The essay was published over the weekend, and by Monday it had a price. Dario Amodei, the chief executive of Anthropic, argued in a piece that the pace of frontier AI development should be slowed. On September 14, markets from Tokyo to New York sold off, and the companies tied most tightly to AI took the heaviest losses.
SoftBank was the most visible casualty. The Japanese company’s shares fell as much as 13.2 percent in Tokyo before closing down about 11 percent at 5,795 yen, the steepest one-day drop since July 17. SoftBank has become a proxy for AI exposure through its stake in Arm and its investments in OpenAI.
The irony of the day was hard to miss. Even as its stock fell on the slowdown narrative, SoftBank was finalizing an $11.87 billion two-year syndicated loan to invest in OpenAI, according to people familiar with the matter. The loan exceeded its original $10 billion target by nearly a fifth, with about 20 banks participating.
Masayoshi Son, SoftBank’s founder, has said he plans to pour nearly $65 billion into OpenAI by October. The contrast between that commitment and the day’s selloff captures the tension running through the market: capital is still flowing into AI at full force even as investors recoil from the companies supplying it.
SoftBank’s outsized role in the selloff reflects its structure. The company’s value is tied to Arm, the chip designer whose technology sits in nearly every AI device, and to a portfolio of stakes in AI companies led by OpenAI. When investors want to reduce AI exposure quickly, SoftBank is one of the most liquid ways to do it.
The selling was broad and global. South Korea’s KOSPI fell 3.37 percent to 6,684.37, with SK Hynix down 5 percent and Samsung Electronics down 3 percent as foreign investors sold more than 1.1 trillion won of shares. Japan’s Nikkei 225 slipped 0.83 percent, and its semiconductor index fell as much as 5.5 percent during the session.
Taiwan Semiconductor Manufacturing dropped 1.2 percent, and the losses extended to Europe, where ASML fell more than 4 percent and Infineon dropped 6 percent. In premarket trading in the United States, Intel was down nearly 6 percent, Micron about 5 percent and Nvidia more than 2 percent, with Nasdaq futures off between 1.3 and 1.5 percent.
The hardest-hit names were the ones that had been the tightest. Memory makers and semiconductor equipment suppliers, the two segments most exposed to AI capital spending, led the declines. KB Securities noted that memory inventories at Samsung and SK Hynix have fallen below ten days of supply, a level that leaves little buffer.
The market’s logic is straightforward to state and harder to defend. If the loudest voices in AI are calling for a slower pace, then the billions committed to chips, memory and data centers may arrive later than expected. Investors who had priced immediate demand responded by marking down the companies that would feel the delay first.
Analysts were careful to draw a line. The selloff, they said, is aimed at the pace of capital spending rather than the underlying demand for AI. The technology’s adoption has not reversed, and the shortage of memory and accelerators shows buyers are still waiting for supply. What changed is the question of how fast that supply needs to arrive.
Amodei’s essay is part of a larger movement. Researchers and some industry leaders have argued that frontier development should slow while safety catches up, and that argument has gained traction in Washington even as the White House has dismissed it. The markets are now pricing that debate directly into valuations.
The White House and much of the industry have pushed back. President Trump has called AI doom warnings something that will not happen and framed the race in national-competition terms, while his Treasury secretary has emphasized maintaining American computing capacity. The disagreement leaves investors without a clear signal on policy.
The market has been here before. AI-related shares corrected sharply earlier in 2026 on fears about spending, only to recover as earnings confirmed that demand was real. Whether this episode follows the same arc depends on whether the slowdown argument changes actual purchasing decisions, which has not yet happened.
The OpenAI loan itself underscores how much capital the sector is still absorbing. An $11.87 billion facility to fund an equity stake is unusual in any market, and the fact that banks oversubscribed the deal even as AI stocks fell suggests the two sides of the trade are moving on different time horizons.
The memory market adds a twist. Inventory below ten days of supply suggests the industry cannot meet current demand, let alone a surge. That tightness has supported prices and profits at Samsung and SK Hynix. If spending slows, that scarcity could ease quickly, turning a strength into a vulnerability.
The contrast inside SoftBank’s own day illustrates the split. The company borrowed more than it had planned to deepen its OpenAI bet on the same day its shares fell the most in months. Either Son is right that the money will pay off, or the market is right that the terms of the bet have changed.
For now, the selloff is a repricing, not a reversal. The companies that make memory and the machines that make chips are still sold out for quarters ahead, and no customer has publicly cut an order. The question the market is asking is what happens when the orders that have not yet been placed are decided.
What the day established is that the slowdown debate has moved from the opinion pages into prices. The market has not concluded that AI demand will collapse. It has concluded that the pace of investment, which had been assumed to keep rising, is now a variable, and it is repricing everything downstream of that assumption.
