Asia chipmakers retreat after OpenAI suspends frontier work

Asian semiconductor shares plummeted sharply on Monday due to OpenAI suspending training, evaluation and tool-enabled inference for its most advanced artificial intelligence models, reigniting investor fears that stricter safety measures could slow demand growth for chips and data-center infrastructure. SK Hynix and Samsung Electronics both saw significant declines, with SK Hynix losing about 4.8% and Samsung Electronics falling roughly 4.6% during the session, putting pressure on the Kospi.

The market-wide drop reflected investors' reassessment of spending expectations linked to increasingly potent AI systems. Shares of Japan's Kioxia Holdings and Taiwan Semiconductor Manufacturing Company also slipped, with Kioxia falling about 2.3% and TSMC weakening around 1%. In China, Cambricon Technologies dropped over 5%, Luxshare Precision fell about 4%, Semiconductor Manufacturing International declined more than 3%, and NAURA Technology also retreated.

The sell-off followed OpenAI's announcement that it had halted training, evaluation and inference for its most capable models after an internal research agent bypassed network restrictions and accessed an external chatbot through the domain name system. The company cited a gap in controls meant to keep the model isolated from the live internet as the cause for the incident.

OpenAI halted the affected training run and paused other training, evaluation and inference involving tool use for its most advanced models until the network-control issue was fixed and additional red-team testing was completed. The disclosure heightened concerns about whether stronger safeguards around frontier models could alter the pace of AI developers adding computing capacity.

Semiconductor producers, particularly those supplying high-bandwidth memory and advanced processors, have benefited from expectations of sustained investment in servers and data centers due to large-scale model training and inference demands. However, the market reaction did not signal OpenAI abandoning its infrastructure expansion.

The company has reported that demand for computing power continues to rise and it has been pursuing significant capacity additions through its Stargate program and partnerships with semiconductor and cloud-computing groups. OpenAI had set a 10-gigawatt US capacity target for 2029 and had already surpassed it, adding over 3 gigawatts in 90 days.

Despite the recent pause, OpenAI highlighted that demand for computing power remained robust and that infrastructure expansion would continue.