DeepSeek disrupts AI chip market with low-cost model
Cheaper AI infrastructure sparks overcapacity concerns among chipmakers.
Chinese startup DeepSeek has disrupted the semiconductor industry with a low-cost AI model using less advanced chips, sparking concerns of AI infrastructure overcapacity as demand shifts to cheaper hardware.
According to Associate Professor Goh Puay Guan from the Department of Analytics and Operations at NUS Business School, DeepSeek’s model “definitely changes the assumption of how the AI industry will develop.”
He explained that the prevailing belief in the industry was to rely on high-end chips and large data centers for AI processing. However, DeepSeek’s approach demonstrates that “it's possible to provide reasonable quality at a lower cost,” signaling a shift in how AI can be developed.
“This concept is aligned with frugal innovation—using fewer resources to develop a lower-cost product. It potentially means the creation of two different market segments: one for complex, computing-intensive AI models and another for consumers who can use lower-cost models that meet basic needs,” Goh added.
The implications of DeepSeek’s model extend beyond just the technology. As governments worldwide—including those in China, the U.S., Europe, Japan, and Korea—encourage investments in semiconductor capacity for strategic reasons, there is a risk of global overcapacity. Goh warned, “When investments are driven by geopolitical considerations rather than purely commercial factors, it could lead to overcapacity at some point.”
Roger Sheng, Vice President Analyst at Gartner, noted that DeepSeek’s approach could significantly impact the demand for high-performance AI infrastructure. “The distilled model size will reduce the memory demand for AI training and inference,” Sheng said. He further explained that this reduction would lessen dependence on costly GPU clusters, such as Nvidia’s B2 100 NVL 72, which could see declining demand.
“Nvidia will be the first impacted vendor in the market,” Sheng said. “[However,] it will provide a great opportunity to other AI chip suppliers such as AMD, Huawei, and smaller chip vendors to develop chips focusing on workload capabilities using DeepSeek models.”
Additionally, Sheng highlighted that DeepSeek’s technology would drive AI adoption among smaller organisations. “Even individuals can leverage cheaper servers and affordable AI systems to develop AI software for broader industry use cases. This will generate more differentiation and innovation in enterprise and user vertical AI applications.”
Goh also emphasised the role of software-as-a-service (SaaS) applications in this evolving landscape. “More SaaS applications are being developed, allowing people to rent services or use software on a pay-per-use basis. This will enable many more companies to access AI resources without heavy infrastructure investments.”
