Start with the year, not the quarter.

Broadcom opened 2026 at $346.10. It closed Wednesday at $367.24, before it reported. Eight months, 6%.

The semiconductor sector is up about 60% over the same stretch.

In those same eight months, Broadcom’s AI chip revenue went from under $6 billion a quarter to $16.7 billion a quarter.

Wednesday night was the fourth time this year the company posted enormous numbers and the stock went down anyway.

The quarter

The fiscal third quarter ended 2 August. Broadcom reported it on 2 September.

Revenue was $29.591 billion, up 86% from $15.952 billion a year ago. Adjusted earnings were $3.32 a share against the $3.24 expected. Free cash flow was $13.665 billion, which is 46 cents out of every revenue dollar, up 95% on the year.

AI semiconductor revenue was $16.7 billion. Up 221% in a year. Up 54% in three months.

Look at that last one again. AI is now 56% of the entire company, and it grew by more than half in a single quarter.

Then Hock Tan gave the forecast.

“In 2027, we have secured the supply to again double AI revenue to approximately $115 billion.”

“In 2028, we expect the trajectory of growth to continue. We have line of sight for fiscal 2028 AI semiconductor revenue growth to again double to $230 billion.”

About $58 billion this year. $115 billion next year. $230 billion the year after.

The stock fell 5% after hours.

What it fell on

Broadcom guided the current quarter to $34.8 billion. Analysts had $35.03 billion, according to LSEG.

The gap is $230 million. On a $35 billion quarter that is 0.66%.

A company forecast quadrupling over two years, came in two thirds of one percent light on one quarter, and lost value.

Now, the two forecasts are not the same kind of promise, and almost every write-up this week has treated them as if they were.

For 2027, Tan said Broadcom has secured the supply. That means manufacturing capacity the company has already locked down and paid for. It is not signed orders. It is still a firmer commitment than most chief executives make about a year that has not started.

For 2028, he said line of sight. That phrase lives somewhere between a forecast and a hope. It carries none of the accounting weight that the word backlog carries. It means management thinks it can see the demand, based on customer build plans that are not contracts yet.

One number has factories behind it. The other has visibility behind it. If you are deciding how much of this to believe, that is the whole question.

The margin, and where I saw this two weeks ago

Regular readers will remember Nvidia’s gross margin slipping from 75% to 74%, and the reason the company gave: memory got expensive.

Broadcom’s gross margin came in at 75%, down 210 basis points from the previous quarter.

Amie Thuener gave the reason. This was her first earnings call as Broadcom’s chief financial officer, after taking over from Kirsten Spears in June.

“Gross margin was 75% of revenue in the quarter, down 210 basis points sequentially as AI semiconductor revenue was a greater proportion of our total revenue mix.”

She also pointed at the rising memory content in the custom chips as those become a bigger share of what Broadcom sells.

So: two companies with genuinely different businesses, one selling standard GPUs and one designing bespoke chips to order, reporting a fortnight apart, both with margin compression, both naming memory.

When two competitors independently point at the same input, it is not a company problem. It is a supply problem. The money in this build-out is moving one step up the chain, toward whoever makes the memory that sits next to the processor.

Broadcom still guided next quarter’s operating margin to about 66%, flat on the year. So the pressure is being absorbed. But a margin that drops 210 basis points a quarter four quarters running stops being a rounding error.

Read more

Gold is up roughly 25% over the past year. Gold is up 3.31% this calendar year, through late August.

Both are true. The difference is entirely where you start the clock. Gold ran to a record of about $5,590 an ounce on 28 January 2026, then fell hard, and it now trades around $4,350. Anyone who bought in January is down 22%. Anyone who bought last September is up a quarter of their money.

That is the first thing to understand about buying gold, and it is the thing the ads leave out. It is not a one-way asset. It just had one of its worst drawdowns in years, and it happened seven months ago.

The second thing is that the route you pick costs you more than most people expect. Not in a dramatic way. In a slow, 4% here and 8% there way that quietly eats years of return.

There are three routes. Here is what each one actually costs and where each one belongs.

Route one: physical metal

You buy coins or bars and you own the actual thing.

What you pay above the gold price. Dealers quote a premium over spot, and it covers refining, minting, their margin, insurance and shipping. In normal conditions, one-ounce bars run 1% to 3% over spot. Coins like the American Gold Eagle or the Krugerrand run 2% to 8%, higher because they are minted, recognised and easier to resell anywhere in the world.

Anything above 3% for a bar or 8% for a coin from an established dealer is too much.

Small coins are where beginners quietly overpay. A tenth-ounce coin carries a far higher percentage premium than a one-ounce coin, because minting a small coin costs nearly as much as minting a large one. You are paying for the stamping, not the gold.

Where to buy. The established online dealers are JM Bullion, APMEX and BullionVault, all accredited and shipping insured. Costco sells one-ounce bars to members at roughly 1% to 2% over spot, which is genuinely competitive, and is worth knowing if you are already a member.

What to stay away from. Television advertisements, subscription programmes that ship you a coin a month, anything sold on scarcity or a countdown, and anything described as proof, commemorative, graded or collectible. Those carry markups that have nothing to do with the metal inside, and some of them are gold plated rather than gold. If you are buying gold as an investment, you want the boring bullion product at the lowest premium, and nothing else.

Sales tax, which almost nobody checks first. Forty-five states now exempt investment bullion, either fully or above a dollar threshold. Five states plus Washington DC still apply their general sales tax to it with no exemption: Hawaii, Maine, New Mexico, Vermont and Washington. And the rules move. Maryland added a 6% tax in 2025, Washington’s 6.5% started on 1 January 2026, and Virginia’s exemption lapsed on 1 July 2026.

A 6% sales tax on a gold purchase is a 6% loss on day one. Check your state before you order, not after.

Storage. A home safe is free and uninsured. Professional vaulting through Brinks or Loomis runs roughly 0.1% to 0.5% a year and includes insurance. On a small position the vault fee is not worth it, which means a small physical position usually sits in a house, which is a security question rather than a financial one.

Route two: a physically backed ETF

The fund holds real bars in a vault and you hold shares. You get the gold price without the safe, the shipping, the dealer spread or the storage question, and you can sell it in seconds.

The only meaningful difference between these funds is the fee, and the gap is larger than people assume.

They all do the same job. GLD is the biggest and most liquid, which matters if you trade options on it, and matters not at all if you are buying and holding. GLD charges four times what GLDM charges for the same exposure. On $10,000 that is $40 a year against $10. On $100,000 held for a decade it is a few thousand dollars, paid for liquidity you were never going to use.

For most people buying gold to hold, GLDM or IAU is the sensible default and GLD is a habit.

SGOL is worth a look if you want the bars vaulted in Switzerland rather than London, which is a preference rather than an argument.

Route three: mining shares and mining funds

Instead of the metal, you buy the companies that dig it up.

GDX is the standard vehicle, holding around $30.5bn at 0.51% a year.

Over the past year GDX returned 50.92% against GLD’s 24.74%. Year to date it is up 14.69% while the metal is close to flat. Over the past decade GDX has compounded at roughly 14.6% a year against 12.7% for GLD.

The miners have beaten the metal on every one of those windows.

Here is why, because it cuts both ways and the second half is the part that gets skipped.

A mining company’s costs are largely fixed. Wages, diesel, equipment, permits. Those barely move when the gold price does. So if it costs a company $1,800 to produce an ounce and gold rises from $2,300 to $2,800, the metal went up 22% while the company’s profit per ounce went from $500 to $1,000. It doubled.

That is operational leverage. It works exactly as hard in reverse. When gold fell 22% from January’s record, the miners fell harder.

You are also buying companies, with everything that comes with companies. Management, debt, floods, strikes, permitting fights, and governments that occasionally rewrite mining law. A mining stock can fall on a day gold rises because something went wrong at a single mine in Peru. If your reason for buying gold is that you want an asset that does not depend on anybody’s competence, miners are not that asset.

The middle option. Royalty and streaming companies like Wheaton Precious Metals and Franco-Nevada finance mines in exchange for buying future output at a locked-in price. Wheaton’s cash cost sat near $400 an ounce in March 2026 while gold traded above $3,200. They get the leverage without operating the mines. Wheaton returned roughly 48% over the past twelve months and overtook Franco-Nevada as the largest streamer by market value in early 2026. Franco-Nevada has raised its dividend for more than 17 consecutive years.

The tax rule that decides where each one goes

This is where the routes stop being interchangeable.

The IRS classifies gold as a collectible, in the same bracket as art and antiques. Long-term gains on collectibles are taxed at a maximum federal rate of 28%, against a maximum of 20% for stocks.

That classification follows the metal, not the packaging. A gold ETF that holds real bars is a collectible for tax purposes. GLD, GLDM, IAU and SGOL all fall in it. It looks like a fund and it trades like a fund, and when you sell it the bill arrives as though you sold a painting.

Mining shares and mining funds are ordinary equities. Maximum 20%.

Same view on gold. Eight percentage points of difference, decided by what the fund holds rather than what it tracks.

One correction to something repeated everywhere online: the 28% is a ceiling, not a flat rate. The gain stacks on your ordinary income and is taxed at your marginal rate, capped at 28%. Somebody in a lower bracket may pay well under it. Higher earners feel it in full, and they are the ones likely to be holding a large position.

And all of this concerns gains on positions held more than a year. Sell inside twelve months and it is taxed as ordinary income at up to 37%, gold or miners, and the whole comparison stops mattering.

So the placement rule is simple. Physical gold and physically backed ETFs belong in a retirement account, where the collectibles rate never applies. Mining funds are the more sensible taxable holding.

If you want physical gold inside an IRA

You cannot just put coins in a drawer and call it retirement money. Holding gold in an IRA requires a self-directed custodian and storage at an IRS-approved depository. Storing it at home violates IRC Section 408(m) and is treated as an immediate distribution, with the tax bill that implies.

Eligible products are bars at 99.5% purity from refiners like PAMP Suisse or Valcambi, plus the American Gold Eagle, which gets a specific exemption at 91.67%.

The costs: setup of $50 to $200, custodian fees of $75 to $900 a year, storage of $100 to $300 a year. Call it $250 to $400 annually all in.

Run that as a percentage before you commit. Three hundred dollars a year on a $10,000 position is 3% a year, which will overwhelm anything gold does for you. On a $100,000 position it is 0.3%, which is reasonable. A gold IRA is a large-position product, and it is sold hard to people with small positions.

The 2026 IRA contribution limit is $7,000, or $8,000 if you are 50 or over.

How much

The World Gold Council’s own research, which is not a neutral source but is the most thorough on this, puts the strategic allocation at 5%, with a sensible range of 2% to 10%. Their finding is that a 5% gold position contributes about 28% of a balanced portfolio’s total diversification benefit, more than any other single asset.

Higher than 10% and you are no longer diversifying. You are taking a position on gold.

Selling, which people plan for last

Physical gold has a spread on the way out as well as the way in. The dealer buys back below spot. That round trip is the real cost of owning coins, and it is the reason a physical position needs to be held for years rather than traded.

On reporting: dealers file a Form 1099-B when you sell gold bars or rounds of 99.5% purity in quantities of one kilo (32.15 troy ounces) or more, with sales to the same customer aggregated over 24 hours. Coin sales generally do not trigger it, because no active futures contract lists coins as deliverable.

Read that carefully. No 1099-B does not mean no tax owed. It means nobody filed a form on your behalf. The gain is still reportable, and it is still yours to report.

Gold is one of the few assets where the wrapper matters as much as the view. You can be completely right about the metal, hold it five years, watch it do exactly what you expected, and end up several percentage points behind somebody with the identical opinion who filled in the paperwork differently.

Disclaimer: Nothing here is financial advice, and it is certainly not tax advice. I am not a licensed financial advisor. Tax treatment depends on your own circumstances, so speak to an accountant before acting on any of it. I may hold positions in securities mentioned.

Leave a comment

The supply agreement runs through December 2047.

Twenty-one years, from a company whose entire business is built on things that did not exist a decade ago. Google will pay the full cost of adding 1,600 megawatts of new renewable capacity and 480 megawatts of battery storage to make it work, and it has accepted minimum monthly charges and termination fees along the way.

Nobody signs a twenty-one year fixed obligation because they are optimistic. They sign it because they are worried the thing might not be available later.

Everyone is watching the wrong shortage

The conversation about artificial intelligence has been about chips for three years. Who makes them, who gets them, how many, at what margin.

Meanwhile the four largest cloud companies plan to spend roughly $725 billion on capital expenditure in 2026, up 77% from $410 billion last year. Amazon alone raised its guidance to about $220 billion, from $125 billion in 2025, with nearly all of it going to data centers, custom chips and power.

You cannot spend that without electricity, and there is a fixed amount of it.

The International Energy Agency expects global data centre electricity consumption to roughly double, from about 485 terawatt hours in 2025 to around 950 by 2030. China and the United States account for nearly 80% of that growth, with America alone adding around 240 terawatt hours. Kevin Warsh, in his Jackson Hole speech last week, attributed more than half of this year’s total capital spending growth to the AI buildout.

Chips can be manufactured faster. Capacity gets added, suppliers get qualified, and the constraint eases within a couple of years.

A power station takes a decade. A transmission line takes longer, because it has to cross land somebody owns.

There is a number out of Texas this month that makes the scale of this unmistakable, and after going through the valuations on every listed way to buy into it, only one of them looks genuinely mispriced.

Read more

Central bankers do not say what they mean. They say something adjacent to what they mean, in a register designed to be defensible later, and the job of anyone reading them is to work out which ordinary-sounding sentence is doing the heavy lifting.

Kevin Warsh gave a speech at Jackson Hole last Friday called “In Our Time.” It was his hundredth day as chairman. The coverage treated it as vague, and it was not vague at all. It was one of the most direct speeches a Fed chair has given in years, and the directness is hidden in phrasing that reads as mild until you know what the mild version is standing in for.

The Fed meets on 15 and 16 September.

So here they are again, slowly.

The responsibility for 65 months of sustained, elevated inflation sits squarely with the central bank.

Start here, because a sitting Fed chairman does not normally say this out loud.

Sixty-five months back from August 2026 lands in March 2021. He is putting a date on the beginning of the failure, calling it five and a half years long, and assigning the blame to his own institution rather than to supply chains, or a war, or a pandemic, or any of the other explanations his predecessors reached for.

A chairman who believes the Fed caused this is a chairman who believes the Fed has to fix it. That is the frame for everything that follows.

Price stability is not self-executing, nor is inflation necessarily mean-reverting.

Plain English: it will not come down on its own.

Mean reversion is the idea that things eventually drift back to normal. He is rejecting it. Which means waiting is not a strategy, and every argument that begins “if we are patient” has just been removed from the table.

Progress over the past two years has been modest.

Two years. Modest. From a man being deliberate about his words.

On balance, I would be hard pressed to describe broad financial conditions as restrictive.

This is the one. If you read nothing else, read this.

Financial conditions being restrictive means that interest rates are actually squeezing the economy, which is the entire point of having them high. Warsh is saying he cannot honestly describe current conditions that way.

Rates are 3.50% to 3.75%. He is telling you they are not doing the job.

He then lists his evidence, and it reads like a prosecutor.

Credit spreads on corporate bonds and leveraged loans are near the low ends of their historical ranges.

Banks tell us that standards for commercial and industrial loans are on the easier end of their historical range.

Overall equity market volatility is low.

Real consumer spending has been healthy despite the shocks, increasing more than 2 percent over the past four quarters.

None of that describes an economy under pressure.

Read more

This week they trade around $935.

That is not a typo. The stock is up roughly 975% in twenty months. Ten thousand dollars put in at the start of last year would be worth about $107,000 today. Sold at the June peak of $1,253, closer to $144,000.

Micron is now worth $1.06 trillion. A company that makes memory chips, a business the market spent forty years dismissing as a boom-and-bust commodity, has joined the trillion dollar club.

It happened in public, over twenty months, while most of us watched and did nothing.

Which is why everyone is hunting for the next one

That is the itch. Not greed exactly. Something closer to having stood next to something obvious and not moved.

So the internet is looking for whatever does that next, and one name keeps surfacing.

IREN

It reported full-year results on Wednesday, which gave me an actual set of numbers instead of a thumbnail. So I read them, then read the accounts underneath, and then the two quarters before that. The last part is where it got interesting.

The shares closed the regular session at $40.53, up 2.4%. The next morning they were around $37.69 in pre-market, down 7%.

What IREN actually is

Two Australian brothers, Daniel and Will Roberts, both formerly of Macquarie, founded it in Sydney in 2018 as Iris Energy. The idea was bitcoin mining powered by renewable energy, starting with hydro in British Columbia. It listed on Nasdaq in November 2021.

Mining needs two things above all else: buildings, and enormous amounts of electricity. Which is exactly what an AI data center needs.

So in 2024 they renamed the company IREN and changed jobs. Instead of using the power to mine bitcoin, they started renting it to companies running AI.

This week we found out how far along that switch is.

AI cloud revenue for the year was $128.8 million, up from $16.4 million. Roughly eight times bigger in twelve months.

And in the final quarter, AI revenue reached $70.5 million and passed bitcoin mining revenue for the first time.

That crossover is the story. About three months ago, IREN stopped being a bitcoin company with an AI side project and became an AI company with a bitcoin remnant.

The $703 million loss is not what it looks like

The headline was ugly. A net loss of $702.6 million, against a profit of $86.9 million the year before.

But $638.8 million of that is one line: writing off old bitcoin mining hardware.

IREN deliberately switched off and scrapped mining machines so it could reuse the electrical infrastructure and buildings for AI. Accounting rules make you write that equipment down to nothing when you do.

No cash left the business. It is the accounting version of admitting the thing you bought three years ago is not what the building is for any more.

Strip it out and revenue for the year was $707 million, up from $501 million.
The transformation is real and the contracts are real. Then I read the quarter before this one, and stopped being sure.

Below: what the previous quarter reveals, why the funding gap everyone quotes is wrong, the four risks that matter, and a straight answer on whether this can do what Micron did.

Read more

The shares had ended the regular session at $209, having drifted lower in the days beforehand as investors weighed nearly doubled revenue against the size of the commitments the company has taken on. Once the numbers came out, the stock rose to around $219 in after-hours trading, a gain of roughly 5%.

Some of that was relief that nothing broke.

Here is what actually happened, what beat, what worried people, and what the internet made of it.

Revenue was up 106% from a year ago. It doubled. At this size, that is worth sitting with for a second.

Nvidia now makes more money in three months than most large companies make in several years.

One note on the labelling, because it confuses people. Nvidia calls this the second quarter of fiscal 2027, even though we are in 2026. Its financial year runs ahead of the calendar. The quarter actually ended on 26 July.

Where the money comes from

This is the number that tells you what Nvidia has become.

Data centers brought in $89 billion of that $96.2 billion. Up 117% from last year.

That is 92% of the entire company.

Nvidia used to be a gaming company. It sold graphics cards to people who wanted better looking video games. That business still exists and it is now a rounding error. Nvidia sells equipment to people building data centres, with a small gaming hobby attached.

The number that went the wrong way

Everything above beat expectations. One thing did not, and it is what analysts spent the evening on.

Gross margin. That is the share of each sale left over after paying to make the product.

This quarter it was 75%. Next quarter, Nvidia says it will be 74%.

One percentage point sounds like nothing. On a $108 billion quarter it is more than a billion dollars of profit.

Why is it falling? Because memory chips have become expensive. Nvidia buys memory to put in its systems, and memory prices have gone through the roof this year. Micron is up 255% for exactly that reason.

So Nvidia is paying more for its ingredients, and right now it is absorbing some of that rather than passing all of it on.

The margin is the smallest story in this release. Below: the sentence Jensen Huang said that actually moved the stock, the $500 billion announcement that has nothing to do with chips and everything to do with how this boom is being funded, the three camps that formed online within the hour, and the one line almost nobody is talking about.

Read more

Which company? There are nine of them and they all sound plausible.

And is now even the right moment? It’s already run up 40%. If you buy today and it falls next week you’ll feel like an idiot for months.

So you close the app and do nothing. Again.

That hesitation costs more than a bad pick does. It comes from two questions you genuinely cannot answer: which company wins, and whether today is a good day.

You can delete the first one and take most of the sting out of the second. Buy the industry.

Which in practice means buying an ETF

Why the basket works

HUMN is a humanoid robotics fund holding 47 companies. Its biggest position is UBTech Robotics, at just over 5% of the fund. Tesla sits third, at around 4%.

If Tesla halves tomorrow, HUMN drops about two percent. If Tesla was the only thing you owned, you lost half your money. And the winner, whoever it turns out to be, is in there somewhere.

You give up the chance of picking a ten-bagger. In exchange you stop needing to be right about which company survives.

There were hundreds of American car manufacturers in the early 1900s and almost all of them died. Betting that people would drive cars was easy. Finding Ford in that crowd was not.

The nine sectors

Everything here is a one-year total return as of 25 August, pulled the same day from the same source.

Read down the fee column and something jumps out. The cheapest fund on the list had the best year and the two most expensive ones had the worst.

What the table doesn’t tell you

Five things worth knowing before you buy any of these.

CHPS beat SMH while owning less of the winner. Nvidia is about 5% of CHPS and about 22% of SMH. In a year when Nvidia did well, the fund holding less of it won by 46 points. That is the opposite of what most people expect, and it is the single best argument in this article.

CHPS is also tiny. Around $109 million. That means wider trading spreads and a genuine chance it closes if it never gathers more money. SMH holds $66.8 billion and is not going anywhere. IBOT has the same problem at $99 million. Small funds with great years are still small funds.

UFO is not a pure space play. Its largest holding is Garmin, the satnav company. Nothing wrong with that, but if you bought it expecting rockets, read the holdings list.

MARS is the one to think twice about. It launched in March, charges 0.75%, holds $47 million, and puts a fifth of its money into a single position. You are paying premium fees for a concentrated bet in a very small fund.

NLR’s biggest holding is Constellation Energy, which I own directly. Worth saying out loud.

And one that isn’t in the table at all. Skip RBOT. It’s a robotics fund, but it trades in Canada, and your broker probably won’t sell it to you.

Picking the fund is now the decision

Look at what happened inside single sectors this year.

Same industries, same twelve months.

That happens because funds with near identical labels are built on different rules. Some weight by company size, so the giants dominate. Some spread evenly. Some hold 26 companies, some hold 156. Those are not the same product however similar the name sounds.

Buying an ETF removes one decision and hands you another.

Three things to check first

Open the fund’s own page and read the holdings list. Two minutes, and it tells you more than any article including this one.

Look at the top position. If one company is a fifth of the fund, you’ve bought a concentrated bet with extra paperwork.

Then the size of the fund. Anything under a few hundred million trades thinly and can close.

Then the fee, and where it’s listed.

What I’d tell a friend

Pick two or three themes you actually believe in, not all nine, each small enough that a bad decade doesn’t hurt you.

Buy in instalments. A third now, a third in three months, a third in six. You are not going to time it, and pretending otherwise is how people end up doing nothing at all.

And read the holdings first. The label tells you the story the fund wants to sell. The holdings tell you what’s in it, and this year those were frequently different things.

You can be right about robots and still lose money. Ask anybody who was right about the internet in 1999 and bought Pets.com.

Disclosure: Nothing here is financial advice. I am not a licensed financial advisor. I may hold positions in securities or funds mentioned in this article. Please do your own research.

Leave a comment

That sounds like good news, and for anyone holding it, it is. But what caused the move matters more than the move, because it answers a question people have been arguing about for fifteen years.

What Bitcoin is supposed to be

The pitch has never changed since 2009.

Governments borrow too much. They print too much. Your savings quietly lose value. Bitcoin has a fixed supply that nobody can increase, so it protects you from all of that.

That is the entire argument. It is why people call it digital gold.

This year was the test

Look at what 2026 actually delivered.

US government debt crossed $40 trillion in August, months earlier than forecast. Inflation is still running at 3.4%. There is a war in the Middle East and oil went above $100 a barrel. The dollar’s share of global central bank reserves fell below 57%, the lowest since 1995.

If someone had asked you a year ago to describe the conditions under which Bitcoin finally proves its point, you would have described this year.

So what happened?

Bitcoin peaked near $126,000 last October. By February it was around $60,000.

It fell 52%.

The wider crypto market went from about $4.4 trillion to $2.3 trillion. Two trillion dollars disappeared during the exact crisis the asset was built for.

And gold, the old-fashioned version of the same idea, hit an all-time record of $5,589 an ounce in January. Right in the middle of it.

One of them did the job. The other did not.

So if the debt didn’t move it, and the war didn’t move it, what did?

What actually moved it this week

The US Treasury announced it would buy back more of its own bonds.

That pushed interest rates down slightly. And when interest rates fall, safe savings pay you less, so money goes looking for something riskier. Some of it went into Bitcoin.

That is the catalyst. Not the debt. Not the war. Rates moved a little lower.

Two things then made the move much bigger than it deserved to be.

About $1.9 billion flowed into Bitcoin funds in a single week, with one day above $600 million.

And more than $4 billion of bets against Bitcoin had to be closed as the price climbed. Closing a bet against something means buying it. So the price rose, which forced buying, which pushed the price higher, which forced more buying.

Cheap money and a pile-up. That is the whole story of this week.

Now look at the pattern

Put the two halves together and something obvious appears.

Rates went up through the first half of the year. Bitcoin fell 52%.

Rates ticked down this month. Bitcoin jumped 24%.

Government finances looked their worst in August, when the debt crossed $40 trillion. Bitcoin had already halved by then.

Something that protects you from the financial system does not move like that. Something priced off the cost of money does.

There is a reason for this, and it is the same reason chip stocks fell hard this month while the wider market barely moved.

When safe savings start paying well, anything whose payoff arrives far in the future becomes worth less today. Why wait years for an uncertain return when a government bond pays you over 5% now?

Bitcoin produces no earnings at all. It pays no dividend and generates no cash. Its entire value is a payoff at some undefined point in the future.

That makes it the most rate-sensitive thing you can own. So in a year when the safe return went to a nineteen-year high, Bitcoin should have been one of the worst performers.

It was.

The part almost nobody reported correctly

There is a second thing worth knowing, because the headlines got it backwards.

All year the story was that institutions were abandoning Bitcoin. The outflows were real. Bitcoin funds had their worst six months ever, losing $5.4 billion in the first half. June alone saw $4.5 billion leave.

But look at who was selling.

Over three months those funds shed about 77,000 Bitcoin. Ordinary investors accounted for roughly 100,000 Bitcoin of redemptions on their own, which is more than the entire net outflow.

Over the same period JPMorgan bought 3,000 Bitcoin. Wells Fargo bought 4,000. Citigroup disclosed a position for the first time.

Small investors sold near the bottom. Large institutions bought. The price is now about 34% above where it sat in February.

That is not a story about who is smarter. It is a story about size. A bank holding a position it can comfortably lose behaves very differently from a person watching a number that represents their savings.

What this actually changes

None of this means Bitcoin is worthless or that you should sell.

It means the reason most people give for owning it is not the reason it moves.

If you hold Bitcoin as insurance against government debt and currency debasement, this year was your test case, and the result was a 52% loss while gold set records. The insurance did not pay out.

If you hold it as a high-risk bet that money will get cheaper again, this year makes complete sense. It fell when money got expensive and rallied the moment that reversed. That is a coherent position and plenty of thoughtful people hold it.

Those are two completely different investments that happen to share a ticker.

The practical consequence is about size. Something that can halve in four months belongs in a position small enough that you never have to make a decision about it under pressure. This year’s fund flows show what happens when it is not: ordinary investors sold at the bottom, and the price recovered a third without them.

Work out which bet you are actually making. Then size it for the version where you are wrong.

Because whatever else Bitcoin is, this year settled one thing. Protection is not supposed to lose half its value in the year you needed it most.

Disclaimer: This is not a recommendation to buy or sell any security. Nothing here is investment advice. I am not a licensed financial advisor. Please do your own research.

Leave a comment

Nvidia reports on Wednesday after the close. You already know it will beat. It has beaten on revenue and earnings every quarter for two years, and it has beaten its own guidance thirteen quarters running.

Here is what the stock did the day after each of the last eight beats:

−6.4%, +0.5%, −8.5%, +3.2%, −0.06%, −3.2%, −5.5%, −1.8%.

Six down out of eight. Average −2.7%. The last five in a row were flat or negative. Nvidia has not had a good day after earnings since May 2025.

That is genuinely strange when you think about it. This is the most closely watched company on earth. Thousands of analysts model it. And for two years the market has guessed wrong about the reaction, in the same direction, over and over.

Which tells you the reaction was never about the results.

Where the stock actually is

$214.72 on Friday. A $5.2 trillion company sitting 9.2% below the record it set on 14 May.

Two things about that most people haven’t noticed.

Nvidia is underperforming the S&P 500 this year. Up 15% against about 22% for the index. The largest company in the world, in the middle of the biggest capital spending boom in corporate history, is a market laggard in 2026.

And going into Friday it had fallen five sessions in a row, on its way to a sixth. That would be the longest losing streak since January 2022. It shed roughly $204 billion in a week with no warning, no guidance cut, no company news at all. For scale, $204 billion is more than Disney is worth.

Now hold those next to a third fact. Earnings are up 110% this year and the price is up 15%.

That is a multiple contracting, not a business slowing. The stock trades at 21 to 24 times forward earnings, having spent most of the last three years between 35 and 40. The business kept compounding and the market kept paying less for each dollar of it.

Which matters more than it sounds. A stock priced for perfection falls on a good quarter, because good isn’t perfect. A stock that has already de-rated for four months doesn’t need perfection. It needs the story to stay intact.

So the question isn’t whether Nvidia beats. It’s what the number has to be for beating to count.

Below the line: the three specific figures Wednesday has to hit, including the one Bank of America thinks sits four billion dollars above the rest of Wall Street. Plus the segment Nvidia quietly stopped reporting, where a quarter of last quarter’s profit actually came from, why half a million chips are sitting in a warehouse because Beijing rather than Washington is saying no, and the one bear argument I think is wrong.

Read more

It crossed $39 trillion in March. Five months for the next trillion.

For scale, it took the country nearly two hundred years to borrow its first trillion dollars. It now adds that much in under five months. The government is currently borrowing about $6 billion a day.

Most coverage of this stops at “big number, be worried.” That is not useful. So here is what the number actually is, why it arrived early, who is getting paid, and the part that shows up in your own bank account.

What $40 trillion actually is

It is not a bill anyone is going to send you.

It is the running total of every dollar the US government has borrowed since 1776 and not yet paid back. Every year the government spends more than it collects in tax, and it covers the gap by borrowing. Those gaps, stacked up over two and a half centuries, are the debt.

You will see it divided up. About $116,500 per person. Around $295,500 per household.

Those numbers make for good headlines and they are basically meaningless. Nobody is coming to collect your share. The debt is never repaid the way a mortgage is repaid. Old loans come due, the government borrows new money to pay them off, and the pile rolls forward.

Which sounds like it should be fine. And for a long time it was.

Why it arrived months early

Two things went wrong this year.

In February, the Supreme Court struck down Trump’s tariffs, ruling he had gone beyond his authority in imposing them. Those tariffs were bringing in real money. Now that money is gone, and the government has already refunded over $100 billion of duties it had collected.

That single ruling blew roughly a $200 billion hole in this year’s budget, and something close to a trillion-dollar hole over the next decade once you count the extra interest on the extra borrowing.

Add the cost of the conflict in the Middle East and the deficit for this year is now heading for $2.1 trillion.

So the milestone did not arrive because of one big spending decision. It arrived because revenue disappeared while costs kept climbing.

Now the number that actually matters

Forget $40 trillion. Watch this one instead.

This year the US government will spend around $1 trillion on interest. Roughly $83 billion a month, just to service money it has already spent.

Put that next to the rest of the budget:

• More than the entire national defence budget, by about $150 billion

• More than Medicaid

• More than all non-defence discretionary spending combined

• About the same as Medicare, and projected to overtake it within two or three years

Sit with that for a second. The single fastest-growing thing the American government does is pay interest.

And interest buys nothing. Not one teacher, one road, one hospital bed or one soldier. It is the fee for money that was already spent, on things that were already delivered, years ago.

$40 trillion is the size of the loan. $1 trillion is the rent. The rent is what actually takes things away from you.

So who is collecting that trillion?

Almost nobody asks this, and the answer is more interesting than the headline number.

Roughly 30% of it goes abroad. Japan holds the most American government debt of any foreign country, around $1.13 trillion. Then the UK at about $807 billion, then China at roughly $750 billion. Notice that China, the country most people assume owns America, is third.

The other 70% stays at home. Which means most of that interest is paid by Americans, to Americans.

But “Americans” is doing a lot of work in that sentence. Break the domestic pile down and it is mostly mutual funds and money market funds, then banks, then pension funds, then insurance companies. Behind those institutions sit whoever owns the funds.

And ownership of financial assets in America is not evenly spread. The richest 1% of US households own about 35.6% of all financial assets in the country.

So a large share of that trillion-dollar interest payment, funded by everyone’s taxes, flows back to the people who already own the most.

One honest caveat, because this argument gets oversimplified constantly. Those pension funds hold government bonds on behalf of ordinary retirees, not only the wealthy. Social Security’s own trust fund holds $2.3 trillion of government debt. Money market funds are where a lot of ordinary people park cash.

So the interest is genuinely concentrated at the top. It is not exclusively going there. Anyone telling you it is a pure transfer from the poor to the rich is skipping a step.

Worth knowing where the debt came from, too. Tax cuts since 2001 have reduced government revenue by about $10.6 trillion, and roughly two thirds of that benefit went to the richest fifth of the country. Less tax collected means more borrowing, and more borrowing means more interest owed.

Is this actually a crisis?

Here is where serious people disagree, and it is worth understanding both sides rather than picking a team.

The relaxed view. America borrows in its own currency. Greece could not print euros, which is why Greece had a genuine debt crisis. The US can always print dollars, which means it can never be forced to default the way Greece was. There is no historical example of a country borrowing in its own currency having that kind of collapse.

Treasury Secretary Scott Bessent takes this line, saying there is “nothing magic about the $40 trillion number” and that the country will “grow our way out of this.” He argues the deficit has already peaked.

The worried view. Debt is now around 134% of everything the US economy produces in a year. Debt held by the public passed 100% of the economy this spring, the first time since the Second World War, and is closing on the all-time record of 106% set in 1946.

Marc Goldwein of the Committee for a Responsible Federal Budget put it best. He called the milestone:

that gigantic flashing “check engine” light

Both sides are right about different things, and that is the useful conclusion.

America will not default. That is not the risk, and anyone selling you that story is selling you something.

The real risk is slower and much duller. A government spending a trillion a year on interest has less room to do anything else. Less room to cut taxes. Less room to spend in a recession. Less room to respond when something goes wrong. Nothing dramatic happens. Options just quietly disappear.
That is the government’s problem. The part almost nobody explains is that it has already become yours.

It is showing up on your mortgage, your credit card, your car loan, your student loan and your retirement, in numbers you can check on your own statements this month. Below is what each one is costing you, and the eight things worth doing about it.

Read more

Here is what actually happened that day.

The entire market moved five hundredths of one percent. That is nothing. A small group of stocks fell hard, and everything else carried on as normal.

So why those stocks? The answer has almost nothing to do with AI, and everything to do with something most people never look at.

The thing that actually moved

On Tuesday, the 30-year US Treasury yield touched 5.323%. The highest it has been since 2007.

That sentence means nothing to most people, so let me take it apart.

When the US government needs money, it borrows from the public. You hand over your cash, the government pays you interest every year, and at the end you get your money back. That is a Treasury bond. A 30-year one is simply a thirty-year loan.

The yield is just the interest rate on it. What you get paid.

It matters because this is considered the safest place on earth to put money. In two and a half centuries, the US government has never once failed to pay it back.

So 5.323% is not a number for traders. It is the return anyone can now get for taking no risk at all. Lend for thirty years and your money roughly doubles in fourteen, and you never have to make another decision after the first one.

Why it went up

Three reasons, all straightforward.

The government is borrowing far too much. US national debt is heading towards $40 trillion. Every month the government sells more IOUs. When anyone floods a market with something, they have to offer a better deal to shift it. So it pays lenders more.

Prices are still rising. Inflation was 4% in May and 3.5% in June. Now think about lending money for thirty years while prices rise 4% a year. If your loan pays you 4%, you have earned nothing at all. Your money buys exactly what it did before. So lenders demand more, just to stay level.

And inflation is not calming down, mostly because of oil. Crude went above $100 a barrel in July for the first time since 2022, after the ceasefire in the Middle East broke down. Oil feeds into transport, food and manufacturing a few months later, which means into almost everything you buy.

The new head of the Federal Reserve does not want to cut rates. Kevin Warsh has said there will be no tolerance for high inflation. Investors who spent this year betting on cheaper money have spent the summer being told no.

None of those three fix themselves in a week.

Why this pushes share prices down

Here is the part worth understanding, and it is genuinely simple.

For about fifteen years, safe savings paid you almost nothing. If you wanted your money to grow, you had no real choice but to buy shares and live with the risk.

At 5.3%, you have a choice.

Every investor now faces a real decision. Take a guaranteed 5.3% and do nothing, or buy an expensive company and hope the story comes good. Some of the money that only ever bought shares because there was nowhere else to put it now has somewhere else to put it. So it leaves. And it leaves the expensive stocks first, because those are the ones where you were paying most for hope.

When the safe option starts paying well, expensive hope gets cheaper

Why some shares and not others

This is the bit the headlines skip, and it is the reason the whole market did not fall.

Imagine two people owe you money. The first will pay you next year. The second promises to pay you in 2035.

Now safe savings start paying 5.3%. Which promise just became less attractive to you?

The 2035 one, and it is not close. Waiting nine extra years costs you a lot more when you could have been earning 5.3% the entire time instead. The payment arriving next year barely changes at all.

Companies work exactly the same way.

A company selling soap and soft drinks makes most of its money in the next few years. That is the near promise. A chip company’s share price, though, is built on demand that mostly arrives in the 2030s. That is the far promise.

When safe money starts paying well, the far promise is worth much less today, even though nothing about the company has changed.

That is Tuesday in one sentence. Nothing about AI changed. What changed is the value, right now, of a promise about the 2030s.

It hits these companies a second time

One more thing, which almost nobody mentions.

The four biggest cloud companies plan to spend around $725 billion this year building data centers, up 77% on last year. That spending is not a nice bonus for chip and memory companies. It is where nearly all of their money comes from.

And building data centers means borrowing. At 5.3% instead of 3%.

Nobody cancels a data center because a bond moved. But budgets get reviewed, and things get built a little more slowly. Slower is enough. These share prices were not built for slower.

This is not only a stock market story

The average US 30-year mortgage is now 6.75%. That is the same thing arriving at your front door.

When the government has to pay more to borrow, everyone pays more to borrow. Mortgages, car loans, credit cards, business loans. And at the same time, anyone with savings finally gets paid something for holding them.

Higher rates are not automatically bad news. They are bad news for expensive things bought on hope, and quietly good news for anyone sitting on cash.

What to actually do

Honestly, not much. But if something you own fell hard this week, two questions are worth asking.

When does this company actually make its money? If it is already selling and already profitable, a rate move is background noise. If the exciting product does not ship until 2027, the rate move is the entire story.

Was it expensive before Tuesday? A stock up 300% this year that falls 6% has not become cheap. It has become very slightly less expensive.

That is it. No panic and no ticker list required.

The one thing to take away

A market can look completely calm and be violent underneath. The whole US market moved five hundredths of one percent on Tuesday while a trillion dollars quietly shifted around beneath the surface.

The thing doing the shifting was not AI news. It was the price of safety, and it just went to its highest level in nineteen years.

Every promise about the future is worth a little less than it was in July. That is not a crash. It is a repricing, and it has barely started.

Disclaimer: This is not a recommendation to buy or sell any security. Nothing here is investment advice. I am not a licensed financial advisor. I hold shares in Broadcom. This is a record of what I am doing with my own money and why. Please do your own research.

Leave a comment

On 10 August 2026, Cristiano Ronaldo and Georgina Rodríguez arrived at an office in Lisbon and signed a document. The wedding happened the following day, in Cascais, in near-total privacy. Four witnesses. Their children. No press.

Two people who have spent a decade being photographed chose to be married in front of almost nobody.

But the notary came first, and that’s the detail worth sitting with. Not because signing something before a wedding is unusual among the wealthy. Because of what the document had to solve.

Most people’s assets are simple to describe. A house, a pension, some savings, maybe a car. Every one of those can be valued, split, sold, or handed over without breaking.

Ronaldo’s cannot.

His wealth isn’t a balance in an account. It’s image rights held inside corporate structures. Endorsement contracts running for years with performance conditions attached. Hotel and business interests spread across countries. Intellectual property in his own name and face.

Marriage, legally, is a merger. And when what you own doesn’t behave like property, a merger raises questions that ordinary marriage law was never designed to answer.

The wedding was the small part.

The document was the interesting one.

What they actually signed

Portuguese couples get to choose the property regime their marriage will run under. It’s an active decision, made in advance and formally registered.

They chose separation of assets.

Under that regime, only property held in both names would ever be divided. Anything owned individually stays individual, including things acquired during the marriage.

They also both kept their existing surnames.

That’s public record. Everything after this point isn’t.

The reported terms

Several outlets described further provisions in the days after the wedding. Before I repeat them, the sourcing: these circulated widely across English-language media, and I could not trace a single one of them back to a primary document or a named lawyer.

Treat them as reported, not confirmed.

One note on the figure. You’ll see it quoted as €100,000, £85,900 or $115,500. Those are the same number converted at different rates on different days. The euro figure is what Spanish and Portuguese outlets carry, so it’s what I’ll use.

Three things were described.

A property. Rodríguez receives the family home in Madrid’s La Finca district, reported at around €5.8 million.

An income. A lifetime monthly payment of €100,000, continuing regardless of whether the marriage does.

A legal status. Rodríguez recognised as legal mother of all five children in the household, reported to hold independently of the marriage.

The property number is the one that grabs attention. A €5.8 million house, transferred outright, is an enormous sum by any normal measure.

It’s also, by some distance, the smaller of the two.

Run the second number out

€100,000 a month is €1.2 million a year.

Ten years is €12 million.

Twenty years is €24 million.

Thirty years is €36 million.

Against that, the house is close to a rounding error. The property is the headline; the payment is the actual transfer.

Which raises a question that doesn’t have an obvious answer.

If you’re prepared to move that much value, why structure it as a payment at all? Why not simply hand over a share of the businesses and let her own a piece of what generates the money?

It would be simpler. It would be cleaner. Most divorce settlements do exactly that.

There’s a reason this one reportedly doesn’t.

Some things break when you divide them

A house divides beautifully. You value it, one party takes it, the other takes cash or an equivalent asset. Nothing about the house stops working. It’s still a house.

Now try it with an image-rights company sitting in the middle of a ten-year endorsement contract.

There’s no line down the middle. The contract has performance conditions attached to one person’s continued participation. The counterparty signed on the basis of who controls it. Split the ownership and you haven’t transferred half the value, you’ve introduced a second decision-maker into an arrangement whose entire commercial logic assumes one.

The asset only functions whole.

A prenup at this level is less about who gets what, and more about the fact that some assets stop existing the moment you try to divide them.

Once you accept that, the reported structure stops looking like a negotiation and starts looking like an engineering solution. Hand over what can be handed over. For everything else, manufacture a substitute: a guaranteed income that delivers the economics without touching the machinery.

The third provision fits the same pattern. Fixing legal parentage independently of the marriage isn’t a financial clause at all. It removes a question from the category of things that could ever be reopened by circumstance.

That’s not the language of romance, and it isn’t the language of suspicion either.

It’s the language of people identifying which questions they’d rather never answer under pressure, and answering them early.

Which turns out not to be unusual at all. The uncomfortable part is what happens once families start treating marriage as a structural problem to be designed around, because designs get refined. Some of what’s now being written into these agreements is considerably stranger than a monthly payment, and the profession has a name for the category it now belongs to.

Below the line: what family offices have quietly reclassified marriage as, the structure I found genuinely difficult to think about, and why the same problem is already arriving for people with far less money than Ronaldo.

Leave a comment

Read more

In May, Xi Jinping hosted a state dinner at the Great Hall of the People in Beijing. Trump was visiting for two days and had brought a delegation of American business and technology leaders.

Consider who was in that room.

Tim Cook, running Apple. Jensen Huang, running Nvidia. Lei Jun, who built Xiaomi. Two heads of state. Collectively, several trillion dollars of market value and a meaningful share of the world’s technology supply chain.

A video from that dinner went on to collect more than 82 million views.

It was not about trade policy. Not semiconductors, not tariffs, not the global supply chain.

It was Elon Musk pulling faces.

Executives took turns dropping into an empty seat beside him to pose for photographs. He grimaced. He huffed. By the time Lei Jun arrived and gestured for a selfie, the expression had become something the internet could not stop replaying.

The hashtag about that single photograph collected over 20 million views on Weibo. Clips involving Musk with Huang, Cook and others gathered another 52 million.

Here is the thing I keep coming back to.

That room contained an extraordinary concentration of economic power. Rank everyone present by companies built, capital deployed, or people employed, and you get one ordering.

The attention produced a completely different one.

Nobody in that room decided this. Not the organizers, not the governments, not the executives queuing for photos. It emerged in real time, from millions of people independently deciding what was interesting.

Attention is not distributed by value created. It is distributed by whoever the story is about.

That sounds like a comment about celebrity culture.

It is a comment about finance, and the selfie tells you why.

The rest is for paid subscribers. Below: what Tesla’s actual position in China is right now, why the man asking for that photograph is the more interesting figure in it, what the research says about celebrity CEOs when things go wrong, and the risk both companies have already disclosed.

Read more

Analysis as of Friday, August 14, 2026

More than half my portfolio sits in one theme. I didn’t plan that. Those positions grew while everything else stood still, which is how concentration happens to everyone.

So I went looking for something different. Not the best stock in the market, just something I didn’t already own, in an industry that wouldn’t fall on the same day as everything else.

I found three. A miner, a food company, and an energy producer. Three industries, three continents, nothing obviously in common.

All three traded at about ten times earnings, while the S&P 500 sat at a record high on twenty-two.

Then I spent the week taking them apart, and found the thing they had in common. It changed what I was willing to pay, and it’s the most useful thing I’ve worked out this year.

Why all three looked cheap

A price-to-earnings ratio has two parts. Most people watch the price. With these three, the interesting half is the earnings.

The miner makes about $2,550 of profit on every ounce it digs up, because production costs $1,866 and gold sells for $4,417. That is the widest margin in the industry’s history.

The food company was earning a record profit on every bushel it processed, roughly two and a half times its own five-year average.

The energy producer covers every dollar it spends at an oil price of $50. Oil is $81.

Now put those back into the ratio. Record profits make the denominator huge, which makes the P/E small. The number that looks like a bargain is really just telling you this year went unusually well.

For a commodity business, the P/E is lowest at the top of the cycle and highest at the bottom. It is the exact opposite of a value signal.

Peter Lynch wrote that decades ago. I’d read it. I still had to meet it three times in one week before it properly landed.

None of these are bad companies. I intend to own all three. What changed is the price I’ll pay, because I’m not paying ten times a number that only exists because gold, oil and crop margins happen to be near records at the same moment.

So I’ve set three limit orders. Below are the three names, the specific number on each order, the upside at today’s price versus at mine, where I’d sell, and the evidence that one of the three has already started rolling over. Plus the honest argument against everything I’ve just said, because this reasoning has cost people money before.

Read more

Japan’s ten-year government bond now pays around 2.9%.

That sounds unremarkable. Slightly dull, even. But it’s the highest it has been since 1996, and understanding why that matters explains more about global markets right now than almost anything else you’ll read this month.

Lend Japan money for thirty years rather than ten and you now get more than 4%. Longer loans paying more is ordinary. What isn't ordinary is either of those numbers existing at all.

To grasp what a break that is, you need to know what came before it.

Thirty years of almost-free money

For most of the last three decades, Japan had the cheapest money on earth.

Interest rates sat at essentially zero. Sometimes below zero, meaning you paid the bank for the privilege of holding your own cash. Japanese government bonds paid so little that fund managers described them, quite openly, as un-investable.

Why would anyone lend money for thirty years at almost nothing?

They mostly didn’t. And that had two enormous consequences.

Japanese savers went abroad. Pension funds, insurers, and ordinary households sent their money overseas hunting for a return they could not find at home. A great deal of it went into American government bonds. Japanese investors now hold roughly a trillion dollars of US Treasuries, making them the largest foreign owner of American government debt.

And everybody else borrowed in yen. If you can borrow at nearly zero in Tokyo and lend at five percent in Texas, you do it. Investors did it at enormous scale for decades. That’s the carry trade, which I wrote about a few weeks back, and the short version is that trillions of dollars of global investment were funded with cheap Japanese money.

Japan was, quietly, the world’s lender of last resort. Not because anyone planned it. Because its rates were so low that money simply flowed outward, year after year, looking for somewhere better.

For thirty years, a large part of the world was running on borrowed yen.

Which is fine, right up until Japan stops being cheap.

What changed

Japan’s central bank has been raising rates. Not dramatically, but steadily, after decades of holding them at rock bottom.

Underneath that is something more fundamental. Japan spent thirty years fighting deflation, meaning prices that fall rather than rise, which sounds pleasant and is actually economically corrosive. Nobody buys today what will be cheaper tomorrow.

That’s ending. Wages are rising. Prices are rising. The economy is behaving normally for the first time in a generation.

There’s also a fiscal element. The new government’s spending plans have made bond investors nervous, and nervous bond investors demand higher yields.

So yields climbed. And climbed.

Why 1.75% is the number that matters

Here’s the detail that makes this concrete, and it’s the part I find genuinely fascinating.

Large Japanese institutions have a rough internal threshold, somewhere around 1.75%, at which domestic bonds become worth owning again.

Below that, you may as well take the currency risk and buy American. Above it, you can meet your obligations at home, in your own currency, with no exchange rate exposure and no political risk from another country.

Japanese yields crossed that line and kept going.

Which means for the first time in three decades, the largest pool of exported savings on the planet has a reason to come home.

And it has started.

Japanese investors sold $29.6 billion of US government debt in the first quarter of 2026 alone. The fastest pace in four years.

Some of that selling had a second purpose. When Japan and the US intervened jointly to prop up the yen last week, the dollars Japan spent came from selling exactly these bonds. Two different reasons to sell, both pointing the same way.

Why this lands in your account

You don’t own Japanese bonds. You’ve probably never thought about them. Here’s why it reaches you anyway.

American interest rates are set by supply and demand for American debt. When the largest foreign buyer stops buying, and starts selling, prices fall and yields rise. Every mortgage, car loan and business loan in the United States is priced off that curve. I went into that mechanism in more detail here, and it's the most convincing explanation I've found for why Washington turned up at all.

One estimate puts the scale of it plainly. If Japanese investors repatriate just 5 to 10% of what they hold abroad, that’s $250 to $500 billion of selling pressure landing on US Treasuries, European bonds, and emerging market shares at the same time.

Not because anyone is panicking. Because the maths at home finally works.

And the carry trade has to unwind. All those investors who borrowed cheap yen to buy assets elsewhere now face a rising cost of borrowing. Some will close those positions, which means buying yen back, which pushes the currency up, which forces others to close theirs.

We have already seen what that looks like. In August 2024, a small Japanese rate rise triggered exactly this cascade, which is the same mechanism I covered here, from the borrower’s side. The Nikkei fell 12.4% in a single session, its largest one-day drop in history. Markets from New York to Frankfurt had a spectacularly bad Monday.

Nothing had happened to Japanese companies. They were fine on the Friday and fine the following week.

A modest interest rate decision in Tokyo moved the value of assets across thirty countries within four days.

The part I keep thinking about

There’s a quiet irony in all this.

For thirty years, everyone treated Japan as the world’s cautionary tale. The economy that stagnated. The stock market that took decades to recover its 1989 peak. The place where policy ran out of ideas.

And during all of that, Japan was underwriting the returns of everyone else. Its citizens’ savings flowed outward and funded asset prices in America, Europe and the emerging markets. Its near-zero rates made leverage cheap for anyone who wanted it.

Japan being broken was, in a strange way, load-bearing for everyone else.

Now Japan is recovering. Wages are up. Prices are normal. The stock market is at record highs and the currency finally has a real interest rate behind it.

That’s genuinely good news for Japan.

It just means the rest of us have to start paying for our own money.

What to actually take from this

You cannot do much about any of it directly, and I’d distrust anyone selling you a trade based on Japanese bond yields.

But three things are worth carrying.

The cheapest input in global finance is getting more expensive, and that has to show up somewhere. Probably in bond yields, possibly in how much asset prices can support.

The scary moves start in boring places. Nobody worries about Japanese government bonds. That’s precisely why the 2024 unwind caught everyone off guard. When something enormous and dull starts moving, it matters more than something small and dramatic.

And your portfolio is connected to things you’ve never heard of. If you own an index fund, some part of what you own has been supported by cheap Japanese money for years. Nobody sends you a note explaining that.

Japan spent three decades as the world’s quiet lender.

That era is closing, slowly, and the bill is being distributed to everyone who didn’t realize they were borrowing.

Two companion pieces if you want the rest of this story. Why America bought yen for the first time since 1998. And the version of all this already sitting in your own account.

Sources: Japan's bond market back in play, CNBC · How Japan's bond market affects your portfolio, Forbes · Wall Street braces for Japanese capital repatriation · 30-year JGB breaks 4% · Global impact of Japan's rate shift, OFX

Leave a comment

It depends entirely on what you want your money to do.

First, what these companies actually are

A water utility is a company that pipes water to your house and sends you a bill. American Water, Essential Utilities, California Water. Unglamorous businesses, and about as predictable as businesses get.

One fact shapes everything else about them, and it’s worth understanding before you buy one.

They are legal monopolies that aren’t allowed to set their own prices. Nobody else can pipe water to your street, and in exchange a state regulator decides how much profit they may earn.

The company invests in pipes and treatment plants, adds up what that’s worth, and asks the regulator for a return on it.

In 2026, regulators granted 9.55% in Pennsylvania, 9.75% in Maryland, and 9.80% in West Virginia.

Notice how narrow that band is. In each case the company asked for around 10.75% and got roughly 9.7%. That number is a ceiling on the entire business, and it explains both halves of this article.

Where water genuinely wins: the income

This is the part that’s real, and it’s more impressive than most people realize.

American States Water has raised its dividend every single year for 72 years. That is the longest active streak of any publicly traded company in the United States. Not the longest in water. The longest, full stop.

Think about what 72 years covers. Every recession since 1954. The oil shocks. Double-digit inflation. The dot-com crash. 2008. Covid. Through all of it, the cheque went up.

York Water goes back further still on a different measure. It has paid a dividend for over 210 years, since 1816, which it believes to be the longest unbroken run in America.

And these payments aren’t creeping up by pennies. Over the past decade, dividends grew at:

For context, the entire S&P 500 pays a dividend yield of just 0.97%. Every water utility on that list pays you two and a half to three and a half times more.

If your goal is income that reliably grows faster than inflation, without you having to do anything, this sector delivers exactly that. It has done so for longer than most countries have had their current constitutions.

Where water loses: the growth

Now the other half, and I want to be straight about it.

Total return means dividends plus whatever the share price does. It’s the number that actually matters for building wealth. Here it is over ten years:

Now over twenty years, which is the fairest test I could find. The oldest water ETF launched in December 2005.

$10,000 in the water fund became about $53,000. The same $10,000 in the S&P became about $89,244. Same two decades, same starting dollar, roughly $36,000 apart.

And the recent picture is harder still. Over the past five years, with dividends reinvested, the pure water utilities have actually lost money:

That’s not underperformance. That’s a smaller pile of money than you started with, five years later, after collecting every dividend.

Why the two halves point in opposite directions

Go back to that 9.7% ceiling.

A water company earns roughly 9.7% on its invested capital because a regulator says so. It pays out about 60% of that to shareholders, which is where your reliable dividend comes from. That leaves under 4% to reinvest for growth.

To grow faster, it has to raise money by issuing new shares, which slices the pie into more pieces.

The dividend is dependable precisely because the growth is capped. Those aren’t two separate facts. They’re the same fact seen from two angles.

A technology company can double its profit in a year. A water utility legally cannot. That’s the trade you’re making.

The catch nobody mentions

Here’s the fact that surprised me most while checking this.

A high-yield savings account currently pays around 4.15%. A money market fund pays about 3.7%. Not one water utility yields that much.

The highest, Essential Utilities at 3.63%, still pays less than cash sitting in a savings account, with none of the risk.

The dividend growth is what justifies owning the stock. Your savings account will not pay you 8% more next year, and 8% more again the year after. Over ten years, a 2.5% yield growing at 8% annually overtakes a flat 4.15% comfortably.

But if you’re buying water for income you need this year, you are taking equity risk to earn less than a savings account. That’s worth knowing before you start.

One trap to avoid

If you decide water is for you, be careful which fund you buy.

PHO, the largest water ETF in America, yields 0.55%. That is less than the S&P 500 itself.

The reason is that it isn’t really a water fund. Its biggest holdings are Roper (a software conglomerate), Waters Corporation (which makes laboratory instruments), IDEXX (veterinary diagnostics) and Agilent (more lab instruments). Exactly one water utility appears in its top ten holdings.

If you want water utilities, CGW is the only fund that actually holds them. American Water, Xylem, SABESP, United Utilities, Essential and Severn Trent fill its top six. It yields 1.51%.

Better still for income: just buy the utilities directly. The ETFs dilute exactly the thing you’d be buying them for.

So who is this actually for?

Water utilities make sense if you:

Are retired or approaching it, and want income that grows without you managing anything. Already hold enough growth assets and want something that behaves differently. Value predictability over performance. Have a ten-year-plus horizon, so the dividend growth has time to work.

They don’t make sense if you:

Are trying to build wealth from a small base. Have twenty or more years and can tolerate volatility. Want to beat the market. Need income immediately, in which case cash currently pays you more.

What I’d actually do

If I wanted this exposure, I’d buy two or three utilities directly rather than a fund. American States Water for the 72-year record. Essential Utilities for the highest yield, noting it’s being acquired by American Water in a deal expected to close early next year. California Water for the 59-year streak.

I’d size it as an income sleeve, not a growth position. Somewhere around 5% of a portfolio, not 20%.

And I’d go in with the right expectation: roughly 7 to 8% a year in total, made up of a 2.6% dividend plus modest growth. Perfectly decent. Meaningfully below what the index has delivered.

One thing to watch: these stocks are sensitive to interest rates, because a 2.6% yield competes directly with savings accounts and government bonds. The five years of losses above were caused almost entirely by rising rates, not by anything the companies did wrong. Their earnings grew the whole time.

That cuts both ways. If rates fall, these recover quickly. But right now the Federal Reserve is holding at 3.50–3.75%, three of its members voted for a hike at the last meeting, and markets are pricing more increases rather than cuts. That’s a headwind, not a tailwind, as things stand today.

If you missed it, the piece that started this one is here: The Rothschilds Bought Water in 2019. Everyone Else Followed.

Not financial advice. This is my own research and analysis. Please do your own.

Leave a comment

I want to tell you about a story that has almost nothing to do with technology, even though it looks like it does.

It starts in 2024, with a young man getting fired.

How it started

Leopold Aschenbrenner worked at OpenAI. He was on a team called superalignment, whose job was working out how you keep control of something smarter than you are.

He was in his early twenties. Columbia at nineteen. And then, in April 2024, he was fired.

The reason is disputed and it matters. He had written a memo to OpenAI’s board warning that the company’s security was, in his words, egregiously insufficient to protect against foreign actors stealing model weights or algorithmic secrets. He meant China. He got a warning from HR. Then he was fired, officially over leaking an internal document, which he describes as a harmless brainstorming note shared with three outside researchers.

He says he was told the security memo was a major reason. OpenAI denies that, and says the firing had nothing to do with retaliation.

Remember that, because the argument he’d just been fired for making becomes the most distinctive part of what he wrote next.

Most people, at that point, go quiet for a while.

He wrote 165 pages instead.

The essay was called Situational Awareness: The Decade Ahead. He published it in June 2024, for free, on the internet.

And here’s the thing that makes it worth your attention. It wasn’t a complaint about being fired. It was a forecast.

What he actually said

His argument, stripped of the technical parts, went roughly like this.

AI systems are improving on a predictable curve. Not a mysterious one. You can measure how much computing power gets thrown at them, how much more efficient the software gets each year, and how many artificial limitations get removed from models that already exist.

Extend those three lines forward and you land somewhere startling. He put artificial general intelligence, meaning systems broadly as capable as people, at around 2027.

Then he followed the money.

If that’s coming, someone has to build it. That means data centres the size of small cities, chips by the million, and enough electricity to run a mid-sized country. He put the eventual bill at roughly eight trillion dollars a year in infrastructure spending.

And then the part that made governments read it. Something that powerful becomes a national security matter. He predicted the American government would effectively take over the leading labs by 2027 or 2028, and that a race with China would become the organising fact of the decade. Which is, almost word for word, the memo that got him fired.

Why it mattered so much

Plenty of people write predictions about AI. Almost nobody reads them.

This one was different, for a reason that has more to do with finance than with technology.

Wall Street had a problem in 2024. Enormous amounts of money were about to be spent on chips and data centres, and nobody could quite explain, in one place, why it was rational rather than a bubble.

Aschenbrenner’s essay was that explanation. It laid out a coherent story in which the spending made sense, with numbers attached, written by someone who had been inside one of the labs.

If you’ve ever wondered why capital has flooded into memory, semiconductors and nuclear power over the last two years, this document is a real part of the answer.

It gave the buildout a justification you could hand to an investment committee.

Then he did the thing almost no forecaster ever does.

He bet his own money on it.

And he was right

He started a hedge fund. He called it Situational Awareness, after the essay. Getting money was not the problem. His backers came to include Patrick and John Collison of Stripe, Nat Friedman, formerly of GitHub, the investor Daniel Gross, and the trading firm Jane Street.

Not naive money. Some of the most sophisticated capital in technology, backing a man in his early twenties with no track record whatsoever.

The Financial Times reported returns of 1,551% since it opened.

In the first six months of 2026 alone, it was up 439%.

At its peak the fund controlled around $45 billion, a figure inflated by borrowed money, almost all of it in the physical layer of the AI buildout. Memory manufacturers, cloud providers, chip companies, the boring infrastructure his essay had said would be needed.

Let me put that in normal terms. If you’d given him a thousand dollars when he started, you’d have had about sixteen thousand back.

That’s not luck. That’s an argument, turned into a portfolio, working.

Which is where the story stops being about being right.

The rest is for paid subscribers. Below: the $8.5 billion in insurance he was quietly carrying, the single listing that set off the collapse, why both halves of a hedged portfolio failed at once, and what actually survived.

Read more

I closed two positions this week. Both were up. Both are companies I still think are good businesses.

Let me explain that before I get to which ones.

The backdrop

Stocks are expensive by a measure that only gets this extreme twice a century. The Shiller CAPE ratio, which compares prices to ten years of inflation-adjusted earnings, closed Friday at 42.39. That is the highest reading since August 2000.

And the macro picture flipped in a single session. Through late July, markets were pricing better-than-even odds of a Fed rate hike in September. Then Friday’s jobs report showed the US shed 23,000 jobs against expectations of adding 83,000. Hike odds collapsed to about 44% within hours.

An expensive market and an economy showing its first real crack. That isn’t a forecast. It’s a reason to be deliberate about what I own and how much of it.

Why I sold two things that were working

Neither sale was about the company.

One had quietly grown into too much of my portfolio. Not by my adding to it, just by it going up while everything else stood still. That’s how concentration happens. It accumulates while you’re not looking.

The other ran very hard, very fast, on a move driven more by sector rotation than by anything the company did. A stock can travel from cheap to fully valued in a fortnight without the business changing at all.

Selling a stock is not a verdict on the company. It’s a statement about the price, the position size, or both.

You are allowed to like a company and still think its shares have got ahead of themselves. Those two beliefs don’t contradict each other.

How I invest, and how you probably should

I’d hate for anyone to mirror my activity without understanding it.

I trade actively. When I see a short-term gain worth taking, I take it. That comes with real costs: fees, tax, and the risk of outsmarting myself.

Every company I own, I own because I’d be happy holding it for years. The trading is something I do around good businesses, not instead of owning them.

If you buy one of these companies and simply hold it, you haven’t made a worse decision than I did. In several cases you’ll have made a better one.

Nothing here is a signal to sell within weeks. When I take a profit it’s usually because a position grew too large or the price outran the business. Neither means the company is finished. If your position size is sensible and you’re investing for years rather than months, the correct action is very often to do nothing at all.

What happens to the money

Honestly, I don’t know yet, and I’m not rushing. The one consistent pattern in my record is that patient decisions make money and quick ones lose it.

I do expect to buy back into both, at prices I’ll name below.

Below the paywall

Exactly what I sold, what I paid, what I got, what each made, and where I’d re-enter both.

Read more

Uber reported yesterday morning and the stock dropped 5.29% to close at $68.18, down $3.81 on the day.

If you only saw the headline, you’d think something broke. Nothing did.

The quarter was really good

And the number I think matters most isn’t even in that table. Uber generated $2.79 billion of free cash flow in the quarter, taking its trailing twelve month total to $10.12 billion.

First time in company history it’s cleared $10 billion in a year. This used to be a business famous for setting money on fire.

So everything sped up, and the stock fell 5%. Here’s why.

The “guidance miss” was 0.13%

What analysts wanted: $59.33 billion
What Uber guided: $59.25 billion
The gap: $80 million, or 0.13%

Eighty million dollars on a fifty nine billion dollar forecast. A rounding error.

The earnings guide came in three cents light too, at $0.84 to $0.88 against $0.89 expected. And even that still means earnings growing 28% to 35%.

Nobody wipes $7 billion off a company over three cents. So something else was going on.

The margin scare that wasn’t

Revenue grew 12% while bookings grew 22%. In Mobility it looked worse: bookings up 20%, revenue up 1%.

Read fast, that says Uber is suddenly keeping far less of every ride. It isn’t.

Back on January 2nd, Uber changed how it operates in parts of the UK for tax reasons. Uber is no longer the responsible party for those rides, which means money paid to drivers now gets subtracted from revenue instead of counted as a cost lower down the page.

Same money in, same money out, smaller revenue number.

CFO Balaji Krishnamurthy said it plainly on the call:

“About 400 basis points is entirely related to this U.K. business model change, and it’s an optical impact. It moves cost from cost of revenue.”

Here’s the cleanest way to see it. The UK change only affects Mobility. Delivery is untouched. Look at what happens to the revenue column:

Delivery’s revenue tracks its bookings almost exactly, like you’d expect. Mobility’s doesn’t, purely because of where the driver payments now sit on the page.

And look at the last column. Mobility’s operating income grew 28% with margins holding at 7.6% of bookings. The profit was always there. Only the revenue line moved.

What actually spooked people

This part is real.

Uber committed more than $10 billion to self-driving cars. Roughly $2.5 billion for stakes in the companies building the technology, and about $7.5 billion toward fleets and commitments to buy around 120,000 vehicles.

Dara Khosrowshahi didn’t hedge:

“Our ambition is straightforward, to become the world’s leading commercialization platform for autonomous vehicles.”

For ten years Uber’s entire appeal was that it owned nothing. No cars, no depreciation, no fleet. Just software taking a cut. Now it’s putting billions into assets that lose value the moment they leave the lot.

That’s a different company, and investors aren’t sure they want to pay the same price for it.

Some perspective though: self-driving cars are currently under 0.5% of Uber’s roughly 300 million weekly trips. This is an argument about 2029.

On Waymo, since everyone asks: the Phoenix partnership ended in June when the contract ran out. Austin and Atlanta continue. Waymo launches its own app in both cities in January 2028, though its cars stay on Uber through at least May 2028.

What it costs today

At $68.18, Uber trades at about 17 times forward earnings. DoorDash trades at roughly 35 times, while Uber’s own Delivery business grows bookings at 25%.

But here’s the chart that actually made me stop and look. These are the analyst price targets, including the most negative views on Wall Street:

Every single one is above the current price. Even the analyst who rates it a sell thinks it’s worth $73.

That doesn’t mean they’re right. It does mean the stock has fallen below the point where anyone covering it, bulls or bears, thinks it belongs. The entire argument is about 2029, not about the business today.

So what am I doing?

Uber is my biggest holding, about a fifth of my portfolio, bought at an average of $71.58. I’m down roughly 4.8%.

Below: whether I’m holding or selling, and the exact price where I’d take money off the table.

Subscribe now

Read more

Last Friday, the United States Treasury walked into the currency market and bought Japanese yen.

Read that again, because it’s stranger than it looks.

America does not do this. The official position for decades has been that markets set exchange rates and governments keep their hands off. The US Treasury even publishes a report twice a year naming countries it suspects of pushing their own currencies around.

The last time America bought yen alongside Japan was 1998. Bill Clinton was president. Most people trading currencies today were in school.

By Monday both governments had confirmed it, and the dollar had fallen from above 163 yen to around 156.

So something was worth breaking a twenty-eight-year habit for.

It wasn’t the yen.

What actually happened

Currencies get priced like tomatoes. More buyers, price up. More sellers, price down.

For years now, the world has been selling yen. Steadily, relentlessly, every single day.

Intervention is what happens when a government finally decides it has had enough and becomes the buyer itself. Japan’s Finance Ministry reaches into its dollar reserves, walks into the market, and starts buying yen.

That’s it. That’s the entire mechanism. No clever financial engineering. The government simply becomes a very large customer for its own money.

And it works precisely as you’d expect. An enormous buyer shows up, the price jumps.

The interesting question is why they had to.

The first thing everyone gets wrong

Most people assume Japan wants a weak yen.

It’s a fair assumption. A cheap currency makes your exports cheaper abroad, so your carmakers and electronics firms sell more of everything. Countries have been accused of weakening their currencies deliberately for exactly this reason. And Japan is the export nation.

But Japan has a problem that swallows the benefit whole.

It imports almost all of its energy and much of its food.

Every barrel of oil, every shipment of gas, every container of imported ingredients gets paid for in dollars. When the yen weakens, all of it costs more, and that lands directly on household electricity bills and supermarket shelves.

So a weak yen is wonderful for Toyota and miserable for anyone in Osaka filling up their car. When the currency hit a 40-year low, that misery got loud enough to force the government’s hand.

Which gives you a rule worth keeping. A cheap currency is neither good nor bad. It quietly moves money from the people who buy things abroad to the people who sell things abroad, and governments step in when that transfer becomes politically unbearable.

The second thing everyone gets wrong

The assumption that intervention works.

Japan has been doing this on repeat. September 2022, roughly $20 billion. A month later, another $43 billion. Spring 2024, about $62 billion across two rounds.

Then in July 2024, they spent $20.7 billion in a single day. The largest one-day currency intervention in recorded history.

Four years. Well over a hundred billion dollars. And this month the yen hit 163 anyway.

Here’s the uncomfortable part.

Every single one of those interventions worked. For about a week.

That’s not incompetence. It’s structural, and once you see why, you understand almost everything about how currencies move.

What’s actually going on

I wrote a while back about carry trades Read here. This is that story reaching its conclusion.

Money flows to wherever it earns the most. For twenty-five years, Japan had the cheapest money on earth, with interest rates near zero and occasionally below it. Meanwhile the United States was paying four or five percent.

So you borrow yen for almost nothing, park it in dollars earning five percent, and pocket the difference. Millions of institutions did exactly that, at enormous scale, for a generation.

Now here’s the bit that matters.

Borrowing yen means selling yen.

Every one of those trades pushed the currency down a fraction. There were an extraordinary number of them.

So the yen isn’t weak because traders are picking on Japan. It’s weak because a gap in interest rates made selling yen profitable, and that gap stayed open for twenty-five years.

Set that against what intervention actually does. Twenty billion dollars in an afternoon is a colossal sum of money. Currency markets trade trillions of dollars a day.

You’re bailing water from a boat that still has a hole in it. The level genuinely drops while you’re bailing. Nobody has touched the hole.

The part almost nobody mentions

There’s a detail in how this gets paid for that explains America’s presence better than anything either government said out loud.

Japan’s foreign reserves are not stacked in a vault as cash. They’re held overwhelmingly as US government bonds.

So when Japan buys yen, it’s selling dollars. And to have dollars to sell, it sells Treasuries.

Japan is one of the largest foreign holders of American government debt. When it sells at scale, US bond prices fall and American interest rates drift upward. Every mortgage and car loan in the country is priced off that curve.

Japan already sold nearly $30 billion of Treasuries in the first quarter of this year, the fastest pace in four years.

Which reframes the whole thing. A coordinated intervention isn’t diplomatic courtesy. If Japan is going to defend its currency by selling American debt, Washington has a very direct interest in how that’s done and in it finishing quickly.

Why America turned up

Two details give away the rest.

This wasn’t improvised on a Friday afternoon. The intervention was carried out under a joint statement the two finance ministries quietly issued back in September 2025. The framework was already built and sitting there, waiting for the moment.

And the language both sides used was carefully chosen. They called the problem “excessive volatility and disorderly movements.” Not the level of the yen.

Nobody is officially saying 163 is the wrong price. They’re saying the speed was dangerous.

There’s a recent lesson in why.

In August 2024, a small Japanese interest rate rise forced everyone holding those borrowed-yen trades to unwind at once. The Nikkei fell 12.4 percent in a single session, the largest one-day drop in its history, worse than Black Monday in 1987. Markets from New York to Frankfurt had a spectacularly bad Monday.

Nothing had happened to Japan. Japanese companies were fine on Friday and fine the following week.

That is what “disorderly” means, and it’s why the American Treasury showed up. Not for the yen. To stop a chain reaction that begins in Tokyo and ends in everybody’s retirement account.

Japan has said it will not hesitate to do it again.

The yen story was never really about Japan.

It’s the clearest visible proof of something true everywhere. The number on your statement is not simply what your money is worth.

It’s a price that somebody, somewhere, has decided is worth defending.

Leave a comment