasons and nearly froze Situational Awareness for this one. The instrument changes generation to generation. The mismatch between how long your conviction takes to pay off and how long your lender is willing to wait does not.

None of this means the AI capital expenditure story is wrong, any more than Amaranth being wrong about position sizing meant natural gas spreads were a bad market. Bank of America still has hyperscaler capex for calendar 2026 running near $859 billion, up close to 80% on the year, which is the actual demand curve underneath both the SK Hynix move and the Anthropic stake that made Aschenbrenner's numbers so good on the way up. If you want exposure to that curve, the boring version is a diversified basket or an index sleeve funded with your own permanent capital, term-matched to a thesis that plays out over years, not a single manager's concentrated book funded overnight by three banks who can and will pull the line together. If you are the family office deciding whether to write Aschenbrenner a cheque now that Sequoia's partners are cheering from the sidelines, the only question that matters is not whether he is right about 2034. Ask him what happened to the leverage. Ask which bank is providing it and what percentage of the book sits in one name. Ask what redemption terms the fastest investor in the fund gets, because that is the term that determines what happens to you.

Silicon Valley is going to keep calling this a scar that made him stronger, because Silicon Valley genuinely rewards conviction and treats a survived blowup as proof of it. Wall Street is going to keep calling it a forced liquidation that happened to end in a wedding photo, because Wall Street has been sizing margin calls since before Aschenbrenner was born. Both of them are describing the same seventy two hours. Only one of them is pricing the next one correctly.

Happy National Day to all my Singapore followers, and a happy Sunday to everyone else around the world!

Today I want to talk you through a business.

It charges $379 a year. It has no capital at risk, no custodian, no auditor, no prime broker, no high water mark, no redemption window and no benchmark. It cannot suffer a drawdown, because there is nothing to draw down. Its revenue does not move when its product turns out to be wrong. Demand for the product rises when the customer is frightened, which is a hedge most funds would pay real money to own.

Describe that structure to a compliance officer and she will tell you it is not a fund. She is right. It is a Substack….. no not ours… as we believe sharing is caring, not something to monetize…

That said, Michael Burry announced on 12 July 2026 that Cassandra Unchained had crossed 300,044 subscribers, 231 days after launch, spread across 212 countries. The list price is $39 a month or $379 a year, and there is a free tier, so the paid count is not public and he has not volunteered it. Take a 5% conversion, which is on the conservative side for a paid finance publication (again not ours). Fifteen thousand paying subscribers at $379 is $5.7 million of gross annual revenue. Substack takes 10%, card processing takes roughly another 3%, and about $4.9 million lands in the account of a man who no longer manages anyone's money.

Set that against the thing he closed. Scion Asset Management reported roughly $155 million of regulatory assets under management in 2025, down from $387 million five years earlier. Two percent of $155 million is $3.1 million, before any incentive fee, and the incentive fee requires him to be right, inside a specific window, above a high-water mark, with clients who are legally entitled to leave. At a 5% conversion the newsletter out earns the entire management fee base of the fund. At 10% it grosses $11.4 million, which is what a $570 million fund collects at two and twenty before it has generated a single dollar of performance.

He shut a $155 million fund and opened something with the fee economics of a fund three to four times larger, and the new one cannot lose money. That is the finest trade of his career, and no filing will ever show it.

The filings are the point. Scion moved to terminate its registration in November 2025, which is a Form ADV-W submitted through the IARD and effective on receipt. What departs with the registration is the scoreboard. There is no longer an annual ADV carrying a regulatory AUM figure that anyone can line up year against year. There is no longer a quarterly 13F under Section 13(f), the report that managers holding more than $100 million of listed securities must file within 45 days of quarter end.

Those 13Fs were half theatre in any case. The November 2025 filing showed put positions against Nvidia and Palantir, disclosed at notional, which generated headlines containing extremely large numbers. He then clarified that he had spent $9.2 million on options to sell Palantir at $50 in 2027, with the stock at $174 at the time. Scion's own footnote noted that puts may hedge long positions which are not themselves reportable. A form that discloses the notional of an option and not the premium paid for it is a form that discloses a rumour.

Strip the mythology out and the structure is clean. A reputation built on one correct call in 2007 has now been placed somewhere it can never be marked. He has no redemptions to meet, no registration to maintain, no positions to report and no outside capital to lose.

Which is a strange place to stand while warning everyone else about forced selling.

On 4 August he wrote that we may be near a major top and possibly a 1987 type fall, that he holds SOXX puts and a short position in the equity, and that a market rising on falling volatility forces vol targeting funds to lever up and drags other momentum strategies in behind them.

That sentence is correct. It describes long division.

A volatility targeting fund runs on one equation. Exposure equals target volatility divided by realised volatility. Set the target at 10% annualised, which is roughly the industry convention, and measure realised over a trailing 20 days. Realised at 8% gives you 1.25 times leverage. Realised at 16% gives you 0.625. The fund has just halved its equity exposure and nobody inside it has formed an opinion about anything. The fund does not think the market is dangerous. The fund does not think. A number moved, and the number sits in the denominator.

Notice what sets the size of the sale. Not the fall in prices. The ratio. A market that has been very quiet is a market where the denominator is small, leverage is at its maximum, and doubling a small number is the easiest thing in the world. Barclays sized the vol-target complex at $500 billion in February 2018 and estimated $225 billion of equity sales in the days following that VIX spike. The European Central Bank's Financial Stability Review of May 2020 counted around $300 billion across roughly 100 risk parity funds, up to $2 trillion in volatility strategies more broadly, and modelled a stylised risk parity portfolio that would have been required to sell assets worth 225% of its capital in March 2020, finishing the exercise about 25% in cash. In August 2024 the sell-side estimate was $30 billion to $50 billion of vol-control de-risking inside 48 hours.

The machine is real, it is enormous, and it requires no fear whatsoever to operate.

There is a detail here that almost nobody says out loud. Realised volatility counts the good days. The S&P 500 rose 3.58% last week and closed Friday at 7,757.64, a record. The Nasdaq Composite added 5.19%. Every one of those green sessions entered the denominator on exactly the same terms a red one would, which means the vol-target complex spent part of last week mechanically selling the rally to whoever was buying it. Being right about market direction and being on the right side of these flows are separate achievements.

So he has the mechanism. The year is where it falls apart.

On the morning of Tuesday 20 October 1987, the American stock market did not really open. Prices existed. Trades largely did not. Part of the reason was a settlement mismatch that reads like an accounting footnote and behaved like a tourniquet. Stocks settled in three days. Options and futures settled in one. A broker who sold your shares on Monday could not hand you the cash until Thursday, while the margin call on the futures leg was due Tuesday morning.

Picture the phone call. I sold your position. Where is my money. Thursday.

Multiply that by every clearing member in Chicago and New York and you get a market that has stopped, not because nobody wants to buy, but because nobody can pay for anything. Alan Greenspan's entire public response the following morning was a single sentence confirming that the Federal Reserve stood ready as a source of liquidity. Underneath it, the ten largest New York banks close to doubled their lending to securities firms during the week of 19 October, on terms Ben Bernanke later described in print as money-losing.

Every pipe that turned 1987 into one afternoon has since been replaced.

US cash equities moved to T+1 settlement on 28 May 2024, which closed the mismatch. Market-wide circuit breakers now halt everything at 7%, 13% and 20% declines in the S&P 500 under NYSE Rule 80B, and the first two levels stop applying after 3:25pm New York time, specifically so the market cannot be frozen into the close and reopened the next morning on top of a two-day queue of unexecuted panic. Individual names have their own limit up, limit down bands.

The modern version of the event he is describing has already run, and we know its shape. On 5 August 2024 the VIX printed 65.73 intraday, the largest single-day spike in its history, and closed the same session at 39. The TOPIX fell 12%. Vol-target funds and yen carry positions unwound together into the deepest liquidity hole of the year, and the whole thing was over in three sessions and fully retraced within weeks. That was the machine firing at close to full power against modern plumbing, and it produced a genuinely terrifying 48 hours rather than a 22.6% afternoon.

You can be exactly right about the mechanism and completely wrong about the shape it takes. The 1987 comparison is not analysis, it is casting. It puts the reader in a film with a known ending and a legendary supporting performance, and films are much easier to sell than probability distributions.

July is the funnier evidence, and it embarrasses both sides of this argument. SOXX fell 21.2%, its worst month since 2002. The short worked. Investors responded by pushing a record $6.9 billion into that same fund while it was falling, taking 2026 inflows to roughly $14.67 billion through 31 July, and the ETF closed at $528.81 on 6 August against a 52-week range of $237.10 to $655.95. The forced sellers of July were not the vol-target funds at all. They were leveraged AI-dedicated hedge funds, one of which, Situational Awareness, detonated outright. Bank of America still has hyperscaler capital spending running at roughly $859 billion for calendar 2026, up 79% on the year. A short that is right about the tape for a month while the fundamental spending accelerates and record money arrives to fund the other side of it is a trade with an unusually short shelf life.

This issue is about scoreboards, so here is ours. On 30 July this letter argued against being short SpaceX into the 6 August unlock, with invalidation set at two consecutive closes below $110.85 before 4 August. The stock breached that level once, on 5 August, falling almost 14% to an all-time closing low of $108.27, one calendar day after the invalidation window shut. Then 911.5 million shares unlocked, the stock rose 6.1% to $114.92 on the day itself, and on Friday it added 15.8% on a revenue beat. The call was right and the reasoning held, with a margin of one day on the invalidation clause. That goes in the ledger here because a track record you grade yourself is worth precisely what you paid for it.

CAN WE TRADE THIS? (Of course we can… this is Caffeinated Capital)

Thesis. The vol target complex is near maximum leverage because realised volatility is low, not because anyone involved is bullish. That is a fragility rather than a forecast, and fragility is at its cheapest to insure at exactly the moment it is at its highest. Own the convexity the machine is structurally short. Do not sell the equity because a newsletter has cast you in a 1987 remake.

Expression. Keep the index exposure. With the S&P 500 at 7,757.64 and VIX at 14.90 as of Friday's close, buy three to six month S&P 500 put spreads roughly 5% to 15% out of the money, sized so total premium runs 50 to 75 basis points of the portfolio per quarter, and book it as an expense the way a landlord books building insurance rather than as a position you expect to make money on. Where the mandate allows, express any semiconductor caution by writing covered calls against existing SOXX or SMH holdings into strength rather than shorting outright. That 52-week range of $237.10 to $655.95 is what a linear short does to a person in a whipsaw.

Invalidation. Twenty-day realised volatility on the S&P 500 holding above 15% with VIX sustained above 22 for five consecutive sessions. At that point the deleveraging has already happened, the insurance has already repriced, and you are buying a fire extinguisher from a man standing in the ashes. Take the structure off and stop paying for it.

Horizon. Three to six months, spanning the September Federal Reserve meeting, where Friday's payrolls print of minus 23,000 against an expected plus 80,000 has already moved the odds, and the September Nasdaq-100 rebalance that will reflect SpaceX's enlarged float.

The reusable instrument this week takes about ninety seconds and works on anything. Call it finding the mark. For any asset, any manager, any warning and any confident stranger, ask who prices this, how often, and what happens to that person when the price turns out to be wrong. The answer sorts the world into two useful piles. Entities that get marked daily by a formula become forced sellers, which makes them predictable and worth front-running. Entities that cannot be marked at all become opinion-havers, which makes them entertaining and worth discounting. Private credit funds carrying loans at par, venture portfolios held at the last round, an insurer's Level 3 book, a fund that has deregistered and a newsletter warning you about all of the above sit in the same pile for the same reason. The daily-marked tell you what will be sold and roughly when. The unmarked tell you where the surprises are being stored.

Cassandra's curse, in the original, was that she prophesied truly and nobody believed her. Three hundred thousand people believe him, in 212 countries, at $379 a year. Nobody checks.

Another big session in the books!

Yesterday DBS fell on heavy volume and we called it profit taking into strength rather than a top. It took one session to find out which. Today DBS rose 2.08% to a fresh record S$75.08 on turnover of S$566.42 million, the single heaviest day for the stock in this entire run, comfortably above the S$468 million peak of mid July. OCBC jumped 2.44% to S$29.33 and UOB added 1.21% to S$43.58 the day before it reports. The banks did not just come back. They came back at a record, and they came back loud.

This is what conviction into earnings looks like. The three lenders that are half the index all rose together, on the heaviest bank volume of the month, in the same week the sector reports. Yesterday's dip was a single day of profit taking and today the buyers returned in size. When the largest, most liquid name in the market makes a new high on its busiest day of trading, that is not a market drifting. That is a market deciding.

And tonight, after the close, the running character of this note for a month finally shows its books. SGX rose 1.29% to S$24.32 today, off the lows it spent July and half of August sulking at, as the market started to believe what its own trading statistics have been shouting all year. The house reports at six. The mood, at last, is coming around to the numbers.

Underneath the green, the currency story got its sharpest illustration yet, and it lived on the REIT board. As the banks celebrated a stronger Singapore dollar, the US-dollar trusts were taken apart: the standout being Acrophyte Hospitality Trust, down 14.35%, but on a real catalyst this time rather than the usual sub-cent theatre. More on that below.

THE TAPE

The Board, Top to Bottom

Our read of the SG close, 6 Aug 2026. Banks and a defensive retailer up top, the property trusts below the line.

DBS at a record S$75.08 on S$566.42 million is the number to carry into the weekend. For three weeks this note has tracked the turnover in DBS as a conviction gauge: S$285 million on the quiet days, S$468 million at the July peak, and now a fresh high on the heaviest day of them all. A record set on the biggest volume of the run is worth far more than a record set on a whisper, and it comes two days before the bank reports its own half. The market is positioning for a good number, out loud.

OCBC led the trio at 2.44% and UOB rose 1.21% into its Friday result. UOB is the interesting one: yesterday it announced the sale of its asset management arm to Allianz, and today it firmed ahead of tomorrow's print. The disposal gives management a clean strategic story to open with, and the market is giving it the benefit of the doubt on the numbers underneath. Tomorrow we find out whether the benefit was earned.

DFI Retail was the top gainer at 3.71% to S$3.91, the supermarket-and-health-and-beauty operator continuing its slow rehabilitation as a focused, cash-generative consumer name. On a day the banks surged, a defensive retailer leading the whole board is a small reminder that this rally is not one trade, it is several running at once.

SGX rose 1.29% to S$24.32 ahead of tonight's results, the clearest sign yet that the month-long funk in the exchange's shares was sentiment rather than substance. The stock spent July and early August near the bottom of its own index while its trading business had a record year. Today, for once, it climbed into the print.

Keppel DC REIT was the worst name at 2.21% lower, a near-perfect round trip of yesterday's 3.20% gain that led the board. Up the most one day, down the most the next, with nothing changing at the company in between. CapitaLand Ascendas, Mapletree PanAsia and Seatrium filled out a shallow red tail, with Seatrium continuing to hand back a little of its strong post-earnings run.

THE TAPE, PART TWO

The REIT Board

S-REIT movers, 6 Aug 2026. The US-dollar and European trusts wore the strong Singapore dollar, with one real result.

This is the currency trade drawn in a single picture. The bottom of the board is a clean sweep of foreign-currency names: Acro HTrust down 14.35%, Manulife US REIT down 4.00%, and the Stoneweg European pair both off around 3.7%. When MAS deliberately strengthens the Singapore dollar, a trust that collects rent in US dollars or euros and reports in Singapore dollars faces a translation headwind on top of whatever its buildings are doing, and the market has spent this week marking exactly those names down. The gainers, such as they were, were the modest Singapore-dollar and Asia names led by DigiCore at 1.05%.

The Acro HTrust fall is the one that is real, and it deserves separating from the noise. Unlike our departed friend Landmark, whose double-digit headlines come from single sub-cent ticks, Acrophyte Hospitality Trust released its first-half results this morning, and a 14.35% drop to S$0.197 is the market's verdict on the numbers, not a rounding artefact. A US hotel portfolio reporting into a firmer Singapore dollar and a softening American consumer is exactly the profile that gets punished right now. If a client asks about a REIT that genuinely moved today, this is the one to talk about. The rest of the red on this board is the currency, doing what a tighter currency does.

THE WRAP

Records Up Front, Results After Dark

The through-line this note has drawn since the MAS surprise is now the entire market, and today was its cleanest expression. A stronger Singapore dollar and a hawkish Fed reward the banks and punish the foreign-currency trusts, and today did both at maximum volume: the lenders to fresh records, the US-dollar REITs to the floor. There is no ambiguity left in the signal. Own the names that benefit from a firmer local dollar and higher-for-longer rates, be careful with everything that does not.

The next twenty-four hours turn the theme into numbers. SGX reports tonight after the close, and UOB reports tomorrow. Between them they answer the two live questions: whether the exchange's record trading year became record profit, and whether the largest bank's margins are holding as the currency is pushed higher. The market has voted with volume ahead of both, DBS at a record and SGX climbing into its print. Now the results have to ratify the vote. A market that makes new highs on its heaviest volume the day before earnings is confident. Tomorrow morning it finds out if it was right.

AROUND THE GROUNDS

News You'll Get Asked About

•  SGX reports FY2026 tonight after the close, and the setup is all about the gap between the business and the stock. The exchange had a record trading year: full-year securities turnover up 35% to S$455.7 billion, daily average value at an eighteen-year high, retail activity the strongest in twelve years, and first-half adjusted profit already up around 12% to S$357 million. The stock, meanwhile, spent the month as the index's dog. Tonight's numbers show whether the second half kept the pace and whether costs stayed inside the 4 to 6 percent guidance. The published dividend path points to about 44.5 cents for the year. Today's 1.29% rise says the market is finally leaning toward the numbers.

•  Acro HTrust's 14.35% fall was a results-day move, not sub-cent noise. Acrophyte Hospitality Trust released its first-half figures this morning, and the market did not like them. A US hotel portfolio faces the double bind of a firmer Singapore dollar on translation and a cooling American consumer on demand. It is the clearest single example on today's board of why the foreign-currency trusts are the wrong place to be while the local dollar is being pushed up.

•  UOB reports tomorrow, one day after selling its asset manager. The disposal to Allianz for S$555 million gives management a clean strategic narrative about pivoting to distribution-led wealth, but the S$330 million gain is a 2027 event and will not be in this print. The numbers that matter tomorrow are net interest margin into a stronger Singapore dollar, wealth and fee income, and provisions. The stock firmed 1.21% today into it.

•  DBS made its record on the heaviest volume of the run and reports its own half shortly. The record close on S$566 million of turnover is the market pricing in a strong result. That cuts both ways: it rewards a good number and leaves little room for a soft one. The bank is priced for success into its print.

THE READ

Where We'd Lean

Two results tonight and tomorrow, and a market that has already placed its bets. Directionally, here is where the desk sits.

Banks: yesterday's dip was a day, and today settled it, but records on record volume raise the bar for the prints. The buyers came back in size, which is genuinely constructive, and a tighter policy stance is marginally supportive for margins. The flip side is that DBS at a record and OCBC firm leave very little room for a disappointing quarter. We stay constructive on the franchises and alert to the valuation. Let the actual margin numbers, tonight and tomorrow, set the next move rather than the pre-earnings enthusiasm.

SGX into tonight: we are more constructive than the month's price action implied, and today the market started to agree. A record trading year that spent the month ignored by its own shareholders is the kind of gap results can close. If the second half held the pace and costs behaved, the funk looks like an opportunity. The steadily rising dividend is the anchor. The risk is simply that a stock that has already climbed into the print has less room to celebrate a merely good one.

US-dollar and foreign-currency REITs: the macro is actively against them, and Acro's results-day collapse is the warning shot. A firmer Singapore dollar is a direct, mechanical drag on any trust that earns abroad, before the underlying assets even enter the conversation. We would stay firmly at the Singapore-dollar, quality end and treat the headline yields on the foreign-currency names as the trap they have been all week.

Seatrium and the industrials: still constructive, but the easy re-rating is behind them. Seatrium handing back a little each day after a strong result is healthy digestion, not deterioration. The order book and the cash-conversion proof point are intact. We would accumulate quality cyclicals on this kind of weakness rather than chase the strength that has already happened.

DESK PREP

Ready for the Morning

The results that define the next two sessions:

•  Tonight, after the close: SGX FY2026. The month's running character reports at last. Watch second-half trading momentum against the record full-year statistics, cost discipline against the 4 to 6 percent guidance, and the final dividend on the path toward 44.5 cents.

•  Friday: UOB first-half results. Net interest margin into a stronger Singapore dollar, wealth and fee income, provisions, and any update to the 2030 wealth-doubling ambition now that the asset manager is being sold. The disposal gain is a 2027 event, so it will not appear here.

•  DBS reports its own half shortly, into a record share price on record volume. Priced for a strong number, with little margin for a soft one.

•  Then the tail: SIA's dividend goes ex on 11 August and ST Engineering reports on 19 August. The macro grind of a firmer Singapore dollar, the oil price and the September Fed question runs underneath all of it, pointing the same way.

THREE THINGS TO SAY BEFORE 9AM

•  "The banks came back on the heaviest volume of the run." DBS to a record S$75.08 on S$566 million, the busiest day of the month. Yesterday's dip was a single session, and today settled it.

•  "SGX reports tonight, and the stock already believes." Up 1.29% into a print that should show a record trading year. The month-long funk was sentiment, not substance.

•  "Read the currency, not just the yield." Acro HTrust down 14% on results, the US-dollar trusts hammered, as a firmer Singapore dollar does its mechanical damage. The foreign-currency yield is a trap right now.

ONE MORE THING

The House Finally Believes Itself

Yesterday this note said the market had spent a month acting as though SGX's business had never been worse, even as its trading statistics screamed the opposite, and that our money was on the numbers rather than the mood. Today the mood blinked first. The exchange that has been the worst company in its own index for most of the month rose 1.29% into its own results, the market finally lining up behind the record year the tables clearly had.

It is a small thing, a single green session ahead of a print, but it is the right kind of small thing. For weeks the croupier counted a record haul at packed tables while its own shareholders looked at the stock and concluded the casino must be empty. Tonight at six the books settle the argument, and today, for the first time in a month, the share price walked in expecting good news rather than dreading it. Sometimes the mood comes around to the numbers, and sometimes it does so exactly one session before it has to. See you tomorrow, with the tally in hand.

The Daily Morning Brew

Indicative and directional only. Nothing here is investment advice or a recommendation. Prices and levels are sourced from end-of-day SG tape (6 Aug 2026) and public market data. Macro, analyst and news items are paraphrased from public sources and current as at publication. Views are the desk's own and may change without notice.

Good morning,

It's been awhile since we published The letter of marque and in true fashion our thesis has thus far been fair.

Today, one hundred and sixteen billion dollars of SpaceX stock is set free to trade, in what is very plausibly the largest single day jump in the tradeable supply of one company in the history of markets. Nine hundred and eleven million shares, unlocked at once, dumped onto a market that until this morning had a float of fewer than two hundred and eighty million. Roughly four times as much stock as exists today, cleared to sell, on a single Thursday. Every headline says brace for the crash.

The stock might go up.

Not because that would be pleasant, and not as a contrarian party trick, but because of a specific number that almost everyone is reading backwards, and that we can rebuild together on the back of a napkin in about thirty seconds. Stay with the arithmetic, because it is the whole issue.

Back in May, before this stock existed as a thing you could buy, we also wrote a piece called The Dam . The argument was not especially original and we said so: SpaceX would go public with a tiny sliver of its shares trading, the price would float up on scarcity and mythology into a thin, airless market, and then on a scheduled date the lockups would expire, a wall of insider stock would hit, and the whole thing would reprice downward the way every mega IPO of the last twenty years has. We ended it, and we quote ourselves only because the receipt matters today, with three words: bring patience or do not come.

Here is the receipt. SpaceX priced at $135 in June, spiked to $225 within four days on exactly the scarcity we described, and has since fallen to around $108, down roughly 20% from its offer and more than 50% from that first-week high. The dam we described has been leaking hard for six weeks. Today is only the morning the gate is officially, ceremonially, opened.

So the easy column to write is the victory lap: we told you the dam would break, here it breaks, aren't we clever. We are not writing that column, because it would be wrong, and the reason it would be wrong is the single most useful idea in this whole episode, the thing that separates people who understand market plumbing from people who read the calendar.

The flood everyone is standing outside to watch has, quietly, already happened. The dam breaks today. The water left weeks ago, carried out by the very people betting against it.

Now the napkin, because this is a Number issue and the number is the argument. As of this week, about 35% of SpaceX's available shares are sold short. Do the small sum with me. The tradeable float is under 280 million shares; 35% of that is roughly 95 to 100 million shares that have been borrowed and sold by people betting the price falls. To sell a share short you must first borrow a real one and hand it to a buyer today. So about 95 million shares have already been delivered to buyers, weeks early, by sellers who now owe those shares back and whose single most obvious place to buy them cheaply is the flood unlocking this morning. Ninety-five million shares of today's supply were sold into the market before today. The reservoir behind the gate was drained, by hand, in advance.

That 35% is one of the highest short-interest readings for any recently public company, ever, and everyone is reporting it as raw pessimism about the unlock. It is that. But mechanically it is something almost nobody is saying out loud: it is the measure of how much of today's flood has already been carried downstream. Each of those short sales was a bucket of tomorrow's water, delivered before the gate ever opened.

A short sale isn't a mood. It's the physical delivery of a future share into the present. Thirty-five percent short into a known unlock means a third of the flood already came, quietly, on the way down.

This is the same trade, run in reverse, that hedge funds do when a stock joins an index. When everyone knows a wave of forced buying is coming on a fixed date, arbitrageurs buy ahead of it and sell into it, and the price adjusts before the event, not on it. A lockup expiry is the mirror image: everyone knows a wave of forced selling is coming on a fixed date, so arbitrageurs sell ahead of it and buy it back into the supply, and the price adjusts before the event, not on it. The 35% short interest is not a prediction that the stock will fall today. It is the measure of how much of today's fall has already been dragged into the past.

The market did not fail to see the dam coming. It saw it so clearly that it drained the reservoir in advance and left the ceremony of the gate opening for the television cameras.

Which is why the genuinely correct read of today, the one the calendar-readers will get wrong in both directions, is this. The expected price of SpaceX after the unlock should be roughly the price before the unlock, because a scheduled, universally-known supply event is the single most efficiently pre-priced thing a market ever handles. If the stock craters today, it will not be because the supply arrived; it will be because the supply arrived and turned out to be even larger than the shorts had already sold, which is possible but is a different and much narrower bet. And if the stock rises today, which would violate every intuition, it may be for the most beautiful reason in finance: once the feared event actually happens and the forced selling is finally, visibly done, the shorts have to buy their shares back, and there is no one left to sell to them cheaply. The thing everyone feared becomes the thing that clears the overhang.

We watched this exact mechanism three weeks ago in a different costume, and regular readers will remember it: a leveraged fund blew up on crowded AI bets, was force-sold into the market, and the instant the forced selling stopped, the very stocks it had been dumping turned and ripped, handing the firm that bought the wreckage a fortune inside a week. The rule holds again here. A market does not bottom when the bad news peaks. It bottoms when the last forced seller runs out of stock to sell, and those are different mornings, sometimes weeks apart. A price pinned down by mechanical, obligated selling is a coiled spring, and the release is not sentiment improving. It is the seller, simply, finally, being empty.

None of which resolves the question our May piece actually posed, and here honesty requires drawing the line clearly, because there are two completely different questions tangled together in today's headlines and the whole art is keeping them apart. The first question is the plumbing question: does the unlock crash the stock today? And the answer, for all the reasons above, is probably not in the way people expect, because the plumbing has been quietly draining for six weeks. But the second question is the one that matters for your actual wealth, and the plumbing has nothing to say about it.

The second question is the one from The Letter of Marque, the very first thing we ever wrote about this company. Strip away the lockup mechanics and the short interest and the index-inclusion timing, and SpaceX's governance reduces to a single wager: do you trust the captain? Musk holds his shares under a separate lockup that runs to the middle of 2027; he is not selling into today's flood. His control is absolute, his other companies compete with this one by charter, and the stock you can buy today is a licence to follow him wherever he goes with no way to steer and no way to get off except selling to the next believer. The unlock changes the supply of that licence. It changes nothing about the terms printed on it.

So the two questions pull apart cleanly, and you should hold them in separate hands. The lockup is a technical event that the market has very efficiently pre-arranged, and betting on a guaranteed crash today is betting against thirty-five percent of the float that already did your selling for you. The company is a decade-long bet on one man's ability to keep pulling rockets out of the fire, and no supply schedule will ever tell you whether that bet pays. The mistake, the expensive one, is letting the noise of the first question talk you into or out of the second. Today is loud. Today is also, almost certainly, not the day anything about the actual company is decided.

THE TRADE

Don't Trade the Gate. Trade the Silence After It.

The structural edge is that a scheduled, universally-anticipated supply event is the most efficiently pre priced moment in a stock's life, and 35% short interest is hard proof the selling has been pulled forward. Betting on a big directional move on the day itself is betting against a crowd that has already done the trade. The opportunity is not the gate opening; it is what the stock does in the days after, when the forced supply is finally exhausted.

The expression, for the patient: do not chase a crash today and do not chase a relief pop today; both are traps for the calendar reader. Instead, watch for the tell that the overhang has cleared, several sessions of unusually high volume during which the stock stops making new lows despite the scary headlines. That combination, heavy supply absorbed without new lows, is the signature of a forced-selling bottom, the same one that preceded the sharp reversal in the AI names three weeks ago. That is the entry, not the calendar date. For the SPV holders about to receive their first distributions from the pre IPO vehicles, the same logic applies in reverse: you are now part of the supply, and selling into a universally-known unlock alongside everyone else is the worst timed exit available; the patient distribution beats the panicked one.

Invalidation: If the stock breaks decisively to new lows on the unlock and keeps making them on rising volume, the forced sellers were larger than the shorts pre-positioned for, the overhang is not cleared, and there is a second, larger tranche plus a December wave still to come. Wait for the low-that-holds; do not anticipate it.

Horizon: Days to weeks around each unlock tranche, with the larger structural repricing still running toward the 2027 insider releases. The company question remains a multi-year one, entirely separate from this.

THE PORTABLE TOOL

Here is the tool, and it fits on a matchbook: the event everyone is watching for is rarely the event that matters, so watch the silence just after it. When a supply-moving event is scheduled and everyone can see it coming, don't ask what happens on the date, ask what already happened before it. Efficient markets pre-price the predictable, and short interest is your gauge of how much has been dragged into the present. High short interest into a known unlock does not mean the crash is coming. It usually means the crash already came, quietly, and the date itself is when the sellers run out and the spring is set. Amateurs watch the gate. Professionals watch the reservoir empty weeks earlier, and mark the calendar for the strange quiet morning after, when the forced seller is finally spent and the price is free to move.

In May we said the dam would come down and the water it held back would arrive all at once. We were right about the dam and wrong about the drama, because the water did not wait for the gate. It left through thirty-five percent of the float, sold forward by short sellers who saw the flood coming and drained the reservoir before the ceremony began.

The captain, meanwhile, has not moved. His shares stay locked until 2027, his control undisturbed, his licence to sail wherever he pleases entirely intact. Today the market discovers how much water was really behind the gate, and it will be a loud, heavily-televised discovery. But the thing worth owning was never the water level on any single morning. It was the answer to the question we asked at the very beginning, the one no unlock schedule will ever settle.

Everyone is watching the dam break. The only question that ever mattered is whether you trust the man who built it, and he isn't selling today, or telling.

The Daily Morning Brew is an independent financial newsletter. Nothing here is investment advice. All market levels cited were verifiable at the time of writing.

Ding ding ding… thats a wrap for a big day here in the Lion City!

The tape itself was a quiet, slightly heavy affair. The index has been drifting down from the record 5,713 it set on 29 July, and today the two heaviest names did the drifting: Jardine Matheson fell 2.37% to S$63.44, and DBS eased 1.28% to S$73.55. A data centre trust and a couple of industrials held the top of the board. On any other Wednesday that would be the whole note.

But the story today was not on the screen, it was in a press release. UOB agreed to sell UOB Asset Management to Allianz Global Investors for S$555 million, two days before it reports first-half earnings on Friday. After a fortnight in which this note kept insisting the important things happen off the leaderboard, here is a clean example: the biggest development in Singapore finance today was a bank deciding what business it no longer wants to be in.

THE DEAL

UOB Sells Its Fund Manager

The facts first. Allianz Global Investors, the fund arm of the German insurer, is buying all of UOB Asset Management, which runs about S$42 billion of client assets across eight Asian markets: Singapore, Brunei, Indonesia, Japan, Malaysia, Taiwan, Thailand and Vietnam. The price is S$555 million, the deal includes the unit's excess cash and a ten-year distribution agreement, and all 500 staff move across. UOB expects a pre-tax gain of around S$330 million and a roughly 14 basis point lift to its CET1 capital ratio. It is subject to regulatory approval and closes in 2027. AllianzGI, which runs close to S$900 billion globally, beat out Amundi, KKR and Temasek's Seviora to win it.

Strip away the mechanics and the deal is one sentence: UOB has decided it would rather sell investment products than manufacture them. It is keeping the customer, the shelf and the advice, and handing the capital-hungry business of actually running the funds to someone with the global scale to do it better.

Our read: this is the right move, and a fashionable one, for a reason. Asset management is a scale game, and it has been getting less forgiving to the sub-scale every year. The global giants keep cutting fees, the passive tide keeps rising, and a S$42 billion regional manager sits in an awkward middle: too big to be a boutique, too small to compete with a BlackRock or an Amundi on cost. That same S$42 billion is worth far more bolted onto AllianzGI's platform than it is trapped inside a Singapore bank. When an asset is worth more to the buyer than the seller, selling it is not surrender, it is arithmetic.

The economics are fine rather than thrilling, and it is worth being honest about which. The price is around 2.5 times the unit's S$223 million net asset value and a little over one percent of assets under management, a fair-to-full outcome for an Asian retail manager and a good one given UOB was the seller into a competitive auction. But the S$330 million gain and the 14 basis point capital bump are rounding errors for a bank UOB's size. Nobody should mistake this for a capital event. It is a strategy event that happens to come with a modest cheque attached. The real prize is the ten-year distribution deal and what it says about where UOB thinks the money is: in owning the affluent customer and selling them everyone's funds, not in running its own.

The honest counter argument, because there is one. UOB is handing a growing franchise in exactly the markets it should care most about, Indonesia and Vietnam, to a European competitor, and swapping recurring management fees it controls for distribution fees it shares. If wealth is genuinely the growth story, and UOB says it wants to double wealth income by 2030, then owning the factory in the fastest-growing corner of Asia had real option value, and that option has now been sold. We think the focus argument wins, because a bank that tries to both manufacture and distribute at sub-scale usually does neither especially well. But a sceptic could fairly call this simplification for its own sake, and they would not be obviously wrong.

What it means for Friday. Do not expect the S$330 million in the numbers: the deal closes in 2027, so the gain lands in a future period, not this week's H1 result. What the timing does buy UOB is a narrative, the chance to walk into its earnings on the front foot with a clean strategic story about pivoting to open-architecture wealth. That is smart stagecraft. Just keep the stagecraft and the profit-and-loss statement in separate boxes when you read the print. Friday's numbers are about net interest margin into a tighter Singapore dollar, wealth and fee income, and provisions. The disposal is about the next five years, not the last six months.

THE TAPE

The Board, Top to Bottom

Our read of the SG close, 5 Aug 2026, showing the confirmed movers on the day.

DBS was the most-traded name and it fell, and the combination is the detail to notice. Down 1.28% to S$73.55 on turnover of S$435.86 million, the heaviest single-name volume in weeks, back up at the levels last seen in mid-July. On Monday DBS made a record at S$74.45 on quiet volume. Today it gave a little back on loud volume. A leader that sells off on rising turnover is distribution, people taking profits into strength, and it is worth more respect than a drift lower on thin trade.

Jardine Matheson was the worst name at 2.37% lower, its second straight decline and the continuation of a pattern this note flagged after its 30 July result. The Morgan Stanley re-rating drove it up more than four percent in a single July session on the promise of buybacks and asset recycling. The numbers have since had to do the arguing, and the stock has been handing back the anticipatory run. The thesis is intact. The easy money was made before the print, as it usually is.

Keppel DC REIT led at 3.20% to S$2.26, and a data-centre trust topping the board on a day the banks fell is a small but real tell. The artificial-intelligence and data-centre theme still commands a genuine growth bid, and a quality Singapore-dollar-earning REIT is a completely different animal from the US-dollar yield trusts that were sold on Monday as the currency firmed. SATS at 2.77% and Wilmar at 1.79% round out a green list of hard-asset, industrial and agribusiness names, the same corner of the market that has quietly worked through every twist of the past three weeks.

THE TAPE, PART TWO

One Line on the REITs

Keppel DC REIT led the trusts as well as the index, which tells you the bid was in quality and growth rather than yield. And at the very bottom, inevitably, Landmark REIT "lost" 12.5% to close at S$0.007. For those keeping score at home, this is the fifth time our favourite sub-cent trust has generated a double-digit headline out of a single one-notch move, this time from eight-tenths of a cent to seven. We long ago ran out of new things to say about it, which is itself the point. Read the price before the percentage, forever.

THE WRAP

Two Days of Earnings That Actually Matter

The market spends its days watching the stock that moved three percent, and the direction gets set by the things that do not fit on a leaderboard. Today it was a strategy decision at UOB. The next two days it is the numbers behind two of the names this note has followed all month. SGX reports tomorrow after the close, and UOB reports Friday. Between them they will tell us whether the exchange's record trading year translated into record profit, and whether the largest bank's margins are holding up as the Singapore dollar is deliberately pushed higher. The index easing off its 29 July record is the right posture into that: not fear, just a market declining to place big bets in the forty-eight hours before it gets real information. The macro has not changed since Monday. A tighter Singapore dollar and a hawkish Fed still reward the order book and penalise the bond proxy. Earnings are now the only thing that can move the argument.

AROUND THE GROUNDS

News You'll Get Asked About

•  The UOBAM sale is the headline, and the distribution deal is the substance. UOB keeps selling both UOBAM and AllianzGI funds through its network for the next decade, so it monetises the customer relationship without carrying the manufacturing. AllianzGI gets long-term access to UOB's Southeast Asian retail base and lifts its Asia-Pacific assets above the equivalent of S$250 billion. Both sides get the thing they actually wanted, which is usually the sign of a sensible deal rather than a distressed one.

•  SGX reports FY2026 tomorrow after the close, and the setup is loaded with irony. The exchange has been near the bottom of its own index for most of the past month, yet the business had a record year: full-year securities turnover rose 35% to S$455.7 billion, daily average value hit an eighteen-year high, and retail activity was the strongest in twelve years. First-half adjusted profit was already up around 12%. The question tomorrow is whether the second half kept pace and whether costs stayed inside the 4 to 6 percent growth guidance. The published dividend path points to about 44.5 cents for the year.

•  UOB reports Friday, and it is the one this note has flagged for weeks. De-risked hard in mid-July on no news, quietly recovered since, and now reporting into a backdrop where a tighter policy stance is, at the margin, helpful for bank margins. The disposal gives management a clean strategic story to open with. The numbers to read are net interest margin, wealth and fee income, and provisions.

•  The macro remains a grind rather than an event: the Singapore dollar has an official tailwind after the MAS tightening, oil is the swing factor, and US data is slowly firming the case for a September Fed hike. None of it has a single dramatic date. All of it points the same way, and it is not the way the long-duration trades want.

THE READ

Where We'd Lean

Two earnings prints in two days give the desk something concrete to lean on. Directionally, here is where we sit.

UOB into Friday: the disposal is a strategic positive, not a Friday-numbers event, so separate the two. We like the pivot to capital-light, distribution-led wealth, and we think the market will too once it stops conflating the 2027 gain with this week's print. The setup is the one we have described for a fortnight: a de-risked bank reporting into a marginally supportive rate backdrop. Constructive, and we would let the actual margin number, not the press release, set the direction.

SGX into tomorrow: the business had its best year in nearly two decades while the stock spent the month as the index's dog. If the results confirm the trading boom carried into the second half and costs stayed contained, that month-long share-price funk starts to look like an opportunity rather than a verdict. The steadily rising dividend is the income anchor underneath it. We are more constructive on the numbers than the recent price action would suggest.

DBS and the banks: a record leader selling off on the heaviest volume in weeks is profit-taking, not a top. Nothing has broken. But it is a reminder that the banks are richly priced into their own earnings, and that leaves less room for a soft quarter than there was a month ago. Own the quality, do not chase the record.

Data-centre and growth REITs versus yield trusts: the split is now the whole REIT story. Keppel DC leading while the US-dollar trusts lag is the market correctly separating a growth theme from a rate trade. We would stay at the quality, Singapore-dollar, structural-growth end and continue to avoid reaching for headline yields in the foreign-currency names, where the firmer local dollar quietly erodes the payout.

DESK PREP

Ready for the Morning

The two prints that define the week:

•  Thursday 6 August, after the close: SGX FY2026 results. The running character of this note for a month finally shows its books. Watch second-half trading momentum, cost discipline against the 4 to 6 percent guidance, and the final dividend.

•  Friday 7 August: UOB first-half results. Net interest margin into a stronger Singapore dollar, wealth and fee income, provisions, and any change to the 2030 wealth-doubling ambition now that the asset manager is going. The disposal gain itself is a 2027 event, so do not look for it here.

•  Then the tail: SIA's dividend goes ex on 11 August and ST Engineering reports on 19 August. The airline's high price against cautious analyst targets is still unresolved.

THREE THINGS TO SAY BEFORE 9AM

•  "UOB sold the fund factory and kept the shop." It is selling asset management to Allianz for S$555m to focus on distributing wealth. A modest gain, closes in 2027, so it is narrative for Friday, not numbers.

•  "DBS fell on the heaviest volume in weeks." A record leader pulling back on big turnover is profit-taking into strength. Respect it, do not panic about it.

•  "SGX had a record year while its stock sulked." Tomorrow's numbers show whether the month-long funk in the share price was an opportunity the business did not deserve.

ONE MORE THING

The House Shows Its Books

For a month this note has run a quiet gag at SGX's expense. The exchange that takes a clip of every trade in Singapore spent most of July and half of the month near the bottom of its own leaderboard, the croupier somehow losing at its own tables while everyone else played. Tomorrow, after the close, it finally has to show the books.

And here is the thing the share price has been ignoring: the house had a spectacular year. Securities turnover up thirty-five percent, trading volumes at eighteen-year highs, retail investors piling in at a pace not seen in over a decade. The tables were packed all year. The croupier counted a record haul. And its own stock spent the month acting as though business had never been worse. One of those two things is wrong, and tomorrow at six o'clock we find out which. Our money, for what it is worth, is on the numbers rather than the mood. See you tomorrow.

The Daily Morning Brew

Happy Tuesday all!

Let us start today's note with a fun little pitch, because it is a genuinely good one, and any honest teardown begins by making the opposing case better than its own salesmen do. Here it is at full strength. South Korea is home to the two companies that make the memory chips the entire artificial-intelligence boom runs on, Samsung and SK Hynix, which together are more than half its stock market. Those companies are printing money; forward earnings for the index grew a barely-believable 366% while the whole thing was falling apart. And after a crash of nearly 40% in twenty seven trading days, the market now trades at about five times forward earnings. Five. That is a decade low, half the multiple of China, less than a third of Taiwan, less than a third of Japan. It is, on paper, the cheapest major market on the planet, wired directly into the most powerful industrial trend of the decade.

The brokerages have done the obvious arithmetic and reached the obvious conclusion. One major American bank upgraded the whole market to overweight last week, calling the crash a leverage washout and slapping a target on the index some 36% above where it sits, arguing the selloff was mainly technical and the fundamentals never changed. It is the single most seductive setup in investing: a wonderful business, temporarily on fire, marked down by a panic that has nothing to do with its earnings. Buy the fear. Every instinct you have been trained to trust says back up the truck.

I want to tell you why I think that instinct is, this time, a trap, and why the number everyone is quoting is the wrong number entirely.

A market is not cheap because the price fell. A market is cheap when the price is low and someone is still willing to buy. Korea just spent a month teaching an entire generation to never buy again.

Here is the thing the five-times-earnings crowd is not pricing, because it does not appear on any valuation screen, and it is the entire point of this letter. The most valuable asset a stock market owns is not its listed companies. It is the trust of the people who are supposed to keep their money in it. That trust is a form of collateral, invisible and uncollateralised and absolutely load-bearing, and Korea has just set fire to a decade of it in a single quarter. You cannot see the damage in the P/E ratio. You can only see it in who has left the building and whether they are ever coming back.

Walk through what actually happened to the people who were supposed to be this market's foundation, because the mechanism matters. For decades Korea suffered from what the world politely called the Korea discount, a permanent markdown for weak governance and shareholders who came last. This year a reformist president and a genuine chip boom finally seemed to break the curse, and ordinary Koreans, who had spent their lives being told their home market was a value trap, finally believed. They poured in. They filled the exact hole left by foreign institutions, who were quietly selling the entire way up. Retail bought roughly $9.4 billion of the new single-stock leveraged funds against about $2 billion from foreigners. The locals were the marginal buyer, the believer, the bid of last resort. And then the government handed them a flamethrower and called it a toy.

The single-stock leveraged ETF, which we have written about before and will now watch finish its work, is a product that must buy as its stock rises and sell as it falls, mechanically, every day, amplifying every move. At the June peak, a 5% move in SK Hynix could trigger rebalancing worth 40% of the stock's entire daily volume. When the chips wobbled, the machine did what it is built to do and turned a wobble into a collapse, and the most popular SK Hynix leveraged fund lost as much as 84% of its value from the June peak. Not 84% of the froth. 84% of the money.

Now the number that actually matters, the one that will still be true when the index has bounced and the brokerages are taking victory laps. An estimated 360,000 brokerage accounts were forcibly liquidated, sold out at the bottom by margin calls, and 62% of them belonged to Koreans under the age of 35. Read that twice, because it is the whole issue in one statistic. This was not seasoned speculators losing a bet they understood. It was an entire generation's first experience of their own stock market, and that experience was: they told us it was safe, they sold us the leverage, and then the machine liquidated us at the low and the pension fund was nowhere. You do not come back from that at five times earnings. You may not come back from that at all.

You can rebuild a 40% drawdown in a year. You cannot rebuild a 25-year-old's belief that the game isn't rigged in less than about twenty-five years.

And here is the part that turns a national tragedy into a permanent investment problem, the fact that separates Korea from every other bombed-out cheap market you have ever been tempted by. Koreans are free to leave. Unlike China, where capital controls trap domestic savings inside the domestic market, a Korean investor with a brokerage app can buy the Nasdaq tomorrow morning, and for years, before the value-up dream, that is exactly what they preferred to do. So the marginal buyer this market just traumatised is not locked in. He is one tap away from the exit, and he has just been given the most vivid possible reason to take it. The cheapness assumes he comes back to provide the bid. The trauma is the reason he won't.

So who was supposed to catch him? This is where the second, quieter blunder comes in, and it is the one that should genuinely worry anyone who owns this market. Every functioning market has a shock absorber, a large, patient, rule-bound pool of money that buys when everyone else panics and sells into euphoria, and in Korea that is supposed to be the National Pension Service, a trillion-dollar whale that owns roughly 8% of both Samsung and SK Hynix. Its entire structural purpose is to be the counterweight. Buy low, sell high, cool the mania, cushion the crash. It is the fund that is supposed to be standing there when the 25-year-old gets liquidated.

Instead, earlier this year, the National Pension Service suspended its own rebalancing and quietly lifted its domestic-equity target from 14.9% to 20.8%, a manoeuvre whose transparent purpose was to justify not selling its Korean holdings into the boom, in service of the president's goal of a rising Kospi. The whale stopped acting like a whale. It stopped selling into the rally, which removed the ceiling and let the mania run hotter, and then, having spent its dry powder holding the market up on the way to the top, it had nothing left to cushion the fall. The one participant whose job was to lean against the crowd joined it.

They removed the market's shock absorber and installed an amplifier in its place, then expressed surprise when the ride got rough.

That is the real structural story, and the reusable lesson buried in a very Korean disaster. Every market runs on a balance between stabilisers, forces that lean against the move, and amplifiers, forces that feed it. Pension funds rebalancing, value investors buying dips, buybacks when stock is cheap: stabilisers. Leveraged ETFs, margin debt, dealer hedging, momentum funds: amplifiers. A healthy market is mostly stabiliser with a little amplifier for liquidity. Korea, through two policy choices made for short-term political convenience, sidelined its single biggest stabiliser and supercharged its most violent amplifier, at the exact top of a global tech mania. The wonder is not that it crashed 40%. It is that anyone is surprised.

The bulls are not entirely wrong, and honesty demands I say where they might win. The earnings are real; the memory supercycle is real; the leverage genuinely has been washed out, with the biggest single-stock fund's assets down 40% from the peak and forced selling largely spent. On Friday the index posted its single largest one-day gain in history, a 17.9% moonshot, when a legendary American hedge fund stepped in to buy the wreckage of an over-leveraged rival. If you are a trader with a three-week horizon and a strong stomach, there is a real bounce to be rented here, and the sharpest money is already renting it. A 366% earnings jump against a collapsing multiple is not nothing.

But a rented bounce and an investable market are different things, and the difference is exactly the trust I keep returning to. A market you trade is one you visit for a violent move and leave. A market you invest in is one you hand your savings to for a decade because you believe the rules are fair and someone will be there to buy when you need to sell. Korea just proved, to its own citizens, in the most expensive way imaginable, that it is currently the first kind and not the second. Foreigners have already voted, offloading roughly $80 billion this year and dragging the won down 4.8% despite a record trade surplus, which is the international money quietly agreeing that a market whose own government won't let its stabilisers work is a market you rent, not one you marry.

The valuation is real. The trap is that a cheap price is only a floor if there is a buyer standing on it, and Korea has spent a quarter methodically clearing the buyers off the floor and locking the door behind them. Five times earnings is not the bottom. It is the price of a market that has to win back an entire generation before the number means anything at all.

THE TRADE

Rent the Bounce, Don't Marry the Market

The structural insight is that the mechanical selloff and the trust damage are two different clocks running at different speeds. The leverage washout is nearly complete, which sets up a real, tradeable, technical bounce. The trust damage is structural and will suppress the market's valuation for years. The trade is to respect both, and never confuse the fast clock for the slow one.

The expression, for the tactical: a bounce is genuinely on offer, the leverage is flushed, foreigners set a single-day record buying on Friday, and 5x forward earnings against 300%-plus earnings growth is a real value cushion for a short-horizon trade in Samsung, SK Hynix, or the broad index. Size it as a trade, set a stop, and take the money; do not let a winning rent turn into an accidental marriage. For the long-term investor, the discipline is patience: the re-rating that would make Korea genuinely cheap rather than cheaply broken requires visible proof the government has restored its stabilisers, the pension fund rebalancing normally, the leveraged products actually curbed rather than lightly discouraged, and above all the young retail investor returning. Until that shows up in the flows, the cheapness is a value trap wearing an AI supercycle as a disguise. The cleaner long-term way to own the memory boom may simply be the chips themselves, or the US and Taiwan names, without importing Korea's governance risk.

Invalidation: Genuine, durable reform, NPS rebalancing restored, leveraged single-stock products actually banned or hard-capped, and evidence in the fund data that domestic retail is buying again, would convert this from a value trap into the real thing and justify owning the market, not just renting it. Watch the retail flow data and the NPS mandate, not the P/E.

Horizon: The tactical bounce is weeks; the trust repair is years. Do not blend the two horizons in one position.

THE PORTABLE TOOL

Carry this into every cheap market and every crashed stock you are ever tempted by. Before you buy the dip, ask one question the valuation screen cannot answer: after this fall, who is the marginal buyer, and do they still want to be here? A low price is only a floor if a willing buyer is standing on it. Sometimes the crash itself destroys the buyer, the retail base is wiped out and embittered, the natural stabiliser has been sidelined, the foreign money has a structural reason to stay away, and then cheapness is not a spring but a trapdoor, because there is nobody underneath it. The number on the screen tells you the price. It does not tell you whether anyone is left who wants to pay it. Find the marginal buyer before you assume the floor will hold.

For twenty-seven trading days Korea ran the fastest, wildest market on Earth, and for a brief window at the top it looked like the country had finally broken its ancient curse, the discount, the value trap, the sense that the game was quietly rigged against the little guy. A reformist president, a genuine boom, a generation of first-time believers who finally stopped putting their money in New York and put it to work at home.

Then the government it trusted handed it leverage at the peak, stood its one great shock absorber down, and watched 360,000 accounts get liquidated at the bottom, most of them belonging to people just starting out. The index will bounce; it already has, harder than any index in history. But the thing that actually broke this summer does not trade on the exchange and does not appear in the multiple, and it is the only thing that ever made a cheap market worth buying.

They finally convinced their own people the market wasn't rigged. Then, in a single quarter, they proved it was. Five times earnings is not a bargain. It's the receipt.

Happy Monday All

Welcome back. In the days since our SG Market Close Note last landed, the two central bank decisions we spent a fortnight flagging both came in on the hawkish side, an earnings season began, and the Singapore market absorbed all of it without drama. Today, the first session of August, the tape delivered a tidy summary of where the dust has settled. Seatrium led the index up 3.72% to S$2.23, the shipyards and industrials filled the top of the board, and everything sensitive to a rising currency and higher for longer rates sat at the bottom.

Start with the decision this note would not stop talking about. On 27 July the Monetary Authority of Singapore tightened, and it was a surprise: twelve of sixteen economists in the Reuters poll had expected a hold. MAS raised the rate of appreciation of the currency band very slightly, its second tightening after April, moving preemptively against the renewed oil-price surge even with domestic inflation still contained. For four editions this note argued that the MAS meeting was the live catalyst nobody had positioned for, and that a tightening would strengthen the Singapore dollar and reprice the rate-sensitive names. That is precisely what happened, and today's board is what it looks like a week later.

Two days after MAS, the Federal Reserve held at 3.50 to 3.75 percent, but three officials dissented in favour of a hike and the market now leans toward a move in September. So the backdrop your clients wake up to in August is a Singapore dollar being deliberately pushed higher and a Fed that is more likely to raise than cut. Both roads lead to the same place: this is not the rate relief world the REITs spent all of July praying for.

And yet the index has been resilient, because the other thing that happened over the break was earnings, and the ones that mattered were good. DBS crossed S$74 for the first time, closing at S$74.45, and Seatrium's result on 31 July answered the one question this note kept asking about it. The macro turned less friendly and the market held its ground anyway. That tells you the strength was never only about rates.

THE TAPE

The Board, Top to Bottom

Our read of the SG close, 3 Aug 2026. Post-earnings winners on top, the currency-sensitive names below the line.

Seatrium topped the board at 3.72%, and it earned it on the numbers. Its 31 July result showed first-half net profit up 158% to S$373m, and more to the point, the figure this note said would decide the stock came good: free cash flow swung to positive S$237m from negative S$5m a year earlier, gross margin widened to 8.6% from 7.4%, and adjusted profit excluding divestment gains still rose 54%. For a month we said the order book was known and in the price and that cash conversion was the real test. Seatrium passed that test. The one blemish, and the reason to stay honest, is that it declared no interim dividend and the headline profit leaned on S$172m of divestment gains. Good quarter, not a flawless one, and today the market chose to reward it.

Yangzijiang added 1.79% to S$3.99 and SATS rose 1.69%, so the shipyards and the industrial and logistics names remain the cleanest expression of the market's current preference: businesses with visible order books and hard assets, the opposite of a bond proxy. This has been the persistent theme through every twist of the past three weeks, and a tightening MAS only sharpens it.

DBS rose 0.58% to a fresh record S$74.45 on turnover of S$246.61m. The record is worth marking, given the stock started July below S$72, but keep the turnover in view: participation has settled in the mid-S$200m range, well below the S$430m-to-S$468m surges of mid-July. The banks are making new highs, but on the quiet, orderly volume of a market that has stopped chasing and started holding.

HongkongLand was the worst name on the index, down 3.08% to S$7.87. For a fortnight it was the most reliable name on the board, four straight sessions near the top on the strength of its daily buyback and the Jardine re-rating story. Today it gave back a chunk of that. A US-dollar-denominated property landlord is exactly the profile that struggles when the Singapore dollar is being pushed up and global rates stay high, and after the run it had, some profit-taking needs no deeper explanation. The buyback has not gone anywhere. The macro simply stopped flattering it.

Keppel fell 2.27%, Jardine Matheson 1.85% and Keppel DC REIT 1.35%, a familiar-looking tail of the rate-sensitive and the recently-crowded. Jardine reported on 30 July into very high expectations after its Morgan Stanley-driven run, and a pullback into a demanding setup is the market doing its usual work of separating the story from the price.

THE TAPE, PART TWO

The REIT Board

S-REIT movers, 3 Aug 2026. The US-dollar trusts led the losers, which is what a tighter SGD implies.

The REIT board tells the currency story in miniature. The losers were led by the US-dollar-denominated trusts: Manulife US REIT down 4.00% to S$0.048 and Acro HTrust down 3.92%, with Lendlease behind them. When MAS deliberately strengthens the Singapore dollar, a trust that earns in US dollars and reports in Singapore dollars faces a translation headwind on top of everything else, and the market marks it accordingly. This is not sub-cent noise for once. It is the policy decision showing up directly in the prices of the names most exposed to it.

The gainers were the European and niche names, Elite UK, the Stoneweg European trust and Sasseur, where the currency and rate story runs to a different clock. Manulife US REIT at S$0.048 has now joined the five-cent metronome club alongside our departed friend Landmark: at these prices every tick is a four-percent headline, and the drama on the screen wildly overstates the money involved. The signal on this board today is not the penny-trust theatrics. It is the clean line between what earns in strong currency and what earns in weak.

THE WRAP

The Catalysts Cleared, the Trend Held

Step back over the fortnight and the scorecard is unusually clean. MAS tightened when the consensus said hold. The Fed held but with hawkish dissents and a market now pricing a September hike. Seatrium delivered on cash. Jardine reported into lofty expectations. DBS made a new high. Every one of those was a chance for the rally that began in mid-July to break, and it did not. What it did instead was rotate, decisively and repeatedly, out of anything that needs cheap money and a soft currency and into anything with an order book and a hard asset.

That rotation is now the trend, and today confirmed it rather than started it. The risk from here is no longer a single dated event, it is the slow grind of a tighter Singapore dollar and a Fed leaning the wrong way for equities. The banks and the industrials have carried the market through the macro turn. The question for August is whether they can keep doing it into their own remaining earnings, with UOB on the 7th the next name up and the one this note has flagged for weeks as the most interesting setup on the board. A market that made new highs through two hawkish central banks has earned some respect. It has not earned complacency.

AROUND THE GROUNDS

News You'll Get Asked About

•  MAS tightened on 27 July, and it was the surprise this note had been positioning readers for. The band's rate of appreciation was raised very slightly, smaller than April's move, driven by the renewed oil surge and a widening positive output gap rather than by hot domestic inflation. The practical read for clients: the Singapore dollar has an official tailwind, imported inflation is the stated enemy, and HSBC among others now expects a further tightening possible in October. Anything a client owns in foreign currency has just been repriced by policy, exactly as flagged.

•  The Fed held on 29 July at 3.50 to 3.75 percent, but the composition was hawkish. Three officials dissented in favour of a 25 basis point hike, Chair Warsh stressed there is no soft inflation target, and the market has moved toward pricing a hike in September. For a REIT complex that spent July waiting for rate relief, this is close to the opposite of the hoped-for outcome, and it is why the currency-sensitive names sit where they do on today's board.

•  Seatrium's 31 July result was the genuine positive. First-half net profit rose 158% to S$373m and, crucially, free cash flow turned positive at S$237m with gross margin up to 8.6%. The net order book stands around S$13.3bn with over 95% series-build projects, a higher-quality mix than a year ago. The caveats are no interim dividend and a reliance on S$172m of divestment gains in the headline number. On balance the market read it as the cash-conversion proof point it needed, and marked the stock up today.

•  Jardine Matheson reported on 30 July into elevated expectations after the Morgan Stanley initiation drove it sharply higher in late July. Its pullback today, down 1.85%, is the market re-testing the gap between an attractive capital-returns story and a price that had already moved a long way to meet it. The thesis is intact. The easy money was made before the print.

THE READ

Where We'd Lean

A cleaner macro picture allows for cleaner positioning. Directionally, here is where the desk sits for August.

The through-line is now explicit: own the order book, be cautious on the bond proxy. A tightening MAS and a hawkish-hold Fed reward businesses with hard assets and visible revenue and penalise anything valued off a discount rate and a soft currency. Shipyards, industrials and the banks sit on the right side of that line. Long-duration REITs and US-dollar trusts sit on the wrong side. Today's board is not a one-day event, it is that line drawn in prices.

Seatrium: the result validated the thesis, so the debate moves on. Cash conversion is no longer the question, execution and the pace of new final investment decisions are. After a 3.72% day into a strong print, the easy re-rating is behind it. We would treat it as a quality cyclical to accumulate on weakness rather than chase on strength, and note the absent dividend as the one thing income-focused clients will ask about.

Banks into the back half of earnings: constructive but alert. DBS at a record on moderate volume is a market holding, not chasing. UOB reports on 7 August and remains the most interesting risk-reward on the board, having been de-risked hard in mid-July and then recovered quietly. A tighter policy stance is, at the margin, supportive for bank margins, which is the one part of the rate-sensitive universe that benefits from where policy is heading.

REITs and US-dollar trusts: the macro is now actively against them, so be selective and patient. A strengthening Singapore dollar is a direct translation drag on the US-dollar names, on top of a rate backdrop that refuses to ease. We would stay at the quality, Singapore-dollar-earning end and avoid reaching for the headline yields in the foreign-currency trusts, where the currency is quietly working against the payout.

HongkongLand and the Jardine complex: the story is intact, the tailwind is not. The buyback remains real and shareholder-friendly, but a US-dollar property landlord into a strengthening Singapore dollar has lost its macro help. We would separate the structural case, which we still like, from the near-term tape, which has turned less supportive.

DESK PREP

Ready for the Week

The August earnings run, now the main event:

•  Thursday 6 August: SGX. The exchange that spent July near the bottom of its own index finally reports. After a month as the running joke of this note, its own shareholders get some actual numbers to judge it on.

•  Friday 7 August: UOB. The one this note has flagged for weeks. De-risked in mid-July on no news, quietly recovered since, and reporting into a backdrop where a tighter policy stance is marginally helpful for margins. The most interesting setup on the board.

•  Then the tail: SIA's dividend goes ex on 11 August and ST Engineering reports on 19 August. The airline disagreement between a high share price and cautious analyst targets is still unresolved, and the dividend date is the near-term marker.

On the macro, the watch list is now a grind rather than an event: the strengthening Singapore dollar, the path of oil, and any US data that firms up the September Fed-hike pricing. None of these has a single dramatic date. All of them push in the same direction.

THREE THINGS TO SAY BEFORE 9AM

•  "MAS tightened, the Fed leaned hawkish, and the market held anyway." Both central banks came in on the firm side and the STI still made new highs. Respect that, but know why: it was earnings, not rates.

•  "Own the order book, be careful with the bond proxy." A stronger Singapore dollar and higher-for-longer rates reward shipyards and industrials and penalise long-duration and US-dollar trusts. Today's board is that sentence.

•  "UOB is Friday, and it is the one to watch." De-risked in mid-July, recovered since, reporting into a backdrop where tighter policy helps bank margins. The most interesting print of the week.

ONE MORE THING

The Note That Cried Central Bank

For four editions this note stood in the corner pointing at a meeting on a calendar that nobody else seemed to care about, insisting the MAS decision was the risk that mattered while the market chased brewers and shipyards and sub-cent REITs around the board. It began to feel a little repetitive. Then on 27 July, MAS did the thing twelve of sixteen economists said it would not, and the Singapore dollar firmed, and the rate-sensitive names rolled over, and the corner of the room this note had been shouting from turned out to be the right one.

We take no victory lap, partly because good calls and luck are hard to tell apart in real time, and partly because the more useful lesson is the boring one. The loudest, most crowded trades on the daily leaderboard were never the story. The quiet meeting nobody wanted to discuss was. That is usually how it goes. The market spends its attention on the stock that moved four percent today, and the thing that actually sets the direction is a central banker adjusting the slope of a currency band by an amount too small to see. Keep watching the quiet things. And keep an eye on Friday, because UOB is the next quiet thing with a date on it. See you then.

The Daily Morning Brew

Indicative and directional only. Nothing here is investment advice or a recommendation. Prices and levels are sourced from end-of-day SG tape (3 Aug 2026) and public market data. Macro, analyst and news items are paraphrased from public sources and current as at publication. Views are the desk's own and may change without notice.

Happy weekend all!

It's funny, but there is a specific comfort in deciding that the thing frightening you today is just a rerun of a thing that frightened you before and turned out fine. It lets you feel the fear and dismiss it in the same breath, which is the most pleasant emotional transaction available to an investor. And for two years the market has been running exactly this transaction on artificial intelligence, telling itself a bedtime story called It's Just 1999 Again.

The story is genuinely reassuring, and parts of it are even true. In 1999 a transformative technology arrived, nobody quite knew how to make money from it, and share prices went vertical anyway as investors bought the dream. Then the bubble burst, the recession that followed was one of the mildest in living memory, and the internet went on to eat the world regardless. The lesson everyone extracted: a bursting tech stock bubble is just the toll you pay for real innovation. The market gets ahead of itself, resets, and the technology keeps going. Nobody important gets hurt except people who bought pets dot com on margin.

And on the numbers, the 1999 comparison holds up better than the doom-mongers want you to believe. In 1999 companies with no revenue and no profit went public at deranged valuations. Today the companies at the centre of the AI trade make money hand over fist; a chipmaker's earnings genuinely have gone to the moon, which is the only reason its stock did. Microsoft and Apple are meaningfully cheaper now, on their earnings, than they were at the millennium. If those profits hold, the stocks are not even especially expensive. So relax. It's 1999, we know how this movie ends, and it ends with everyone eventually rich.

This is a lovely story and I would like you to stop telling it to yourself, because over the past three weeks the market has begun, very quietly, to notice it may have been rhyming with the wrong year the entire time.

A burst equity bubble robs you of money that only ever existed on a screen. A burst credit bubble robs you of money that other people were relying on being real. Those are not the same movie, and only one of them has a sequel called The Great Recession.

Here is the distinction that turns out to be the whole ballgame. An equity bubble is people paying too much for shares. When it pops, prices fall, paper wealth evaporates, some people who bought high feel terrible, and the economy mostly shrugs, because the money that vanished was never doing any structural work. It was just sitting there being a high number. Painful, survivable, forgotten in a cycle.

A credit bubble is a different animal in the same fur. It is people borrowing too much, and lending too much, against assets whose value everyone has agreed to overestimate. Houses in 2007. Data centres in 2026. When it pops, the problem is not that paper wealth disappears. The debt is real, the interest payments are real, the collateral turns out to be worth less than the loan, and the loss doesn't stay politely inside the companies that made the bet. It travels, through the banks and funds and insurers who financed it, into the plumbing of the whole system, because that is what debt does. A bubble in debt is a bubble in promises, and when promises break they break in a chain.

So the only question that actually matters about the AI boom is not whether the stocks are expensive. It is whether the thing being built has been paid for with wealth or with debt. And the answer, over the last few quarters, has quietly flipped from the comfortable one to the alarming one, which is why we are here on a Saturday talking about it.

Watch the pivot happen in the plumbing. For most of the last decade, the giants building AI were the most beautiful businesses ever constructed, because they were capital-light. Adding a customer cost them almost nothing; every new subscriber to a streaming service was nearly pure margin, the second million users costing about what the first million did. That is the magic that made these companies mint money, and it is precisely the magic AI does not have. Every query to an AI model costs real money, in electricity, in memory, in a chip that wears out and runs hot. There is no free second million. There is only a meter, running, on every question anyone asks.

The old tech giants got rich because serving one more customer was free. The AI giants get poorer with every customer they serve, and they are calling that progress because the alternative is admitting they are now in the business of renting furnaces.

That single change, from capital-light to capital-devouring, is the fault line, and the evidence it has cracked is now sitting on the cash-flow statements of the largest companies on Earth. One search giant just posted quarterly results far ahead of expectations and got its stock beaten down anyway, because it also burned six billion dollars of cash and promised to spend more. A social-media company watched its free cash flow, the actual money left after the bills, fall ninety per cent in a single quarter, the cash that used to reach shareholders now poured into data centres and chips. And one giant is guiding capital spending for next year toward two hundred and twenty billion dollars, a figure that would recently have been the entire market value of a very large company, now spent annually on buildings full of silicon that has to earn its keep or become the most expensive warehouse space in human history.

Across the whole group, the hyperscalers are now expected to borrow more than two hundred billion dollars this year, up from a hundred and twenty-five billion last year. Read that twice. The most cash-rich companies that have ever existed, whose balance sheets were a punchline about having too much money, are now borrowing at scale to fund a construction boom whose payoff is a projection. That is the moment an equity story becomes a credit story. That is the year turning from 1999 into 2008 while the stock market keeps its eyes politely on the share price.

And now the number. Because everything above is narrative, and this letter does not run on narrative, it runs on the one figure that makes the argument un-ignorable, and it lives in a market most people have never looked at and 2008 taught a small number of people never to stop looking at.

A credit default swap is insurance on a company's debt. You pay a premium; if the company fails to pay its bondholders, you get made whole. The price of that insurance is therefore the market's cold, money-on-the-table estimate of how likely a company is to go bust. It does not care about the share price, the earnings call, or the vibe. It is a number that people who insure debt for a living are willing to be wrong about with their own capital, which makes it the most honest signal in finance.

Oracle is spending enormous sums to become an AI landlord, borrowing tens of billions to build data centres it will rent to companies burning cash. Its five-year credit default swap has climbed to around 215 basis points, up from 144 at the start of the year, an eighteen-year high. Sit with eighteen years. That means it now costs more to insure Oracle's debt against default than it did during the actual 2008 financial crisis. One bank's credit strategist runs the maths on that spread and gets an implied default probability north of sixteen per cent, better than one-in-six odds that a blue-chip technology company does not pay its debts.

The stock market is arguing about whether it's 1999. The credit market has already stopped arguing, priced 2008, and quietly bought the insurance.

Zoom out from Oracle and the tell only sharpens. Total net-notional credit-default-swap protection on the big technology names has hit a record twelve and a half billion dollars, up five hundred per cent in a year, with Oracle alone over half of it. A data-centre operator called CoreWeave, junk-rated and central to the buildout, trades at a swap spread of 855 basis points, the market's genteel way of putting its odds of default at roughly one in two. A coin flip. Even the pristine, fortress-balance-sheet names have seen their insurance costs creep up, the market whispering that it is no longer sure any of them can chew what they bit off.

Regular readers have watched us circle this exact structure for two months, in different costumes each time. We wrote in July about the shovel-sellers, the compute landlords extending ten-year credit to AI labs that lose two dollars for every one they earn, and asked the only question that mattered, which was whether the miners could pay. We wrote about a perpetual money machine that ran on belief until the belief and the borrowing came apart. This is the same creature, grown to the size of the entire market, and the credit default swap is simply the first instrument sober enough to say so out loud.

The bull case has not vanished, and honesty requires stating it at full strength. AI revenues so far have been spectacular and genuinely real, not a 1999 mirage, and as long as that revenue keeps compounding faster than the interest bill, the credit investors stay calm and the whole edifice is not a bubble but a boom that happens to be debt-financed, which is how most great infrastructure gets built. Railroads were debt-financed. The interstate was debt-financed. Debt is not the disease. Debt against an asset that turns out nobody wanted at that price is the disease, and the jury on which of those we are watching is still, genuinely, out.

But the burden of proof has flipped, and that is the actual news. For two years the AI trade was innocent until proven guilty, carried by the reassuring hum of 1999. As of this month it is the reverse. The credit market has moved first, as it did in 2007 while equities kept partying, and put real money on the darker reading. When the people who insure debt and the people who buy stock disagree this loudly about the same companies, history is fairly consistent about which group was paying attention.

THE TRADE

Trust the Insurers, Not the Optimists

The structural edge is that two markets are pricing the same AI buildout at once, and they violently disagree. Equity investors, anchored to 1999, are pricing continuation. Credit investors, anchored to 2008, are pricing risk. When these two diverge this far, the credit market is the one wagering its own capital on being right, and it is historically the earlier and more reliable signal.

The expression, for the sophisticated: favour quality and cash generation over the debt-funded buildout. Within the AI complex, prefer the genuinely self-funding giants (the ones generating tens of billions in real free cash flow, whose CDS has barely moved) over the pure-play landlords borrowing tens of billions against a single concentrated tenant, whose swaps are screaming. The cleanest expression of the caution is in credit, not equity: the debt of a single-tenant AI landlord is priced for a boom that its own insurance market no longer believes in. For everyone else, the discipline is simpler and just as valuable: before you buy any company riding the AI wave, look up its five-year CDS. If the price of insuring its debt is near a multi-year high while its stock is near a high too, those two facts cannot both be right for long, and the insurers are the ones with skin in the game.

Invalidation: Sustained, accelerating AI revenue that comfortably outpaces the rising interest bill turns this from a credit bubble into an ordinary debt-financed boom, and the CDS spreads compress. Watch hyperscaler free cash flow and Oracle's CDS together: falling swaps plus rising cash flow means the bears were early and wrong.

Horizon: Six to eighteen months, the window in which the buildout's interest payments start colliding with its actual, delivered revenue. The catalyst is any hyperscaler guiding capex down or any AI-lab funding round that struggles, either would confirm the credit market read the movie correctly.

THE PORTABLE TOOL

Carry this one into every boom you ever meet, not just this one. When you want to know whether a hot story is a bubble in stocks or a bubble in debt, do not read the equity analysts, who are paid to find continuation. Find the price of insuring the company's debt, its credit default swap, and compare it to its own history. Stocks tell you what the optimists hope. Credit tells you what the people who lose real money if they are wrong actually believe. When the share price says party and the debt insurance says panic, believe the insurance, because it is the only one of the two that has to pay out when the story turns out to be fiction. Two markets, one company, and the quieter one is almost always the one telling the truth.

So the market sits in the cinema, confident it has seen this film before: a comedy from 1999 about overexcited investors who lose their paper profits, learn a gentle lesson, and are quietly rich again by the sequel. Everyone knows the ending. Everyone is relaxed. Down in the front row, in a market with no popcorn and no press, the people who insure debt for a living are watching a completely different film, a slow, grim thing from 2008 about a mountain of borrowing built on an asset everyone agreed to overvalue, and they are not relaxed at all. They have already left to buy insurance in the lobby.

When the audience and the insurers are watching two different movies about the same company, the insurers have read the ending. They always buy their tickets first.

The Daily Morning Brew is an independent financial newsletter. Nothing here is investment advice. All market levels cited were verifiable at the time of writing.

Good morning all!

At some point on Thursday, an official at the New York Fed picked up a telephone, on instructions from the US Treasury, and asked several banks a question. The question was: what price would you buy and sell dollar-yen at right now? The banks answered. Nobody traded anything.

This is called a rate check, and it is the closest thing modern central banking has to a haunting.

The yen then rose as much as 3.3% against the dollar, its largest intraday move in more than two years, closing at 159.53. Japan's Nikkei reported that Tokyo had also intervened directly, and that the rate check had happened. The Finance Ministry could not be reached. The US Treasury declined to respond. Nobody confirms these things and nobody is expected to.

Which is the point at which I should be honest that reporting ends and inference begins. No official has said Japan intervened on Thursday. A newspaper cited unnamed traders, the move was the right size and the right shape, and everybody has drawn the obvious conclusion. Whether the rate check was coordinated policy, a solo Treasury decision or a well timed piece of theatre is not something anyone outside two buildings actually knows, and that includes several of the people confidently quoted saying it was one or the other.

Hold on to the phone call, because we are going to compare it to $73 billion in a moment and the comparison does not flatter the $73 billion.

First, the setup. The yen has been having the sort of decade that gets written about in books. It breached 161.95 in June, its weakest against the dollar since 1986, and the previous visit to this level was on the way up rather than down, midway through the long rally that followed an American-engineered currency accord. It extended to 162.40. Japan imports effectively all of its energy and more than half its food, priced in dollars, into a war-inflated oil market, which means the currency is now an inflation instrument rather than a trade instrument. Prime Minister Takaichi's approval ratings have fallen sharply and she is preparing a two-year cut to the sales tax on food, which will lower measured inflation from April 2027 and probably raise the underlying kind.

Tokyo has been fighting this. From 28 April to 27 May, the Ministry of Finance spent ¥11.73 trillion, about $73.2 billion, buying its own currency. That is the largest monthly intervention in Japanese history, comfortably. It was funded, as these things are, by selling foreign securities, including US Treasuries, out of a $1.17 trillion pile of foreign currency assets.

Now the arithmetic, which is where this stops being a currency story.

The global foreign exchange market turns over about $9.5 trillion a day. Divide by 1,440 minutes and you get roughly $6.6 billion a minute. Japan's largest monthly intervention on record therefore amounts to about eleven minutes of global turnover.

Eleven minutes. Spent over four weeks. Against a tide that refills itself every six hundred and sixty-six seconds. The exact day by day breakdown of what the ministry actually did lands in early August, long after anybody could have traded on it.

The Ministry of Finance knows this, which is why the honest version of the official position is not that intervention reverses trends. It is that intervention buys time and contains the speed of the move, and by that standard the April operation worked, keeping the rate inside a five yen band during the worst of the Strait of Hormuz panic. Containing volatility is a real objective. It is just a different objective from the one the headlines describe, and the gap between those two objectives is where $73 billion went.

Here is the part that should interest anyone who does not care about the yen at all.

To buy yen, Japan sells dollars. To have dollars to sell, Japan sells US Treasuries. Selling US Treasuries pushes US yields up. Higher US yields widen the interest rate differential between America and Japan. A wider differential is the single largest reason to be short the yen. Which produces yen weakness. Which requires intervention. Which requires selling Treasuries.

The world's largest foreign holder of American government debt has been manoeuvred into a position where defending its own currency requires it to raise America's cost of capital, which then attacks its currency. It is a machine that manufactures the disease it was built to treat.

The timing is exquisite. On Wednesday the Fed held at 3.50% to 3.75% by 9-3, three presidents dissenting for a hike, and Chairman Warsh observed that rising Treasury yields had done some of the tightening for him. The thirty-year promptly rose to its highest level since 2007. So the US long end is being repriced by a central bank that has withdrawn its forward guidance at exactly the moment Tokyo needs to sell into it.

The part that makes a Japanese equity holder wince is Toyota. The company estimates that every ¥1 of yen depreciation adds ¥50 billion to operating profit. Thursday's move was about 5.3 yen of appreciation from the intraday low. On Toyota's own arithmetic, the Japanese state spent public money to remove roughly ¥263 billion, about $1.65 billion, of annualised operating profit from its single largest manufacturer. Japanese carmakers collectively stand to gain $5.8 billion this year from the weak currency, and the government has been spending twelve times that figure to take it away from them.

Which, fine. Consumers vote and Toyota does not. But notice that there is no version of Japanese economic policy in which everyone in Japan wants the same exchange rate, and any policy that costs $73 billion a month while making your largest exporter poorer is not really a currency policy. It is a transfer, and the recipients are households buying imported food and the payers are shareholders, and nobody has said so out loud.

Now zoom out, because the reason a family office in Singapore should read this has nothing to do with Japan.

The yen is the world's funding currency. Borrow at 1%, sell the yen, buy anything with a yield, collect the difference. That trade sits underneath a very large amount of global leverage, and the funding leg is supposed to be boring. A 3.3% adverse move in an afternoon on your funding leg is not boring. It is a margin call transmitted simultaneously to every levered position on earth, which is why August 2024 remains a case study, and why gold, emerging market credit and US technology exposure all quietly share a risk factor most holders have never written down.

The regional picture is already tightening. India's central bank sold about $7 billion defending the rupee last Friday. Bank Indonesia is issuing reassurances about the rupiah that markets are not buying. Argentina's central bank has just ended a 135-day dollar-buying streak as the peso hit a new low. The yuan has climbed to its strongest since early 2023 as the post-Fed dollar slide did the work. The yen is the anchor of that complex. If it goes, the entire Asian FX defensive line reprices at once, and every regional treasurer discovers that their hedging assumptions were correlated after all.

Which brings us to lunchtime today in Tokyo.

All 52 economists surveyed expect the Bank of Japan to hold at 1%, the level it reached on 16 June and the highest since 1995. Traders disagree about what comes next with unusual conviction: overnight index swaps imply roughly a 74% chance of another hike by October, up from about 50% the day after the June meeting. Roughly half of economists say December. About 40% say October. Governor Ueda chairs his first meeting since being hospitalised for a liver cyst infection, and about half of surveyed economists expect at least one dissent. Thursday's two-year JGB auction scraped through on weak demand, which is the bond market's way of saying it would like to see the hike rather than hear about it.

The US Treasury, in its semiannual report last week, cited "substantial yen undervaluation" and pressed the BOJ to keep normalising. Tokyo has not answered that in public.

The positioning follows from the plumbing rather than from a view. Intervention without a rate move is a fade, because it changes the price and not the differential. Intervention followed by a hike is a trend change, because it changes both. That is the whole decision tree, and it resolves at 3:30 p.m. Tokyo time. Short-dated yen calls into the announcement are the honest expression of it, sized as insurance against the meeting being live rather than as a directional bet, because Ueda has spent three years declining to surprise anybody and there is no reason to assume today is the day.

The named variable is 165. Bloomberg strategists have flagged 164 to 165 as the next reference zone from the 1986 price data, and MUFG has said the yen may test 165 if the BOJ stands pat. If today's hold pushes the currency back through 162 and toward 165, then the record intervention has publicly failed, the October hike stops being a 74% probability and becomes an obligation, and Japan's next defence has to be conducted with policy rather than reserves. Below 158 and holding, the intervention worked, the carry trade re-arms, and everything in the paragraph above can be ignored until the next time.

Which returns us to the telephone.

Tokyo has now spent $73.2 billion, roughly 6.3% of its entire foreign currency reserve, over a single month, to move the yen about five figures on a screen for a few weeks. On Thursday somebody in New York asked four or five banks what their price was, hung up, and helped produce the largest intraday yen move in two years at a cost of one phone call.

The most powerful instrument in international monetary policy is a question asked by a government with no intention of trading.

Notes

1 Yen rose as much as 3.3% intraday on 30 July 2026, largest since December 2023, closing 159.53. Nikkei reported Japanese intervention and a New York Fed rate check conducted for the US Treasury. Yen breached 161.95 in June, weakest since 1986, extending to 162.40. Japan's Ministry of Finance intervened to the tune of ¥11.73trn ($73.2bn) between 28 April and 27 May, funded partly from $1.17trn of foreign currency assets.

2 Turnover arithmetic: $9.5trn daily global FX volume equals roughly $6.6bn per minute, so $73.2bn is about 11.1 minutes. Toyota estimates ¥50bn of operating profit per ¥1 of depreciation; Thursday's move of roughly 5.3 yen implies about ¥263bn. Japanese carmakers' estimated 2026 currency windfall is $5.8bn. BOJ holds at 1% since 16 June; OIS implied roughly 74% odds of a hike by October as of Wednesday's Tokyo close.

The Daily Morning Brew is an independent financial newsletter. Nothing here is investment advice. All market levels cited were verifiable at the time of writing.