An analysis said that China's drive to localize semiconductors is not a headwind for Samsung Electronics(005930) but could instead be a factor that increases memory demand. That is because China-made artificial intelligence (AI) accelerators have lower compute efficiency than Nvidia graphics processing units (GPUs), requiring more accelerators and more memory to handle the same level of AI computation.
On the 28th, Kim Dong-Won, head of research at KB Securities, said in a report, "China's semiconductor localization policy is expected to act not to reduce Samsung Electronics' memory demand but rather as a direct factor that creates additional demand," adding, "China-made AI accelerators have lower compute efficiency than Nvidia GPUs, so more accelerators and more memory are needed to process the same amount of AI computation."
The technology gap between Chinese memory makers and Samsung Electronics is also expected to work in Samsung Electronics' favor for the time being. In particular, in high-performance server DRAM, the analysis said there still exists a gap in technology and product competitiveness with China's Changxin Memory Technologies (CXMT).
The expansion of AI investment by China's big tech is also expected to support demand for Samsung Electronics. According to KB Securities, major Chinese cloud service providers (CSPs) such as Baidu, Alibaba, Tencent, and ByteDance Ltd. are asking Samsung Electronics for multi-year long-term memory supply agreements (LTAs). It was also analyzed that they are simultaneously requesting foundry outsourcing cooperation for producing their in-house developed AI accelerators (ASICs).
Kim said, "The more China's AI investment expands going forward, the stronger the benefit to Samsung Electronics will become," rating Samsung Electronics as the biggest beneficiary of the expansion in China's AI data center investment.
The scale of China's AI data center investment itself is also expected to swell rapidly. KB Securities projected that China's AI data center capacity will double from 40 gigawatts (GW) this year to 80 GW in 2030. Over the same period, the United States is also expected to expand from 55 GW to 121 GW, a 2.2-fold increase.
The combined AI data center capacity of the United States and China is expected to rise 2.1 times from 95 GW this year to 201 GW in 2030. Cumulative AI data center investment by the two countries from 2026 to 2030 was estimated at about $5 trillion. The average annual investment was expected to exceed 1.8 quadrillion won.
KB Securities said that considering the sophistication of AI models, the expansion of inference demand, and increased adoption of in-house AI accelerators, growth will not be limited to adding more data centers; compute performance per server and memory installed per server are also expected to increase together.
Kim said, "Global AI data center investment is expected to unfold with the United States leading the market and China accelerating growth in a two-strong structure," adding, "We judge that the AI infrastructure investment cycle is still only at the beginning."