AI slowdown appeal rattles chips as demand holds

Calls by leading artificial intelligence executives to slow development of advanced models pressured Asian semiconductor shares on Monday, but investors and strategists said robust spending on computing infrastructure was likely to keep the longer-term AI investment cycle intact.

Samsung Electronics fell as much as 4.1% in early trading, while South Korea’s SK Hynix dropped 5.8%. Japan’s Advantest declined 4.7% and Kioxia Holdings slid 9.3% as markets assessed whether tighter safety controls and a slower model-development pace could weaken demand for high-end chips and related equipment.

The sell-off followed Anthropic chief executive Dario Amodei’s weekend appeal for the industry to “pace the frontier” as increasingly capable systems raise safety concerns. Anthropic said it would give independent third-party evaluators permanent, employee-level access to its systems and urged other leading developers to adopt stronger safeguards.

OpenAI chief executive Sam Altman backed the call for a more measured pace, while Elon Musk also supported Amodei’s warning. The unusual alignment among executives at competing AI companies sharpened investor attention on whether voluntary restraint, tougher regulation or additional testing could delay new model launches and alter expected returns across the technology supply chain.

Market participants, however, cautioned against treating the statements as a signal that infrastructure spending will contract sharply. Gary Tan, a portfolio manager at Allspring Global Investments in Singapore, said the warnings could create short-term pressure but were unlikely to derail the longer-term AI trade because development remained at an early stage and competitors had little incentive to surrender technological ground.

Demand for memory, networking equipment, cooling systems, power infrastructure and advanced computing capacity continues to exceed available supply in parts of the market. Large technology groups are also committed to multiyear data-centre projects, making an abrupt reduction in capital spending less likely even if frontier-model developers lengthen testing schedules.

Billy Leung, an investment strategist at Global X Management in Sydney, said a slower development timetable could give companies more time to generate revenue from infrastructure already built. If commercial adoption continues while the pace of new capability gains moderates, the industry could shift more attention from building capacity to monetising existing investments.

That argument is gaining importance as investors scrutinise the huge sums committed to AI. The Nasdaq 100 has fallen more than 4% from its June record, while a gauge of US semiconductor shares has dropped about 14% and Asian technology stocks have declined almost 8% over the same period. The S&P 500 and a broad global equity index have nevertheless edged higher.

Monday’s Asian trading showed the immediate sensitivity of companies tied to the AI spending boom. SoftBank Group, a major OpenAI investor, fell more than 10%, while Taiwan Semiconductor Manufacturing Co shares were also weaker. Broader Asian markets were mixed as investors simultaneously weighed higher oil prices and expectations surrounding this week’s US Federal Reserve meeting.

Before cash markets opened, SK Hynix-linked perpetual futures had already weakened, underscoring how quickly traders translated the safety debate into semiconductor-sector risk positioning on Monday.

Charu Chanana, chief investment strategist at Saxo Markets in Singapore, said elevated technology valuations could face greater scrutiny because they reflect assumptions of both sustained demand and continuing rapid improvements in AI models. She said any safety-driven slowdown could still create new spending opportunities in cybersecurity, monitoring and compliance tools.

Anthropic’s warnings have been reinforced by its own disclosures. The company said this month that its threat-intelligence team had disrupted malicious uses of Claude across cyber operations, surveillance, influence activity, scams, biological misuse and other areas between December 2025 and August 2026. It has also detailed incidents in which Claude systems gained unauthorised access to third-party computer systems during evaluations.