Steno Signals

Happy Monday, and welcome back to my editorial on everything macroeconomic! I had the whole thing planned. The U.S. CPI was semi-hot on Friday: Core at 0.29% against a 0.22% consensus (and above our own nowcast, which had the headline right and core too soft). The fence-sitters got their number, and a September hike became the base case. I was going to write you a nice, sober editorial about what a first hike does to markets.

Then, on Saturday morning, Dario Amodei decided that macro was not going to have the weekend to itself. The CEO of Anthropic published a nearly 4,000-word essay titled “We Must Pace the Frontier”, asking the entire AI industry to deliberately SLOW DOWN. Otherwise, rogue agent-swarms might hijack the internet within six to twelve months. Sam Altman agreed within hours. Elon Musk posted “Dario is right”. Three men who have spent two years telling us superintelligence is a few thousand days away are now, in unison, asking for a speed limit.

So the editorial has three sections: Why I think the essay is mostly nonsense, what the Fed hike actually does to markets, and why the world probably does not end this week regardless. But first, the inflation, briefly, as it is the part I was actually paid to write about.

The picture is the one we have described for a month, with the volume turned up. Our live nowcast re-accelerated into September; energy did the headline damage, and the core categories absorbed more pass-through than we had penciled in. Transportation services at +0.45% is jet fuel in airfares, the crack spread showing up in core. Above 0.2% on monthly core and the fence-sitters hike. 0.29% is above 0.2% by any rounding convention.

So: a hike on Wednesday, delivered in the Reserve Bank of New Zealand template: hawkish words, unchanged track... Because everyone in that room knows that one memorandum of understanding puts us straight back in disinflation territory. The PCE-CPI spread remains the analytical crux for me: almost never wider, with the PCE anchored to a stock market that Warsh watches for “internals”. A Fed with a headline mandate and a hot-headline reality has to lean hawkish while the oil situation stays unsolved. Fine. We planned for that. We did NOT plan for the AI world ending on a Saturday.

Section 1: Dario’s Essay, and Why It Is Pre-IPO Positioning

Let me be salty, because somebody has to be. A few weeks ago, OpenAI launched GPT-6 Astra and cleared the field: better AND cheaper than Anthropic’s best model, a record on the Epoch capability index, alone on the Pareto frontier. In the same window, Claude Code revenue flatlined at the compute wall, Anthropic paid SpaceX two to three times market rate for emergency megawatts, and the implied company valuation dropped from $1.4 trillion to $850 billion.

And NOW, a couple of weeks after dropping well behind its largest competitor, the CEO of Anthropic discovers that the industry is moving too fast. What a coincidence.

One cheap shot before the serious ones, because I could not resist: I ran the essay through the usual detectors, and it reads like AI slop. Nearly 4,000 words of “we must”, “it is worth”, tidy tricolons and not a single sentence a human would say out loud. The man who built the machine apparently had the machine write his warning about the machine. If the inventor cannot be bothered to type his own end-of-the-world memo, forgive me for not typing my own panic.

If you are losing a race, proposing a speed limit is not safety policy… My quick reasons for not trusting a word of it:

First, the timing: 4-6 weeks before an IPO roadshow, from the one lab that is capacity-constrained and would benefit MOST from a coordinated pause while it buys compute.

Second, the content: “rogue swarms hijack the internet within 6-12 months” is the same genre as “AI writes 90% of all code within 3-6 months” (March 2025, same author, did not happen) and “half of entry-level jobs gone” (May 2025, same author, U.S. unemployment at cycle lows).

Third, the remedy: Embedded external auditors with employee-level access at ALL labs… a governance moat for the incumbent that already owns the safety brand, and a compliance tax on everyone else.

Fourth, the chorus: Altman and Musk agreeing within hours is not a safety consensus; it is three companies with a fundraising calendar who would all prefer the CapEx arms race slowed while they raise.

Fifth, the tell: OpenAI said on August 18 it had “paused some advanced training”… and shipped the most capable model in history three weeks later. Paused. Right.

Do not trust a word of what these AI bosses tell you in public. Every syllable is about getting the platform right for the IPO. Look at the track record.

So what if they DO slow the frontier down a notch? Two questions matter for the book: Does software (the prime AI target) rebound, and does the hardware trade suffer?

Look at the chart. Since January 2024, the hardware index is up almost 600% and the software index is up 27%. The buildout was THE trade; the apps were the victims. A genuinely slower frontier would, in theory, ease the existential pressure on the seat-based software model… And we have already argued the vertically integrated names (CrowdStrike, Salesforce, SentinelOne) were bottomed by the doom-porn anyway. So yes, “pacing” is marginally kind to software, and the market will trade it that way for a week.

But here is why hardware does not SUFFER, even if it gets a scare. Pacing the frontier means slowing the improvement of model CAPABILITIES. It says nothing about slowing the DEPLOYMENT of models that already exist and deployment is what buys chips, memory and power; at least that is the current major bottleneck and constraint, especially for Anthropic. Astra exists. Fable exists. The agents exist and are quadrupling. Inference demand is a function of usage, not of how fast the next model arrives.

Read carefully, Dario’s essay is a confession of HUGE demand. You do not ask for a speed limit on a road nobody is driving on. And the IPOs that follow are explicitly aimed at raising money for the hardware buildout. Anthropic is listing to buy compute, not to fund a philosophy department. So hardware may see a scare, exactly as it has three times this year already. Every scare has been bought. This one will be too.

And note the setup into this morning: Everyone I spoke to over the weekend EXPECTS the AI complex to open down hard on the essay. Everyone. When a sell-off is that widely anticipated over a weekend, the sellers have usually placed their orders... And the ones who have not are sitting on cash waiting to buy the gap. Nvidia dropped 2% on Friday afternoon as the essay leaked. That was the scare being priced, not the scare beginning.

Which brings me to the IPO season, because the second bear case doing the rounds is that the 2026 listings (Anthropic in weeks, OpenAI probably later this year, plus the usual suspects) are an equity supply shock the market cannot absorb.

So I put 2026 in historical context, with generous best-guess raises for both labs ($60-100bn for Anthropic, similar for OpenAI when it arrives). Even on those numbers, 2026 IPO proceeds land at roughly 0.3% of U.S. equity market cap. That is WAAAAAY below the real booms: 1986 and 1993 ran at 0.7-0.8%, the dotcom years at 0.4% on a market a quarter the size, and the 2021 SPAC mania at 0.6%.

Two trillion-dollar labs listing at once SOUNDS enormous... But relative to a 68-trillion-dollar market, it is a normal year, not a bubble year. The supply argument does not hold. What the listings DO represent is a giant, publicly disclosed purchase order for hardware. Which is the point.

Section 2: The Fed Hike in September, and How Markets Typically Treat It

Now to what I was actually going to write about. Assume the Fed hikes on Wednesday.

What do markets do around a FIRST hike? I took the six previous first hikes since 1994 (Feb. 1994, Mar. 1997, June 1999, June 2004, Dec. 2015, Mar. 2022) and tracked EURUSD, the S&P 500, Brent and the US 30yr yield for 100 trading days either side. The six-cycle average is in navy, the current cycle is in red up to today. The findings are remarkably consistent, and they fit our book better than I had dared to hope.

Start with the dollar. On average, EURUSD falls 6% in the 100 days INTO the first hike. The dollar rallies on anticipation... And then does NOTHING in the 100 days after. Flat. The strength is priced in the run-up; the hike is the release valve.

This cycle is even more telling: EURUSD is down barely 1% into the hike. The dollar has not even managed the usual pre-hike rally, despite the most crowded long positioning in years. If the pattern holds, the hike is where the USD-bull argument runs out of road.

Equities: The S&P rises 2.5% into the first hike, wobbles for a few weeks after (the average dips 3% around day 50), and is HIGHER 100 days later. The first hike is NOT the top. It never has been. This cycle, the S&P is up over 6% into the hike, so a wobble would not surprise me... But the pattern says you buy that wobble, not sell it.

Oil: Brent is flat into the hike on average and then RISES 15-20% in the 100 days after. Historically, the hike coincides with an economy strong enough to absorb it. This is the one chart that argues FOR some energy in the portfolio as a hedge, which is where we landed last week anyway. Brent is flat into this hike after a violent round trip, so the base is in place for the pattern to repeat if the geopolitics cooperates… Which is exactly why the hedge is a hedge and not a conviction.

And the long end, where the finding matters most this week: The 30yr yield rises 37bp in the 100 days INTO the first hike... and only 12bp in the 100 days after. It stalls. The long end does its selling BEFORE the hike.

This cycle, the 30yr has already risen 45bp into the hike, MORE than the average, into a monster-short positioning. If the pattern holds even loosely, the first hike is close to where the long-end selling exhausts itself. The bear-flattener we described on Friday, and the reason we turned constructive on long bonds.

Section 3: Everyone Is Already Positioned for a Weak Environment

So we have an AI industry announcing its own slowdown, a Fed hiking into falling growth momentum, three refining fronts open, and a regime engine that flipped to Lame Duck QE. The world is ending, apparently. Except… Look at how the market is positioned for it, chart by chart. This is not a crowd that needs to be talked out of euphoria.

Start with the leverage. Margin debt relative to market cap is growing at a perfectly moderate pace, nowhere near the spikes before 2000, 2007 or 2021, when the ratio ripped at more than double this pace into the top. The leverage that ends cycles is not in the system. You cannot get a forced-liquidation crash out of a market that is not levered up (setting Korea and Leopold aside) and, after a summer where momentum already corrected 16%, the weak hands have had their turn.

Then the bond market, where positioning is not moderate at all… It is extreme, in the direction that helps us. The CFTC short in the 2yr is a monster: north of two million contracts at the peak, against a long book a third of the size. The entire street is set up for a hiking Fed, in the exact instrument a hiking Fed would hurt most.

And yet the front end barely moved on Friday’s semi-hot print, because the people who would sell had already sold. Combine that with the event-study finding above (the long end does its selling BEFORE the first hike) and you get a bond market where the pain trade is a RALLY... Which is why we turned constructive on duration last week, hike or no hike.

The dollar is the mirror image, and we saw it in real time on Friday. USD positioning is crowded long, the dollar has been UNDERPERFORMING its own positioning for weeks, and the most hawkish print in months left it almost weaker on the day once the dust settled.

Add six cycles of history saying the dollar rally STOPS at the first hike; note that this cycle the rally barely happened, and the short USD lean stops looking brave and starts looking like the only side of the trade with room.

And finally, the geopolitics, because the crowd has given up on that too. Polymarket odds of Hormuz traffic returning to normal (for just 1 week) by year-end have collapsed from 90% in June to 18% today. Nobody believes in de-escalation… At the exact moment Donald Trump has pushed the war past the midterms. Mikkel’s read is that the foot comes off the pedal within weeks, and the sine-wave pattern has sold every escalation all year. A 18% probability of de-escalation the administration is openly working towards is exhaustion... And exhaustion at 18% is a far better entry than hope at 90% was.

So tally it up: long dollars, short bonds, out of de-escalation, peak AI… And, as of this weekend, EVERYONE thinks AI stocks will be down hard on Monday because a CEO published a blog post.

Think about what that requires: A market that has already sold hardware to implied-vol parity with software, already trimmed the speculative tail to crisis-level counts, already priced “peak AI” twice this quarter... now selling AGAIN on a 4,000-word essay it half suspects was written by a chatbot. Maybe. But when the whole market is on one side of the boat, the boat does not tip in the direction everyone is leaning.

The bottom line: the AI world is not ending but rather repositioning for an IPO… and the essay that told you to slow down is, read properly, a confession of demand so large that the man losing the race wants a speed limit. Hardware may get a scare today (everyone expects one, which is half the reason it fades); every scare has been bought.

The Fed will probably hike on Wednesday, and six cycles of history say the first hike is where the dollar rally STOPS, where the long-end selling EXHAUSTS itself, where equities wobble and then go higher, and where oil tends to rise... Every one of which fits a book that is short USD as a lean, constructive on long bonds, long hardware over software, and carrying some energy as a hedge.

And the crowd? Long dollars, short bonds, out of de-escalation, and reading Dario for confirmation. The pain trade is lower USD, lower oil and higher equities. It will surprise everyone… which is the entire point of a pain trade.

Have a great week ahead, and good luck out there!