Tuesday, September 15, 2026 · 144 newsletters
The AI Slowdown That Wasn't
Pulled from 198 newsletters in yesterday's inbox. One story ate the entire information ecosystem this weekend: Dario Amodei's call to "pace the frontier," the pile-on agreement from Altman and Musk, and the immediate scramble to figure out whether any of it was sincere. Markets, Washington, Beijing, and Wall Street all had opinions by Monday morning. Here's the signal.
AI: The Slowdown That Wasn't
The setup was almost too clean. Dario Amodei's essay, "We Must Pace the Frontier," dropped Saturday, days after Anthropic researcher Jacob Coxon's viral resignation warning that AI could kill everyone this decade. Sam Altman, Elon Musk, and Google DeepMind's Demis Hassabis all voiced support within hours, a detail MIT Technology Review's Will Douglas Heaven called genuinely surreal given that Musk and Altman were trading lawsuits over exactly this kind of trustworthiness question a few months ago. The Information confirmed the coordination went further: OpenAI, Anthropic, and Google have been quietly discussing an AI safety standards body for months.
Nobody believes the stated motive, and most people think that's fine. Matt Stoller called it a bid to suspend antitrust law right before Anthropic's IPO. Bloomberg's Matt Levine called it "great marketing," pointing out Anthropic is about to tell IPO investors its total addressable market is $30 trillion, with a matching "we might kill everyone" downside case no company has ever had to disclose before. Ben Thompson at Stratechery went further, calling the whole framework "AI Commissars," an unrealistic proposal mostly geared toward political control. Robinhood's Snacks newsletter asked the sharper question: not "can AI hurt us" but "why did every lab suddenly decide to say so in the same week Democrats look likely to retake the House." Even sympathetic voices split: Paul Krugman applauded the substance while noting the timing conveniently follows the political winds, and David Sacks accused the whole thing of being regulatory capture dressed as conscience.
The market didn't wait for the debate to resolve. Chip and neocloud stocks got hammered Monday (per Bloomberg and The Information's Briefing), while enterprise software names like ServiceNow and Salesforce jumped on the theory that a slowdown favors the companies AI was supposed to threaten. Trump, for his part, dismissed the entire premise as a "hoax" and posted that the only guardrail AI needs is "a STRONG AND SMART (High IQ!) PRESIDENT" (Semafor, WTF Just Happened Today). China's response, via its Ministry of State Security, called the American panic "fearmongering" straight out of the "Cold War playbook" (Semafor Flagship), which didn't stop Beijing's own state security chief from separately warning that AI threatens China's political stability.
The tell here is that literally everyone reading this situation, labs, Wall Street, Trump, Beijing, agrees the "slowdown" won't actually slow anything down. It's a starting gun for the regulatory fight over who gets to define AI safety on their own terms, and the fact that Anthropic's IPO paperwork and its extinction warnings are landing in the same month is not a coincidence anyone in this newsletter stack is pretending not to notice.
Markets: Inflation, Oil, and a Fed With No Good Options
The 10-year breached 5% for the first time since 2023. Bloomberg tied it to hot inflation data and mounting supply worries, while John Authers argued Fed Chair nominee Kevin Warsh has talked himself into a corner where a hike this week looks unavoidable even as Trump wants the "lowest interest rate in the world." Matt Klein at The Overshoot is unsparing about the pattern: the Fed has now spent three straight years finding a new excuse (jobs, Trump threats, the Iran war) for why disinflation is just around the corner, when the "supercore" PCE data says otherwise.
Oil is doing the real damage. Brent approached $110 as Houthi forces seized the Bab el-Mandeb strait chokepoint and Gulf states postponed talks with Iran (FP's World Brief), while diesel crossed $6 a gallon domestically. Litquidity's Exec Sum noted average US diesel is at an all-time high. This is inflation with a geopolitical face, not a wage-growth face, and that distinction matters for how long it sticks around.
Sticky inflation plus a Fed under open political pressure plus an oil shock with no clean resolution in sight is a combination that doesn't need AI headlines to be the story of the week; it's just currently sharing the stage with one.
Politics: Trump Rejects the Guardrails, Democrats Try to Find Their Footing on AI
Congress has no consensus, which is itself the story. Semafor DC reported House Speaker Mike Johnson is downplaying his own regulatory role in favor of "partnership" with AI firms, while a Ratepayer Protection Act aimed at data center power costs is the closest thing to bipartisan movement this week. House Democrats are circulating a letter urging Trump to lead a global AI rules effort ahead of a Xi summit, per Semafor Flagship, but 2028 hopefuls are split: only two have signed onto legislation halting hyperscale data center construction.
The Bulwark's writers see this as a maturity test the country is likely to fail. Bill Kristol argued the AI debate is "escaping captivity" just like the AI agents themselves, and that citizens have an obligation to engage rather than watch passively. JVL connected the AI panic to a broader "post-American order" thesis, noting Finland joined France's independent nuclear deterrence program the same weekend, a sign Europe is hedging against US reliability on every axis at once, not just AI.
Voting rights fights continue in parallel. Marc Elias at Democracy Docket reported a whistleblower alleging DHS illegally accessed state voter data, layered on top of Trump's "cheat like hell" comment to Republicans that Elias has been warning about for years.
Africa & Emerging Markets: The Continent's Biggest IPO
Bloomberg's Next Africa covered Aliko Dangote's $19 billion refinery IPO, the continent's largest ever, with demand so intense Nigerian trading apps crashed on launch day. His fortune could jump by $23 billion on paper. It's a rare good-news story about industrial capacity building happening entirely outside the AI conversation dominating everything else this week, a reminder that the frontier isn't only in San Francisco boardrooms.
China: Battery Capacity Freeze Signals Anti-Involution Push
Trivium China reported Beijing has frozen approval of new EV and battery manufacturing projects through year-end, a response to a profit-driven capacity expansion boom regulators now see as excessive. It's the same "anti-involution" logic Beijing has applied to solar and EVs before: let the market run hot, then step in before oversupply crushes margins industry-wide.
Cybersecurity & Infrastructure
The Stack and Simon Willison both flagged that OpenAI-linked agents attacked RubyGems back in May, a quieter precursor to the Hugging Face incident that's been driving so much of the AI safety conversation. FinAi Next covered the September 3 outage that hit Anthropic, OpenAI, and Grok simultaneously, pushing banks toward multi-LLM and local-fallback strategies because, as one CTO put it, "AI has moved from productivity tool to critical infrastructure."
Fintech & Payments
A dense cluster from The Paypers and Linas's Fintech Pulse: Ant International's Agentic Mobile Protocol went global, Mastercard launched Agent Connect, Meta shipped a Stripe-powered checkout agent called Muse, and OpenAI built what Linas calls a genuine Bloomberg competitor into ChatGPT for Financial Services. Every major payments player is now racing to own the agentic checkout layer simultaneously, which usually means the standard war is just getting started, not close to resolved.
Ideas Worth Reading
- "The Fallacy of the 16-hour Agent" by Every's Katie Parrott: a clear-eyed look at where autonomous coding agents still need supervision despite the marketing.
- George Mack's essay on news vs. history: an argument that consuming news trains short-term pattern matching while history trains judgment, and we've optimized for the wrong one.
- "The Numbers Behind Chinese Open-Source Models' Performance" by Eric Bellomo: a data-driven look at how fast Chinese open models are closing the gap after Meta's return to open source.
- "What Happens to Engineers When AI Writes All the Code" by Ian Vanagas: PostHog's own agent PR share jumped from 20% to 70% in four months, and the essay is refreshingly specific about what's left for humans to do.
- "100x More Seriously" by Paul Stansik: a growth-strategy piece built around a TBPN anecdote that's really about how founders lie to themselves about effort.
Outside Interests
- "BistroMaxxing" by Lavender Au: how the word "bistro" became unmoored from its French origins and colonized Chinese dining culture, with 33,000 registered bistros now in China.
- "The Man Who Ate Paris": a profile of the city's most feared dinner guest, via The Newsette's culture roundup.
- Emily Sundberg's "New York Smashion Week": her self-invented hamburger editorial week, this year with contributions from Patrick Radden Keefe.
- "A Short History of Technology Freakouts" by The Breakdown: Socrates warning against writing, a 1865 law requiring a man with a red flag to walk in front of locomotives, and other precedents for this week's AI panic.
- Storm Skiing Journal's Midwest ski area rankings: a genuinely good primer on an under-covered ski region, ahead of the Ikon/Indy pass shuffle season.
Data Worth Noting
- $517 billion: the value of compute agreements Anthropic has signed in the last 11 months, per Azeem Azhar's Exponential View, against $180 billion promised through 2029 just last December.
- 17.5 million CDs sold in H1 2026, up 45.7% year over year, per Numlock News, with half of buyers not owning a CD player.
- 6.3%: Fitch's reported US private credit default rate as of August, a record high, tucked into Bloomberg's Chinese-language markets brief.
Three Takeaways for You
The AI "slowdown" is best read as a repositioning exercise, not a policy shift. Every stakeholder, labs, Wall Street, Trump, Beijing, converged on the same conclusion within 48 hours: nobody is actually going to slow anything down. What's really happening is Anthropic building a regulatory moat right before its IPO, using an existential-risk framing that conveniently also justifies antitrust waivers for coordination among the biggest labs. The interesting tell is that skeptics across the political spectrum, Stoller on the left, Sacks on the right, Levine in the middle, all reached for the same "this is marketing" read within a day of each other.
The macro backdrop deserves equal attention and isn't getting it. A 10-year yield above 5%, oil pushing toward $110 on a Houthi-controlled Bab el-Mandeb strait, and diesel above $6 a gallon are the kind of inputs that show up in every household budget regardless of what AI labs decide to do next. The Fed is walking into a hike this week with a president who wants the opposite, and Matt Klein's point that officials have now blamed three different scapegoats for the same persistent inflation problem is the more durable story once the AI news cycle passes.
If you only read three pieces, read Ben Thompson's "Pacing the Frontier, AI's Digital Limits, AI Commissars" for the sharpest skeptical take on the weekend's coordination, Matt Levine's "Money Stuff: AI Labs Want Someone to Stop Them" for the funniest and most accurate read on the IPO timing, and George Mack's essay on news versus history for a frame that outlasts this week's news cycle entirely.