“Cloud AI is a monopoly structure of a few big tech companies, but on-device AI is a B2C-centered, multi-product market that is an open opportunity market that is advantageous to Korea. “On-device AI chips such as NPU and AI SoC are the only promising area in the non-memory market that domestic fabless companies can enter.”

Lee Seung-woo, head of the Research Center at Eugene Investment & Securities, presented on the topic of ‘On-Device AI Semiconductor Industry Trends’ at the ‘AI Semiconductor Collaboration Forum’ held at the Westin Chosun Seoul on the 20th.

Director Seungwoo Lee said that one of the most notable areas in the semiconductor industry recently is on-device AI, and that unlike existing AI technologies that were mainly executed in a cloud environment, on-device AI is characterized by being embedded in various devices such as smartphones, cars, and PCs to independently perform AI functions.

He also said that the semiconductor market is changing very quickly, and that the market value of major companies is changing rapidly accordingly. He said that in the past Internet era, Cisco, Intel, and Microsoft were at the center, and in the mobile era, Apple, Qualcomm, and Samsung Electronics grew, and as we entered the AI era, companies such as NVIDIA, OpenAI, and Invidia are leading the market.

In particular, in the on-device AI market, lightweight AI models based on low-power MPUs, DSPs, and SOCs are attracting attention, it said.

Citing recent research results as an example, Center Director Seungwoo Lee stated that the AI semiconductor market is expected to grow from $80 billion in 2024 to over $200 billion in 2030, with a compound annual growth rate (CAGR) of 19%, which is more than double the growth rate of the existing semiconductor market.

In particular, the penetration rates of AI smartphones and AI PCs are also rapidly increasing, and are expected to reach 39% and 53%, respectively, in 2027.

The key to on-device AI semiconductors is model lightweighting, and existing large AI models require high computational power and operate in a cloud environment.If so, he noted that on-device AI must be optimized to operate efficiently even on small devices.

For this purpose, it was reported that technologies such as pruning, quantization, and knowledge distillation are being developed.

He also mentioned that semiconductor manufacturing methods are also changing, and while the key in the existing semiconductor industry was to produce the same capacity at a lower cost, in the future, the development of high-performance semiconductors is expected to become more important.

In particular, it is expected that the development of high-speed memory technology such as HBM (High Bandwidth Memory) will play a decisive role in improving AI semiconductor performance.

The growth of Chinese AI semiconductor companies is also noteworthy, he said, adding that Chinese companies such as Huawei and Cambricon have already established a strong position in the global market and their market capitalization has also increased rapidly.

Global major startups are also showing explosive growth. Cerebras has attracted a total investment of $1.14 billion, is pursuing an IPO in 2025, and is valued at $7 billion. Tenstorrent has also attracted a total investment of $1 billion, including $693 million in Series D in 2024, and is valued at over $2.6 billion.

Center Director Lee Seung-woo said that Korea's opportunities in global competition must be found in on-device AI.

As of 2024, the global demand for on-device AI chips installed in smartphones, laptops, and wearables is growing by more than 30% annually, and the center of gravity is shifting from cloud AI to edge on-device. The AI PC and AI smartphone market is estimated to grow to over 500 million units by 2026.

Korea also has the strength of being a powerhouse in memory semiconductors and the mobile ecosystem.

Director Lee Seung-woo said, “Cloud AI is a monopoly structure of a few big tech companies, but on-device AI is a B2C-centered, diversified market that is an open opportunity market that is advantageous to Korea. On-device AI chips such as NPUs and AI SOCs are the only promising area in the non-memory market that domestic fabless companies can enter.” He added, “Early-growth companies such as DeepX, Furiosa, and Rebellion have the opportunity to internalize system semiconductors ranging from chips, devices, and platforms based on this market.”

He continued, “In order to be competitive in the future on-device AI semiconductor market, not only technological innovation but also the creation of an industrial ecosystem is important.” He added, “As AI semiconductor technology advances, it is likely to become as widespread as electricity, which will create new opportunities across industries.”

He also argued that “Korea can play an important role in the on-device AI market by leveraging its existing strengths in mobile and memory semiconductors,” and that “to this end, the government and companies should work together to actively support the development of AI semiconductors and secure competitiveness in the global market.”