• AI chip stocks in Asia fell despite continued investment in AI infrastructure.
• AI spending is also driving data-centre and power demand across Asia.

Asian semiconductor and technology stocks fell on Monday after Anthropic CEO Dario Amodei called for companies developing advanced artificial intelligence models to slow the pace of capability improvements. OpenAI CEO Sam Altman and xAI chief Elon Musk also backed calls for a more measured pace of development amid concerns over AI safety.

The reaction was felt across several companies tied to AI infrastructure. Japan’s SoftBank fell as much as 13.2%, South Korea’s SK Hynix dropped 6.3%, and Taiwan Semiconductor Manufacturing Company (TSMC) slipped 1.2%, according to Reuters. Samsung Electronics closed 4.05% lower in Seoul.

US and European semiconductor stocks also declined. Nvidia fell 3%, AMD dropped 5.7%, and Dutch chip equipment maker ASML was down 5.8%, while European technology stocks fell about 2.5%.

The selloff came as investors were already scrutinising the pace of AI infrastructure spending across semiconductors, data centres, and power systems.

AI slowdown debate meets long-term chip spending

Amodei called for a slower rate of model development in an essay published over the weekend, arguing that safety work needs more time to keep pace with increasingly capable systems. “We must slow the pace at which we improve the capabilities of AI models,” Amodei wrote.

He cited advances in AI systems assisting with the development of subsequent models, as well as recent incidents involving autonomous AI agents. Amodei said these developments had increased his concerns about safeguards keeping pace with model capabilities.

OpenAI has separately slowed some development work. In August, the company said it temporarily reduced the pace of scaling after the OpenAI-Hugging Face security incident and preliminary evidence that its Astra model could meet the Critical cybersecurity capability threshold under its Preparedness Framework.

OpenAI said the changes included stronger isolation for research workloads, tighter network controls, and expanded monitoring. The company also said some safeguards added additional computing overhead to monitored inference workloads.

Reuters reported that investors were weighing whether slower development of frontier models could affect future spending on chips and data centres.

TSMC entered the second half of 2026 with a larger investment programme than previously planned. In July, the company raised its full-year capital expenditure forecast to between US$60 billion and US$64 billion, from an earlier range of US$52 billion to US$56 billion.

TSMC reported second-quarter revenue of US$40.2 billion and said AI-related demand remained strong. The company also raised its forecast for 2026 revenue growth to slightly above 40% in US dollar terms.

Those investments are planned over several years rather than around individual model launches. TSMC CEO C.C. Wei said the company works with customers and their customers on multi-year product roadmaps and production plans, with the process of developing technology, preparing capacity, and reaching high-volume production taking more than five years.

TSMC is also building 13 leading-edge semiconductor and advanced packaging fabs in Taiwan as part of its broader capacity expansion.

Memory suppliers have also increased production plans around AI servers. Samsung began commercial shipments of HBM4 earlier this year and said it expects its HBM sales to more than triple in 2026 compared with 2025.

Samsung said in its second-quarter results that its memory business recorded another quarterly high while focusing production on server products. The company expects demand for server DRAM, enterprise SSDs, and HBM to remain supported by AI infrastructure capital expenditure during the second half of 2026.

SK Hynix has also tied its capacity plans to multi-year customer agreements. The company said in July that it had completed long-term supply agreements with around 10 customers, including strategic customers, and was discussing further multi-year contracts with other large clients.

The company approved another 54 trillion won in semiconductor investment in August. The plan includes 35.2 trillion won for a new DRAM fab in Yongin and 19.1 trillion won for a NAND fab in Cheongju, with cleanroom openings scheduled for 2029 and 2028 respectively. SK Hynix said capacity additions would be aligned with customer demand.

The company said its longer-term manufacturing and packaging investments would be phased according to customer demand and investment efficiency.

Long investment cycles are also evident in chipmaking equipment. ASML’s current extreme ultraviolet lithography systems, which are used to manufacture advanced chips, are largely sold out through 2027, according to Reuters.

Samsung and SK Hynix plan to introduce ASML’s newer High Numerical Aperture EUV systems into memory production from 2028, while TSMC plans to use the technology for advanced-node high-volume manufacturing from 2030.

The High-NA systems cost about US$400 million each, roughly twice the price of ASML’s existing EUV machines.

Frontier-model training is only one source of AI computing demand. Deployed AI systems also require infrastructure for inference, including agentic workloads.

TSMC said in July that the expansion of agentic AI was increasing demand for CPUs inside AI data centres alongside AI accelerators. Wei said the company was working with CPU customers using x86, Arm, and RISC-V architectures to prepare technology and capacity for those workloads.

Samsung has also linked memory demand to AI inference. The company said earlier this year that it was expanding sales of high-performance NAND products used for inference workloads, alongside HBM, server DRAM, and enterprise SSD products.

SK Hynix’s next-generation HBM4E memory is also being developed for both AI training and inference. The company said the product improves data processing performance and power efficiency for AI data centres and large-scale computing systems.

Spending forecasts also remain high. Morgan Stanley Research estimates that capital expenditure by the five largest US technology companies will increase from nearly US$800 billion in 2026 to around US$1.2 trillion in 2027 and US$1.4 trillion in 2028.

The spending includes chips, servers, data centres, power infrastructure, and other capacity required for AI services. UBS said in July that near-term AI spending remained strong, while flagging pressure on hyperscaler free cash flow and the risk of slower spending growth after 2027.

AI infrastructure spending is reaching data centres and power grids

On September 8, OpenAI and Australian infrastructure company Firmus announced a multi-year agreement covering dedicated AI compute capacity at two Firmus data-centre sites in Malaysia.

Firmus said OpenAI would become an anchor customer and that its contracted capacity across all customers had risen above 900 MW. The company has seven planned or operating AI data-centre sites across Australia, Singapore, Indonesia, and Malaysia.

Malaysia’s Energy Commission has already identified data-centre expansion as one factor behind higher electricity demand. It said electricity demand reached new peaks in 2025, driven by factors including economic activity, electrification, industrial expansion, weather conditions, and data-centre growth.

Reuters reported that data centres accounted for 9.3% of electricity consumption in Peninsular Malaysia during part of August, compared with an average of about 7% in 2026. Officials estimate the share could rise to as much as 31% by 2035.

Tenaga Nasional had signed electricity supply agreements with 49 data centres by September 2025, representing maximum demand of 7.1 GW, according to data published by the Malaysian Investment Development Authority. Actual utilisation reached 710 MW at the time, up 75% from the end of 2024.

South Korea is also planning for higher electricity demand from semiconductor production and AI data centres. Energy Minister Kim Sung-whan said in September that these projects could add between 25 GW and 30 GW to national electricity demand.

The projected increase is being incorporated into the government’s long-term electricity planning. Nuclear power currently accounts for nearly a third of South Korea’s electricity generation.

Grid financing has also become part of the semiconductor expansion plans. Reuters reported on Monday that Korea Electric Power Corp had proposed requiring Samsung and SK Hynix to provide around 25 trillion won in advance payments for electricity infrastructure serving planned semiconductor clusters. Both companies rejected the proposal.

Samsung and SK Hynix declined to comment on the proposal. A company official told Reuters that the chipmakers questioned whether such large upfront payments were necessary, citing uncertainty over the long-term durability of semiconductor demand.

The proposed infrastructure would supply planned semiconductor mega-clusters. South Korea’s government has separately accelerated semiconductor projects originally expected around 2045, with completion now targeted for the early 2030s.

Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

Tech Wire Asia is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.