Asia Put a Price on Pacing the Frontier

Monday's Asian session was the first time anyone had to put a number on the idea that the AI industry might deliberately slow itself down. Over the weekend Anthropic chief executive Dario Amodei published "We Must Pace the Frontier," arguing that the rate at which models improve has outrun the industry's ability to understand and control them. Sam Altman endorsed it and said OpenAI would not go public in 2026. Elon Musk agreed. By Monday's close in Seoul the Kospi was down 3.26 per cent and SoftBank Group — OpenAI's largest outside backer — had fallen 11.2 per cent in Tokyo.

That is a quick trip from essay to balance sheet, and the shape of the move says more than the size.

What moved

What sold off hardest was not AI software. It was memory, storage and the optical-interconnect chain — companies whose order books are a direct function of how much frontier training happens and how soon. Market recaps put Taiwan's Largan down around 10 per cent and Zhongji Innolight down roughly 4 per cent on the same session. Dan Baker of Morningstar framed the logic plainly: AI development "may be slowed by regulators... to avoid the worst case outcomes that Anthropic and Open AI have discussed."

In other words, the market did not price a safety event. It priced a capex event.

The pact is thinner than the selloff implies

What was actually committed to over the weekend is modest, and mostly unilateral. Amodei's post sets out three steps: embedded third-party evaluators with employee-like access inside frontier labs; common standards and limits on the rate of progress agreed among companies in democracies, with government help to clear the legal obstacles; and eventually verification arrangements extending to China.

"We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain." — Dario Amodei, "We Must Pace the Frontier"

Only the first step is something one company can do by itself, and it is a transparency measure, not a capability cap. Altman committed OpenAI to it — "independent evaluators with employee-like access" — and paired it with the IPO delay, saying that "given everything happening with safety, right now would be an ill-advised moment to go public." On Monday Microsoft published a code of conduct for its first-party MAI models and opened it to public consultation, with Satya Nadella writing that any pursuit of superintelligence must be grounded in the principle that AI "not helping humanity and under human control" is "not worth pursuing," and that alignment "cannot be controlled by a handful of entities."

Nobody has announced a compute cap, a shelved model, or a cancelled data centre. The second step, by Amodei's own description, requires governments to solve a coordination problem that is really an antitrust problem. The third depends on Beijing, which has offered nothing.

The tape is not a clean read

Monday had two other reasons to fall. Brent crude rose 3.2 per cent to $107.94 after a Saudi pipeline shutdown following Houthi attacks, and US core CPI came in at 0.3 per cent against a 0.2 per cent forecast, pushing futures markets to price a rate rise this week. KB Securities analyst Lim Jung-eun expected "macro sensitivity in the stock market" to rise as investors digest the Federal Reserve and Bank of Japan meetings.

So attribution is genuinely muddy. What oil and rates do not explain is the dispersion: Hong Kong and Shanghai closed higher on the same day Seoul fell more than three per cent, and the worst single performers were the names most levered to AI training demand rather than to energy or duration.

Listed Chinese AI companies were not spared inside that split — one market recap has Z.ai down about 10.5 per cent and MiniMax down 5.4 per cent — even though no Chinese lab is party to anything Amodei proposed. That is the clearest evidence that this was a trade about compute demand rather than about who has signed what.

The objection with teeth

The sharpest response did not come from an investor. Writing on 13 September, Cohere chief executive Aidan Gomez called the proposal "a sheep in wolf's clothing, a cartel by any other name," and argued that thresholds defined by compute and infrastructure are ones only incumbents can meet.

"A mechanism that slows everyone down while explicitly preserving existing commercial advantage does not make AI safer." — Aidan Gomez, Cohere

That argument is not disposed of by good intentions. Amodei's second step asks governments to make inter-company coordination on the rate of progress legally safe. Whatever else that is, it is a request for permission to agree with competitors about output.

What to watch

Three things will separate rhetoric from commitment. Whether the embedded evaluators arrive with names, scope and published findings, or remain a principle. Whether Microsoft's consultation produces binding text or a document. And whether any lab's next release actually slips. Until one of those happens, Monday's selloff was the market pricing a promise — and promises about capex are the ones it reprices fastest.

Sources: Dario Amodei, "We Must Pace the Frontier" · AP via ABC News: Asian shares mixed, SoftBank plunges after calls to slow AI · Seoul Economic Daily: KOSPI tumbles 3.3% on rate, oil and AI slowdown fears · Nikkei Asia: Tech stocks slump in Asia, driven by AI safety concerns · OfficeChai: Asian AI-related stocks drop as much as 13% · Aidan Gomez, Cohere: Who gets to define the rules for AI · Unite.AI: Nadella announces public consultation on Microsoft's MAI model rules · The Tribune: Altman indicates pause in OpenAI IPO plans, backs call for increased safeguards · CNBC: Anthropic's Amodei proposes plan to slow the pace of advancing AI capabilities