Samsung Electronics has set a goal to increase the sales proportion of artificial intelligence (AI) chips in its foundry (semiconductor contract manufacturing) business to 50% by 2028.
According to sources in the semiconductor industry on Nov. 20, the current revenue breakdown by application for Samsung Electronics’ Foundry Business Division (based on internal projections for this year) shows that mobile chips account for 54%. High-performance computing (HPC) for AI servers and data centers represents 19%, while automotive chips, including those for autonomous driving, constitute 11%.
The revenue portfolio is expected to undergo significant changes by 2028. Samsung Electronics plans to decrease the mobile segment to the low 30s percentage-wise, increase HPC to 32%, and boost automotive to 14%. The company also intends to double the number of external customers by 2028 compared to this year.
Currently, the main clients of Samsung Electronics’ Foundry Business Division are the System LSI Business Division of Samsung Electronics, Qualcomm, and other fabless companies (specializing in semiconductor design) that create chips for smartphones. This focus has been due to the production of mobile semiconductors primarily for Samsung smartphones. Mobile chips make up 54% of this year’s projected sales.
While this has the advantage of stable revenue, there have been continuous criticisms of over-reliance on mobile. Samsung has not accumulated as much experience or performance in producing large-sized AI HPC chips and chips for autonomous vehicles as its competitors. Large orders related to AI and autonomous driving semiconductors have shifted to TSMC, widening the market share gap.
Recently, the trend is shifting. Samsung Electronics’ Foundry Business Division is receiving continuous orders for AI semiconductor contract manufacturing. These include graphics processing units (GPUs) and central processing units (CPUs) for AI servers and data centers. A notable example is AMD, the world’s second-largest CPU company and a competitor of Intel. Foreign media reports suggest that AMD is seriously considering mass-producing its next-generation server CPUs using Samsung Electronics’ 4-nm foundry process. This is due to Samsung’s 4-nm process yield (proportion of good products) reaching a competitive 70% level with TSMC. Samsung is emerging as an alternative to TSMC.
The trend of major tech companies like Google, Microsoft, and Amazon developing their own AI semiconductors also plays a role. These companies, lacking their own factories, outsource production to foundries like Samsung Electronics. Microsoft’s recently unveiled AI semiconductor is being produced using TSMC’s 5-nm process, but there is a possibility that future production may shift to Samsung Electronics. An industry insider explained, “For fabless companies, reducing reliance on TSMC is advantageous in price negotiations.”
Samsung Electronics’ Foundry Business Division plans to increase the sales proportion of HPC from 19% in 2023 to 32% by 2028, and automotive chips from 11% to 14% in the same period. In contrast, the mobile segment will be reduced to the early 30s. Diversifying the service portfolio and increasing the production of high-value-added chips can enhance profitability.
Key to this plan is the advanced process technology for producing high-performance, low-power, high-efficiency chips for AI. Samsung Electronics aims to secure more AI semiconductor clients by enhancing the completion of cutting-edge processes below 3 nm. The company has planned to start producing automotive and HPC chips using a 2-nm process from 2026 and unveil the “dream process” of 1.4 nm in 2027.