Good morning.

Masayoshi Son has had better timing. SoftBank spent last week sealing a two-year facility to fund its OpenAI stake, upsizing it to $11.87 billion from a $10 billion target with commitments from about 20 banks, Bloomberg reported. Then the weekend happened, and the two labs at the center of the AI trade suggested the industry ought to ease off the gas. SoftBank fell as much as 13.2% in Tokyo on Monday.

Borrowing $11.9 billion on Friday to buy the future, then watching the future request a later delivery date on Saturday, is the kind of week that makes a man want to unplug the router and go for a walk.

Artificial Intelligence

The AI Labs Asked for a Speed Limit. Markets Priced One In Overnight.

Wall Street spent three years paying up for acceleration. Over the weekend, the drivers asked for a slower lane.

Anthropic CEO Dario Amodei published a 3,800-word essay on Saturday arguing that frontier labs should deliberately slow how fast model capabilities improve, proposing that companies open up to third-party evaluators, agree on common safety standards, and that democratic governments coordinate with authoritarian ones "to the extent this is possible." OpenAI's Sam Altman agreed on X. So did Elon Musk, who runs xAI. By the Asian open, an equity trade built entirely on compounding capability was getting re-rated.

The tell is where the selling landed. This was not a broad risk-off morning. It was surgical, and it hit precisely the companies whose forward earnings assume the next model arrives soon and arrives better.

"We must slow the pace at which we improve the capabilities of A.I. models. Progress will still seem fast, and we must make wise use of the time we gain."

— Dario Amodei, CEO, Anthropic, in his weekend essay

Braking Distance

Memory and chip-supply names took the worst of it, which makes sense: they are the purest expression of the "more compute, faster" thesis, and they have the least cushion if the build-out pauses. Anthropic and OpenAI are also, awkwardly, among the largest customers underwriting that spending. OpenAI alone told investors in February it was targeting roughly $600 billion of total compute spend by 2030.

By the numbers, per Reuters: Kioxia fell 9.8% at the open in Tokyo and Tokyo Electron 3.7%; SK Hynix slid 5.3% and Samsung Electronics 3.7% in Seoul; TSMC (TSM) gave up 1.2% in Taipei; and Z.ai dropped as much as 10.5% in Hong Kong after a discounted share placement. Stateside, Yahoo Finance clocked Nasdaq-100 futures (NQ) down 1.6% and the VIX up 12.6% to 17.83 before the bell, with Dow futures off just 0.2%.

Not everyone is braking. "Their valuations assume both strong demand and a relentless pace of technological progress," Charu Chanana, chief investment strategist at Saxo Bank in Singapore, told Reuters. "When expectations are this high, even a possible delay can trigger profit-taking." Michael Burry, of The Big Short fame, was blunter, calling the warnings "hype and puffery" in a post on X.

Governor Installed: The uncomfortable part for the bulls is that this repricing did not need a regulator, a bad print, or a broken model. It needed two founders to say out loud that they might go slower, and every listed proxy for the AI build-out decided that was a different business. Whatever the labs actually do next, the Street has now learned how much of the AI premium was riding on velocity alone. That number is no longer a mystery, and it can't be un-learned.

Kalshi News

Kalshi News

Markets move before the news.

Capital Markets

OpenAI's IPO Slips to 2027. Anthropic Is Still Walking Toward the Door.

Two labs, one weekend, and suddenly only one of them is in a hurry to ring a bell.

Altman told Fortune that "right now would be an ill-advised moment to go public," confirming OpenAI will not list this year; finance chief Sarah Friar had already told staff last month that the lab "will be a public company in 2027." Anthropic, meanwhile, is still moving. It confidentially filed its prospectus in June, has picked the Nasdaq, and has been widely expected to list as soon as next month.

So the same argument produced opposite calendars. That is the genuinely interesting part, because both companies confidentially filed, both endorsed the slowdown, and only one decided the slowdown was a reason to wait.

Window Shopping

Anthropic's case for walking through anyway is arithmetic. The company was valued at $965 billion earlier this year and could seek a $2 trillion valuation in a listing, per CNBC, which also reported the lab hit $65 billion in annualized revenue in July, roughly a sevenfold jump from a year earlier. The Financial Times reported Sunday that Anthropic has told some shareholders it will book an operating profit for a second straight quarter. Menlo Ventures partner and Anthropic investor Matt Murphy called the growth "off the charts" and said he doesn't "see why growth would slow or any other reason to wait." The comparison everyone reaches for is SpaceX, which went public in June in the biggest IPO on record and now carries a $2 trillion valuation of its own.

The skeptics aren't arguing with the numbers so much as the motive. Gil Luria, an equity analyst at D.A. Davidson, noted that stricter standards conveniently favor the two labs that can most easily afford them, an argument Gartner's Arun Chandrasekaran echoed. OpenAI has reportedly asked members of Congress whether an industry-wide slowdown would run into antitrust law, according to Wired.

"It feels more and more like a ladder pull."

— Gil Luria, equity analyst, D.A. Davidson, to CNBC

Threshold Question: Public markets have already shown they'll underwrite enormous, unproven AI valuations. What they have not been asked to price is a company that has publicly committed to growing more slowly than it could. Anthropic is about to run that experiment with its own book, and every private AI mark behind it gets repriced off the result. OpenAI, by waiting, gets to watch someone else find the floor.

Economics

The Fed Meets Wednesday With Gasoline on Its Hands

Kevin Warsh wanted a clean inflation print. He got one with a fuel surcharge.

August CPI landed Friday at 0.4% on the month and 3.4% over 12 months, both in line with the Dow Jones consensus, the Bureau of Labor Statistics reported. Core was the problem: 0.3% on the month, a tenth above forecast. Traders responded by pushing odds of a quarter-point hike this week to nearly 90% on CME's FedWatch, and by Monday morning the market was still holding around 86%.

Then oil kept climbing. The FOMC concludes Wednesday with the fed funds rate parked at 3.5%-3.75%, where it has sat all year, and a chair who has said flatly that if the numbers don't improve, "we have work to do."

Premium Unleaded

The energy complex is doing most of the damage, and it is doing it in the components households notice first:

  • Gasoline prices rose 3.9% in August, accounting for more than a third of the entire index gain; the broad energy index was up 16.3% year over year, with fuel oil up 52%.

  • Brent traded near $108 a barrel Monday after Saudi Arabia shut a key regional pipeline, while October WTI (CL) added 3.2% to $103.29, per Yahoo Finance.

  • US diesel hit a record $6 a gallon on Friday, a number that Bloomberg Money's Nikki Waller noted eventually shows up in the price of everything that arrives at a grocery store on a truck.

"Chair Warsh and others signaled that interest rates can remain on hold only if disinflation continues and today's August report did not deliver that," Kathy Bostjancic, chief economist at Nationwide, said after the release. Her firm now expects a quarter-point hike.

Hiking into an oil shock is the least popular trade in central banking, because the thing you're tightening against is the thing you can't tighten away. Warsh appears to have decided that letting expectations drift is the worse of the two mistakes.

— AllThingsWallSt, our take

Pump Fake: A supply-driven inflation problem does not respond to rate hikes the way a demand-driven one does, which is the central bank's whole dilemma this week. Raise, and you slow an economy that isn't the source of the pressure. Hold, and you invite the argument that the 2% target is decorative. Wednesday tells you which risk this Fed is more afraid of, and the long end will grade it immediately.

The Tape

Beijing to Differ: China's state-backed Global Times called Amodei's essay a "Cold War playbook" aimed at curbing the country's technological development, even as Washington and Beijing are expected to hold AI safety talks as part of bilateral discussions this month.

Flight to Somewhere: Bitcoin (BTC) rose 1.1% to about $77,654 on Monday morning while gold slipped 1.3% to $4,353.80 an ounce, an odd split for a session that otherwise smelled like a scramble for cover, per Yahoo Finance.

Trust Fall: More than half of Americans now say they're more concerned than excited about AI in daily life, up from 37% in 2021, according to Pew Research Center data cited by CNBC.

Extra Upside

Thirty-Eight Gigawatts: Microsoft plans to more than triple its data-center fleet to as much as 38 gigawatts from roughly 12 today, Bloomberg's Brody Ford reported on Bloomberg Tech; Microsoft told Bloomberg the numbers "were not accurate."

Creative Differences: Adobe's backlog shrank quarter over quarter, which BNP Paribas global head of software research Stefan Slowinski called very rare for a third quarter, adding that the cracks he had expected for three years are "finally starting to come through."

Doomer Scrolling: Burry's dismissal wasn't a one-liner; he framed the labs' safety warnings as "cover for real uncontrollable slowing growth," per Reuters, which is a considerably more cynical read than the market's.

Just For Fun

Micron's Taiwan staff are getting rewards worth up to 68 months of pay after a record year, and the unions still think the offer is light.

One in five people surveyed by Bank of America called sports betting a form of investing — the same report found roughly 75 cents comes back for every dollar wagered on online betting platforms.

After the Bell

Three years of AI trading came down to one assumption nobody wrote in a model: that next quarter's model would be better than this quarter's. It took two blog posts and a weekend to find out what that assumption was worth. Somewhere, a risk manager is updating a spreadsheet cell labeled "pace of progress" that has never once been filled in.

That's the tape. We'll see you at the open. — AllThingsWallSt

Keep Reading