Monday discovered the brake pedal is connected to the stock market.

Wall Street has discovered that AI safety has a price-to-earnings ratio

AI-linked stocks sold off Monday after Anthropic CEO Dario Amodei called for slower frontier-model development while OpenAI’s Sam Altman and xAI’s Elon Musk expressed support. Nasdaq 100 futures were down roughly 1.6% early Monday, while Nvidia fell more than 2% in premarket trading. AMD, Intel and Marvell were also sharply lower.

The reaction is interesting because nothing happened to the technology over the weekend: no important model failed, no data center disappeared, and no GPU became slower. What changed was investors’ estimate of how aggressively the industry might pursue the next generation. That is a milestone for AI safety, which has long looked like an externality to researchers and governments while capital markets rewarded acceleration.

Monday shows those worlds are connected. If laboratories genuinely slow frontier development, demand assumptions for chips, data centers, electricity and debt-financed infrastructure may need adjustment. That does not mean the AI bubble popped this morning. It means investors have found another variable in the equation: more capable model = more valuable infrastructure, unless we intentionally wait.

China heard “slow down” and immediately asked who gets to remain in front

The proposed frontier slowdown has already acquired a geopolitical problem. China’s state-backed Global Times attacked Amodei’s proposal Monday as resembling a “Cold War” strategy intended to preserve American technological dominance rather than simply improve AI safety. The criticism is understandable in context: Amodei’s proposal sits alongside calls for tighter restrictions on China’s access to advanced AI hardware and U.S. model capabilities. From Beijing, the message can sound like America should keep its lead, restrict China’s ability to catch up, and then ask everyone to slow down.

China’s foreign ministry says international cooperation should avoid confrontational technological blocs. Yet China is not dismissing loss-of-control risk. Reuters reports that Chinese policymakers and researchers are developing approaches to increasingly autonomous systems, including work on models acting beyond human control. The disagreement is less about whether AI risk is real than who decides what safe development looks like, and whether “safety” preserves somebody else’s lead.

Technical coordination among three American frontier labs is difficult enough. Global coordination requires countries to trust that a safety regime is not industrial policy wearing glasses. The United States and China are expected to discuss AI safety in upcoming talks; we may be discovering that alignment is not only a problem for the models.

Reuters — China’s safety work ↗

The $400 million chip machine has apparently won the argument

For years, semiconductor manufacturers faced an uncomfortable question about ASML’s newest lithography technology: is High-NA EUV worth that much money? The answer increasingly appears to be yes. ASML’s High-NA systems cost roughly $400 million each, about twice the price of existing EUV machines. Their larger numerical aperture can print features approximately 40% smaller, potentially extending leading-edge chipmaking well into the next decade.

Intel moved first. Samsung and SK Hynix are expected to begin adopting the equipment from 2028, and TSMC — whose earlier caution was a major question for High-NA — now plans to introduce it around 2030. That gives ASML an extraordinary position: the company held roughly 94% of the lithography market in 2025 and remains the only commercial supplier of EUV systems.

There are few more literal technological bottlenecks. If the next decade’s AI roadmap depends on denser chips, and those chips depend on machines only one company currently knows how to build, a surprisingly large part of the future passes through a Dutch factory. Software likes to pretend scale is infinitely replicable. Semiconductor manufacturing replies: sure. First buy the $400 million machine.