The State of the AI Debate

September 21, 2026 · Episode Links & Takeaways

HEADLINES

Anthropic Pushes Its IPO to November

Anthropic won't go public this month or next, but sources say the November timing was set before Dario Amodei called for an AI slowdown and is about getting strong Q3 numbers into the prospectus. Testing-the-waters meetings leaked that Anthropic went from spending $2.30 per dollar of revenue in Q2 2025 to slightly profitable on that basis this year, and the company says it's on track for $100B in annualized revenue by year end, up from $65B in July. Critics have hammered the adjusted framing, and the disclosure documents will give investors their first look at unadjusted numbers. Former Nasdaq executive Karen Snow says the IPO "went from being a no-brainer to not being a no-brainer." With new Sonnet, Opus and Fable models reportedly in stealth testing, a model release could be the play to shift the narrative from safety back to progress.

Anthropic Quietly Opens a Wet Lab

Reuters reports Anthropic has set up its own biology lab in the Bay Area, focused on fundamental biology rather than drug discovery, though its biosafety rating is unknown. The safety irony drew immediate fire, but the more interesting question is what it means for lab business models, with Nic Carter floating the cynical read that "pacing the frontier" is cover for internalizing breakthroughs rather than selling them as tokens. Those are two separable ideas: whether pacing is the vehicle is one question, but whether labs are hedging against commoditizing token prices by using early frontier access to do the work themselves is worth considering on its own. Plenty of non-cynics simply see a lab excited about medical use cases taking an obvious step.

First Cracks in Data Center Debt

Bonds tied to a data center leased by Jane Street, issued in August at 8.9%, are now trading at 11.3%, a move the Fed can't explain. The caveats matter: the rate is fixed so existing costs don't change, and this is the junk market rather than investment grade, which makes it a sign of funding slowing at the margins. Meta's first junk bond via CleanSpark still drew $10B in demand for $2.3B at 8.25%, while $18B of Oracle-linked debt in New Mexico has been quoted as low as 89 cents on the dollar. For anyone watching for the AI trade rolling over, this corner of the market, not stock prices, is the one to watch.

MAIN STORY

The State of the AI Debate

AI regulation is becoming an issue with real political resonance, and much of the US debate rests on assumptions about what China will or won't do. With Xi Jinping in Washington this week, there's a rare chance to hear directly where China stands, and the early signals suggest the two superpowers are worried about very different things.

WILL XI SLOW DOWN?

AI Force
"I am forming the AI Force, much like I did Space Force"
Trump announced a new military arm, a coming AI Czar and a public poll to rename the field "Superior," "Extreme" or "Supreme" Intelligence, drawing aghast reactions from left and right alike. Strip away the noise, though, and Space Force did consolidate satellite operations scattered across branches; a central body supporting AI-era cyber defense might not be a crazy notion.

Kill Switches
"We're not waiting to act"
Newsom's executive order fast-tracks kill switch policy and third-party safety inspectors, and as a likely 2028 contender who has vetoed AI safety bills before, he's a useful bellwether for Democratic AI politics. The popularity is easy to understand; how much a kill switch does for here-and-now concerns is a different question.

Virginia
Not a moratorium, but some rules could work like one
Spanberger's order bans NDAs for officials, expedites noise rules and ends the by-right approvals that built Data Center Alley. Builders are unhappy and opponents say it doesn't go far enough, which in a democracy can signal a policy has found the middle ground.

Ro Khanna Calls for China Slowdown
Asking Alibaba, DeepSeek and Moonshot to pace the frontier
If China won't join a slowdown, US pause proposals risk being DOA and simply cutting America out of the race. Khanna's letters seek a binding international agreement, with an emergency China Committee hearing set for Wednesday ahead of Xi's visit.

Xi's Visit
An AI version of the Cold War red phone
Xi is in the US September 23 to 25, and while preserving the trade truce is the immediate priority, AI was central to prep talks where Scott Bessent proposed an incident notification mechanism with Vice Premier He Lifeng. Don't underestimate the low-hanging fruit: "moving from opaque to more transparent between the number one and number two AI powers in the world is very important."

How China Sees AI Risk
"AI safety not as a shared humanitarian mission"
China's AI community professes almost no fear of existential risk; the existential worry is to party rule, per Minister of State Security Chen Yixin, and law professor Henry Gao says Beijing views safety "through an adversarial lens." The Foreign Ministry dismissed US slowdown talk as "fearmongering."

Anthropic and China
Espionage terms matter more to Beijing than x-risk
A Claude Code backdoor that tracked Chinese IPs prompted an Alibaba ban, Anthropic disclosed Chinese officials uploading military documents to Claude, and a state TV affiliate now claims Anthropic shares data with US intelligence.

China's AI Economy
"Running too cold" for a tech boom to lift
AI capex has kept the US out of recession, but China's economy shows no AI-driven recovery, with youth unemployment at 18.9%, car sales down 20% and housing down another 14%. Chinese officials are closer to managing a crisis than carving up the spoils of an AI boom.

Mustafa Suleyman
"I don't think we should use China as the bogeyman"
Microsoft's AI chief argues each lab owns its safety in a pluralist system, and that regulation creates "shared norms and standards to increase safety. It shouldn't have to slow us down." Whether that view spreads remains to be seen, but this week will move the conversation one way or another.