Dario Amodei shouldn't be running a public company.

Anthropic isn't listed yet. It wants to be, and soon. Reports have it looking to raise as much as $100bn at around a $2 trillion valuation, which would be the biggest IPO ever.

And this week gave us two pretty good clues about why the man at the top is the wrong fit for that job.

Clue one. On Thursday night, Trump hosted Xi Jinping for a state dinner at the White House, with AI one of the big topics between them.

OpenAI’s Sam Altman was there. So were Jensen Huang, Tim Cook, Mark Zuckerberg, Elon Musk and Jeff Bezos, basically the whole American AI stack sat round one table.

Amodei wasn't even on the White House guest list. Nobody's given an official reason, and Anthropic hadn't said he was going, so I'm not going to pretend I know it was a deliberate snub.

But when you're about to ask the market for $100bn, the boss of one of the two biggest AI labs in America missing the one dinner where the US and China talk AI is… not a great look.

Clue two. The IPO has slipped again.

On Sept 4th Reuters reported marketing would start mid-October at the earliest.

On Sept 28th the WSJ said November.

By Thursday, reports had it sliding past the midterms.

Companies push listings back all the time, fine, but twice in one month is unusual.

Why he's the wrong CEO for a listed company

He's clearly brilliant. He's built a research lab into a business doing a reported run-rate of more than $65bn by the end of July, up from about $9bn at the end of 2025. The business isn't the problem.

The problem is he's dogmatic, and public markets punish dogma.

In February, the Pentagon told Anthropic to drop two limits on how Claude could be used, covering fully autonomous weapons and mass surveillance of Americans.

Amodei refused and put a $200m contract on the line. The administration then labelled Anthropic a supply chain risk and ordered federal agencies to stop using its products.

Anthropic took it to court and has mostly won so far. You might think he was right. Plenty of people do. But a private lab owned by investors who signed up for the mission can afford that fight.

A listed company with pension funds on the register has to explain to shareholders why it picked a fight with the US government on principle.

Then a couple of weeks ago he published a 3,800-word essay calling for the whole industry, his own company included, to slow down:

Read that from the seat of someone planning to buy the IPO.

The CEO is publicly arguing his product should get better more slowly, weeks before he asks you for $100bn.

Some coverage has linked the delay to that essay, although the WSJ says the switch to November was decided before the safety row blew up. Either way, you know where his priorities sit.

The structure backs this up. Anthropic is a public benefit corporation, and a body called the Long-Term Benefit Trust gets to pick board members. It's all legal and disclosed.

It also means that when the mission and your returns clash, your returns don't automatically win.

Investors tend to prefer returns.

Pulling back

I'm not saying Anthropic's a bad business. You don't grow like that by accident, and I'm not calling the top on AI either.

Invest in the IPO?

Ask the boring questions before day one gets you excited. Who controls the board? What can the Trust do that shareholders can't stop?

And the next time Dario's principles cost a government contract, what happens to revenue?

What to watch

The public S-1 and its governance section. The Q3 numbers the bankers reportedly want to show first. And who anchors the deal, because Reuters reported Nvidia is in talks to put in up to $10bn, which is a supplier buying into its own customer. More on that in What We're Ignoring.

Brilliant founders make bad public company CEOs all the time.

I think Dario's one of them.

Back to the dirt

The money is going back into infrastructure. Prices haven't all followed yet, and that gap is the interesting point.

Infrastructure ETFs took in $232m in the week to 18 Sep, $1.38bn over 30 days and $3.2bn over 90 days, per ETF Action.

All while the group was down 1.6% on the week, 5.7% on the month and 8.9% over three months. Investors are accumulating.

Earlier in the month, ETF Themes had infrastructure ETFs pulling in $263m in a single week, more than the whole prior month, with PAVE taking about $236m on its own and GRID about $231m over the month.

Meanwhile, clean energy funds lost about $198m in a week. The market has worked out that data centres need power, full stop, and it doesn't much care where the electron comes from.

This week the picks and shovels started moving. To Thursday's close, Eaton (ETN) is up 10.6% on the week, Quanta (PWR) 3.9%, GE Vernova (GEV) 3.2%, the data centre ETF DTCR 2.5% and GRID 2.3%. Caterpillar is up 2.9% (Stock in Focus below).

Broad industrials haven't caught up. XLI is flat on the week and down 5.7% on the month, and PAVE is down 5.1% on the month.

That's where the interesting gap is.

So far, this month has seen a tech-led (or tech-only) rally.

Where are all of those chips going to live if we don’t build the infrastructure?

Looks like a CATch up opportunity to us.

Tech has just kept running. SMH was up 7.1% on the week and XLK 5.9%.

Utilities are getting hammered by rates. XLU is down 4.7% on the week and 8.9% on the month, with the US 10-year at 5.22%, its highest since 2007. Bond proxies hate that.

Nuclear is worse, with NLR down 10.5% on the month and URA down 10.6%. The concept end of the power trade needs cheap money, and it isn't getting any.

The Lights

🟢 Power picks and shovels - ETN, PWR, GEV, GRID. Flows in, prices turning. Huge order backlogs.

🟢 Semis and big tech - Still out in front, with SMH +7.1% and XLK +5.9% on the week.

🟡 Broad industrials and construction - Money going in, price still down on the month. CAT sits here.

🔴 Utilities and nuclear concept names - A 10-year at 5.22% is doing the damage. XLU -8.9% and NLR -10.5% on the month.

Caterpillar (CAT)

Yes, again. We had CAT here on Sept 4th after it bounced off the 200-day. Three weeks on, it's still sitting on that line, and the picture underneath has drastically improved, which is why it's back.

The chart

CAT closed Thursday at $805.25, down 0.83% on the day, and was indicated at $813.25 pre-market this morning.

It peaked at $1,073.46 intraday on 30 June. That's 25% off the high, and at the low on 2 Sep ($771.39) it was about 28% off.

It's still up around 40% this year, which tells you how big the run up was.

The 200-day moving average is at about $787, just under the price, and CAT has held it all through September. The 50-day (about $832) and 100-day (about $881) are overhead. The old breakout zone around $796 to $803 sits right on top of the 200-day, so you've got two supports stacked in the same place.

Here's what our indicator says.

Momentum positive, velocity accelerating, 20-day bull, 200-day bull and trajectory neutral. The 50-day and 100-day are both still broken, and it's 61 days since the peak.

In plain English, that's a big run, a 25% pullback that flushed the tourists out, and now it's holding the 200-day with momentum turning back up while the crowd's looking elsewhere.

Why Caterpillar is our pick again

Q2 was a record $20.5bn of sales and revenue, up 24%.

Power & Energy did $8.2bn, up 17%, with segment profit up 30% to $2bn.

Power generation retail sales grew 72%, which CAT put down to very strong demand for large generator sets and turbines for data centres.

Backlog hit $72bn, with about 59% due to ship in the next 12 months and some power customers ordering out to 2030.

They're even restarting a 10-MW gas engine line they stopped making in 2022, bringing back 1.5 GW of capacity with shipments from Q4.

Construction was up 35% to $8.3bn, with North America up 50%, and they raised full-year guidance to mid-to-high teens growth.

Incredible growth metrics! This is the physical side of the AI build. A data centre can't wait years for a grid connection, so it buys its own power, and CAT sells the kit.

Asked about AI spending on the earnings call, CEO Joe Creed said nobody is slowing down at the moment.

Where it fits in the process

This sits in the Entry layer of our criteria.

A quality company has pulled back to a big average and the crowd has buggered off, which is exactly what a process is meant to spot.

But the 50- and 100-day are still broken, so it hasn't proved anything yet.

Your line is the 200-day. If CAT closes below it, you should already know what you're doing, because you wrote the rule down before it happened.

If it gets back above the 50-day, it’s probably time to buy more.

What we need to see

We need it to hold the 200-day. We need power generation orders to stay strong next quarter. And we need tariff costs (CAT expects $2.2bn this year before any refunds) not to eat the margin.

Hit reply with CAT in the subject and tell me your rule if it closes under the 200-day. If you haven't got one written down, that's a gaping hole in your process.

The indicators above, plus the process driving our Qualify, Enter, Size, Manage, Exit criteria, is what FINK OS members run every stock through. Have a look at FINK OS here.

Not financial advice. Price data to 24 Sep close. Business figures from Caterpillar's Q2 2026 results.

We're ignoring the Enron comparisons on AI's circular financing. We're not ignoring the maths.

Here's circular financing in plain English. I sell shovels. I give you money, you use my money to buy my shovels, my sales go up, your valuation goes up, and the stake I hold in you goes up too.

Everyone looks great, as long as you eventually dig up something somebody else pays for.

That's roughly what's going on in AI right now:

So why ignore the doom threads in the face of all of this ‘evidence’?

Because the end demand is real. Nvidia did $96.2bn of revenue last quarter, and Anthropic's run-rate is reported above $65bn, so this isn't a company invoicing its own subsidiary.

That same money is paying for the infrastructure boom in Theme Watch, and it's a big reason CAT is sitting on a $72bn backlog.

The maths still matters, though. There is a risk here. All we’re saying is that every risk doesn’t end in disaster.

If a supplier is funding its customer to buy its kit, some of that demand is just the supplier's own cash going round in a circle. That's fine while the customer can keep raising money.

Reuters, citing the FT, reported loans on one Oracle-leased campus being quoted at 89 to 91 cents on the dollar. Nobody's defaulted, but that's where you'd see the cracks first.

The rule is to know which bit of the loop you own. A company shipping 59% of its backlog within a year and getting paid for it is a very different risk from the lender, or the vendor who's also the investor.

And if Anthropic lists with a chip supplier as anchor, ask how much of the demand for the shares is really demand for the chips.

What if Japan keeps going?

Japan's 30-year government bond yield hit 4.223% this morning, the highest since the bond was launched in 1999. The 10-year closed Thursday at 3.075%, its highest since August 1996, and touched 3.115% today.

For a country that borrowed at basically zero for most of the last 20 years, that's a massive shift. The Bank of Japan hiked to 1.25% on 18 Sep, and two of the nine board members voted against it.

So why should you give a shit if you own US tech? Three reasons.

First, the carry trade. For years, people borrowed yen at next to nothing and bought higher-yielding assets everywhere else, US tech included.

When Japanese rates go up and the yen jumps, everyone tries to get out of that trade at once. That's what happened on 5 Aug 2024, when the Nikkei fell 12.4% in a single day.

Second, repatriation. Japan is the biggest foreign holder of US Treasuries, with $1.104tn in July. If Japanese pension funds and insurers can get 4% at home in their own currency, why take the currency risk on Treasuries or gilts? Less Japanese buying means higher long-end yields everywhere else.

Third, valuations. The US 10-year hit 5.22% on Thursday, the highest since 2007, and the 30-year hit 5.50%. Long-duration tech is priced off those numbers, and a higher discount rate usually means a lower multiple.

The twist is the yen's still weak, past 157 to the dollar after the hike, so the carry trade hasn't broken. It's being squeezed rather than blown up, for now.

Let's play a game.

The 30-year JGB goes to 5%. What happens first?

A) US long bonds sell off harder
B) Big tech takes the hit
C) The yen rips and the carry trade unwinds
D) Nothing, markets shrug it off

Reply and let us know what you think and why.