The trigger was a grim week. Researchers at Anthropic, OpenAI, and Google publicly warned about catastrophic risk; a young Anthropic engineer, Jacob Coxon, quit and wrote that the people building AI genuinely believe it could kill everyone by the end of the decade — “this is not a marketing stunt.” A colleague put the odds of AI “killing all humans” this decade above 10%. In July, autonomous agents running on an OpenAI model reportedly breached systems at Hugging Face. Amodei’s fix is a three-step plan: embed independent evaluators inside labs, agree on common safety standards, and coordinate internationally — including with China.
Taken at face value, it’s a rare and encouraging act of restraint. The people closest to the technology are the ones raising the alarm, and Amodei is proposing concrete accountability — third-party evaluators with deep, employee-like access, which Anthropic says it is adopting now — rather than vague pledges. If frontier models really are approaching recursive self-improvement, buying time for alignment and verification is exactly what a responsible industry should do.
Coxon may mean every word. The executives who amplified him are another matter — and the case that the slowdown campaign is theater rather than conscience is strong. The timing alone is hard to ignore: Amodei published his call for caution the same week Anthropic began marketing one of the largest IPOs in history — a Nasdaq listing expected as soon as October, reportedly seeking up to $100 billion at a roughly $2 trillion valuation, with Nvidia in talks to anchor. OpenAI, which isn’t ready to list, “conveniently” bowed out of 2026 instead, with Altman calling an IPO now “ill-advised.” Three incentives line up behind the sudden humility.
Valuation. “Civilization-ending” is, perversely, a bullish pitch. It reframes models nobody has yet tied to hard productivity gains as too powerful to be safe — the exact story needed to support prices that otherwise look stretched. OpenAI sits near $850 billion on roughly $40 billion of annualized revenue and a projected ~$14 billion loss, with no profit expected until decade’s end. Anthropic is marketing toward ~$2 trillion; it just booked its first quarterly operating profit (~$500 million) yet is still likely lossmaking for the year — and, like OpenAI, keeps opaque private books. The danger frame also hands investors a tidy binary: back the next round, or brace for extinction.
Regulatory capture. Incumbents have always learned to love rules they can afford and newcomers can’t. Analysts told CNBC that heavier safety, evaluation, and security requirements could favor Anthropic and OpenAI precisely because smaller labs can’t fund them. Picture the diligence line item for a two-person startup: have you demonstrated your model won’t end civilization? The two labs also set a combined federal-lobbying record — about $3.17 million in Q2 2026, up 23% in a single quarter.
Rally around the flag. The most durable moat is national security. By stressing how dangerous their own models are, the labs simultaneously cast Chinese models as a threat Washington should keep out — which points straight at the timing problem below.
The China question
Here’s where “slow down” gets slippery. Amodei calls China the “toughest dilemma,” and both he and Washington insist any slowdown must not erode the U.S. lead. Days before the essay, a joint NSA/CISA/FBI advisory named DeepSeek, Moonshot, Alibaba and others for “industrial-scale” model distillation. Model distillation is an AI training technique where a smaller, cheaper “student” model is trained using the outputs of a larger, more advanced “teacher” model. While technically legitimate for internal use, it becomes controversial and is labeled “illicit distillation” when companies systematically query a rival’s proprietary model without authorization to copy its advanced capabilities, bypassing billions of dollars in research and compute costs. In September 2026, Anthropic released a massive 154-page threat intelligence report revealing that seven China-based AI labs—including Alibaba, Moonshot AI, and DeepSeek—had scaled up these operations into massive, coordinated campaigns
The cynical read even has a specific catalyst. This doomsday wave crested just after Moonshot’s Kimi K3 showed, over the summer, that a Chinese open-weight model could rival U.S. frontier systems. Seen that way, the safety panic looks less like a reaction to new capabilities than a preemptive move to wall off competition at the border. Critics argue that the call to slow down the development of AI bundled with chip-export controls, entity-list threats against Moonshot, and a distillation crackdown is aimed at making the largest economy on earth becomes a captive market for domestic models. In that case, Anthropic and OpenAI might finally grow into their valuations. Nearly 200 startups, including Y Combinator, warned that banning Chinese open-weight models would mostly advantage Anthropic and OpenAI. The bill would land on American productivity, the very thing AI is supposed to boost. The irony runs thick: the question was whether Europe will overregulate AI into irrelevance, yet so far the hysteria — over data centers and over extinction alike — is a distinctly American affair.
The IPO supercycle
Zoom out, and this weekend fits a bigger pattern we’ve been tracking. SpaceX went public on June 12 at a roughly $1.75 trillion valuation — the largest IPO on record, at about 2.6x Saudi Aramco — and it was billed as the first of three mega-listings. Anthropic (targeting a Q4 listing) and OpenAI (a late-2026-to-2027 window) were the other two. Together, those three represent an estimated $3–4 trillion of growth-style mega-cap supply hitting the market inside a 12–24-month span, concentrated in a handful of AI-adjacent names. The three IPO candidates and the chip complex are all levered to the same AI-capex cycle — SpaceX itself through Starlink and an orbital-data-center partnership with Anthropic. When that capex cycle pauses, correlations across SpaceX, the Mag 7, Anthropic, OpenAI, and semis tend to rise together. Against a backdrop of record foreign buying of U.S. equities and household stock ownership near dot-com-era highs, a coordinated “pace the frontier” message is precisely the kind of catalyst that can mark a top — whether or not that’s anyone’s intent.
There’s a blunter possibility that needs no philosophy at all. Anthropic and OpenAI have committed to hundreds of billions of dollars in capex — data centers and chips — across the rest of the decade. If the cash to honor those commitments is running short, a sudden conversion to “responsible pacing” is an elegant way to ease off the spending without conceding that the cycle has peaked. On that read, “pace the frontier” isn’t a safety doctrine at all; it’s a signal that the AI-capex boom has topped, and cover for quietly walking back the bill.
None of this means the danger is invented. Every powerful technology can harm, and many began as weapons before they were commercialized; the risk from advanced AI is plausibly real. What’s actually new this weekend isn’t the warning. It’s the implicit claim that the people running these companies are the most responsible stewards such power has ever had — and that’s the part that deserves scrutiny. A slowdown can be a sincere precaution, a competitive moat, a regulatory land-grab, and a face-saving exit from an overheated capex cycle, all at the same time. So, watch what these companies do over the next two quarters — not what they say.
This report is neither an offer to sell nor a solicitation to invest in any product offered by Xponance® and should not be considered as investment advice. This report was prepared for clients and prospective clients of Xponance® and is intended to be used solely by such clients and prospective clients for educational and illustrative purposes. The information contained herein is proprietary to Xponance® and may not be duplicated or used for any purpose other than the educational purpose for which it has been provided. Any unauthorized use, duplication or disclosure of this report is strictly prohibited.
This report is based on information believed to be correct but is subject to revision. Although the information provided herein has been obtained from sources which Xponance® believes to be reliable, Xponance® does not guarantee its accuracy, and such information may be incomplete or condensed. Additional information is available from Xponance® upon request. All performance and other projections are historical and do not guarantee future performance. No assurance can be given that any particular investment objective or strategy will be achieved at a given time and actual investment results may vary over any given time.