Asian semiconductor shares fell on Thursday, mirroring heavy losses in U.S. chip stocks after media reports said OpenAI had developed software improvements that materially reduce the computing resources needed to run some of its models. That, together with news that Meta Platforms is considering ways to monetize spare AI compute, weighed on investor sentiment across the region.
According to the report cited by market participants, OpenAI engineers created software optimizations that can cut inference costs by about half and reduce the number of Nvidia graphics processing units required to serve certain ChatGPT users. The development prompted investors to reassess the pace at which demand for AI chips might grow.
Market moves were pronounced among major Asian semiconductor names. South Korea's Samsung Electronics and SK Hynix each fell in the mid-to-high single-digit range, with declines reported between 8% and 10%. In Japan, test and equipment and manufacturing-related stocks were also hit: Advantest Corp shares dropped more than 7% while Tokyo Electron shares declined about 5%. In Taiwan, shares of Taiwan Semiconductor Manufacturing Co fell by more than 2%.
Adding to the market's unease was a separate report that Meta Platforms is working on a cloud offering designed to sell excess AI computing capacity. The plan, reported by Bloomberg News, would allow Meta to generate returns from large-scale investments in AI infrastructure by offering customers access to spare compute and potentially to AI models themselves through a new cloud service.
Combined, the OpenAI efficiency report and the Meta cloud discussions sparked concern that AI firms could extract greater value from existing hardware. Market participants interpreted that possibility as potentially diminishing the urgency for fresh chip purchases to expand capacity.
Market context and immediate effects
Investor reaction in Asia was driven by the twin headlines: software-led reductions in compute requirements at OpenAI and the prospect of a new buyer of spare compute capacity in Meta. The announcements were enough to trigger selling pressure across manufacturers, equipment suppliers and foundries linked to AI chip production and deployment.
What was reported
- OpenAI engineers implemented software optimizations said to lower inference costs by roughly half and to reduce GPU counts needed for some ChatGPT users.
- Meta Platforms is reportedly developing a cloud service to monetize unused AI compute and may offer access to AI models and spare processing capacity to customers.
Implications cited by investors
Market participants expressed concern that these developments could slow future hardware demand growth for AI applications by enabling more work to be done with existing infrastructure.