SoftBank drops 13 percent after lab chiefs ask to slow the AI frontier
Dario Amodei's essay 'We Must Pace the Frontier' drew public support from Sam Altman, Demis Hassabis and Elon Musk. Altman said OpenAI will not list in 2026. SK Hynix fell 6.3 percent.

Tokyo3 min read
Last updated
SoftBank Group shares fell as much as 13.2 percent in Tokyo on Monday, the steepest drop in nearly three months, after Anthropic chief executive Dario Amodei called on frontier labs to slow the rate at which they raise model capability. Sam Altman of OpenAI, Demis Hassabis of Google DeepMind and Elon Musk said they agreed with the Saturday essay, titled "We Must Pace the Frontier." Altman also told Fortune that OpenAI would not hold an initial public offering this year and would match Amodei's plan to put outside evaluators inside the company.
SoftBank is scheduled to have about $65 billion in OpenAI by October. The stock was already about 30 percent below its June peak after earlier reports that the listing might slip. Yugo Tsuboi, chief strategist at Daiwa Securities, said Monday's move reflected worry that OpenAI's value may not be as high as the market had priced and that no one yet knows how much the development pace will actually slow. Dan Baker at Morningstar said the slide "probably reflects the possibility that AI development may be slowed by regulators."
The sell-off was not only a SoftBank story. South Korea's Kospi dropped about 3 percent. SK Hynix fell 6.3 percent. TSMC slipped 1.2 percent in Taipei. Infineon lost 8.4 percent and Siemens Energy 7.4 percent in Europe. Nasdaq futures pointed to a 1.3 percent decline at the US open. Japan's Nikkei 225 was down 0.8 percent at 63,474.58. Hong Kong and Shanghai were slightly higher, a reminder that the shock hit firms tied to US closed models harder than the broader Asian tape.
Amodei's argument is that building too fast is reckless and that a future swarm of agents could cause hundreds of billions of dollars of damage by taking over large parts of the internet. Some researchers have disputed the mechanism. Markets did not wait for that debate to finish. They priced a world in which the hundreds of billions of dollars already committed to data centres, chips and power might be spent more slowly, which cuts the near-term earnings story that has held those stocks up.
Reuters reported, citing sources, that Anthropic is still preparing its own listing and that Nvidia would be an anchor investor. That split is easy to miss in a headline about a slowdown. One lab asks the industry to pace the frontier and still files for a public market. Another lab cancels a 2026 listing and agrees to outside evaluators. Investors have to decide which signal dominates: the safety language or the capital plan.
SoftBank's problem is leverage plus concentration. A single private company accounts for a large share of the story investors tell about Masayoshi Son's holding company. When that private company delays a listing and its chief executive talks about slowing capability growth, the holding company's equity becomes a high-beta bet on a timeline that just moved right. Chipmakers are a cleaner expression of the same fear. If training runs are paced, wafer demand in 2027 is less certain than the bull case assumed in June.
None of the chief executives announced a freeze. "Slow down" in this context means fewer giant capability jumps per year and more external testing between them. That can still be a very expensive industry. It is a less predictable one for people who bought the stocks on the assumption that each quarter would bring a larger model and a larger capex print. Monday was the first session in which that assumption was marked down in public.