Samsung Foundry received a 4nm AI chip order, but advanced packaging may become a manufacturing bottleneck.
Samsung Foundry has reportedly secured an AI chip foundry order from a major data center client, driven by significant improvements in its 4nm process yield rate, which sources indicate has reached 75%. This development marks a notable shift in the competitive landscape, as Samsung aims to strengthen its position in the high-demand AI chip manufacturing sector. The semiconductor industry overall continues to recover more slowly than anticipated, though the second half of 2023 may see traditional peak season effects, with visibility on demand momentum and economic conditions remaining unclear.
The contrast between TSMC and Samsung Foundry's market performance has drawn attention from Korean media outlets including ZDNet and Korea Economic Daily. While TSMC secured high-end GPU-related AI chip orders from Nvidia and AMD during the economic downturn—with order volumes so large that TSMC expanded production capacity rapidly—Korean media previously lamented Samsung's inability to secure comparable orders. However, with Samsung's improved 4nm yield rate and newly secured orders, the competitive dynamics have begun to shift.
According to industry sources and security firms, Samsung has already secured an AI chip foundry order from a data center client, with foreign media identifying AMD as a possible client. Additionally, both Google and AMD have reportedly signed agreements with Samsung to manufacture AI chips using its 4nm process. These developments suggest growing confidence in Samsung's foundry capabilities among major technology companies.
Despite these advances, AMD may not immediately transfer orders from TSMC to Samsung. AMD's CPU technology still relies heavily on TSMC's advanced process, making large-scale order transfers unlikely in the short term. Nevertheless, sources suggest that AMD could leverage an order transfer to Samsung as a strategic tool to impact Nvidia, particularly given TSMC's current prioritization of production expansion to fulfill Nvidia's orders.
The critical limiting factor for high-end AI GPU and HPC chip production is not front-end processing capacity but rather the severe shortage of back-end advanced packaging. TSMC's CoWoS advanced packaging technology remains the market leader in yield rate and customer satisfaction, making it the preferred choice for AI HPC clients. Samsung's 4nm yield improvements alone do not guarantee success in securing high-end AI GPU manufacturing contracts. Instead, Samsung's success depends on presenting competitive 2.5D or 3D packaging solutions or establishing partnerships with high-quality OSAT companies. Only by securing sufficient advanced packaging capacity can Samsung's 4nm process realize its potential in the AI HPC market.