NVIDIA's Blackwell Architecture Adjustment: A Steep Decline in CoWoS-S Packaging Demand What are the Implications for TSMC and the Supply Chain?

NVIDIA's Blackwell Architecture Adjustment: A Steep Decline in CoWoS-S Packaging Demand What are the Implications for TSMC and the Supply Chain?TF Securities analyst Ming-Chi Kuo recently published a blog post analyzing NVIDIA's adjusted Blackwell architecture blueprint. He suggests that NVIDIA will significantly reduce its demand for CoWoS-S packaging for at least the next year

NVIDIA's Blackwell Architecture Adjustment: A Steep Decline in CoWoS-S Packaging Demand What are the Implications for TSMC and the Supply Chain?

TF Securities analyst Ming-Chi Kuo recently published a blog post analyzing NVIDIA's adjusted Blackwell architecture blueprint. He suggests that NVIDIA will significantly reduce its demand for CoWoS-S packaging for at least the next year. This adjustment will have profound implications for NVIDIA itself, its supply chain, and chip manufacturing giant TSMC. This article delves into the details of this architectural adjustment, explores the underlying reasons, and examines the potential impact on related companies.

Kuo's analysis indicates that the 200-series products in NVIDIA's Blackwell architecture will adopt a Dual-die (CoWoS-L manufactured) design, with representative products including GB200NVL72 and HGXB200. The 300-series is more complex, employing both Dual-die (CoWoS-L manufactured) and Single-die (CoWoS-S manufactured) designs. System products include GB300NVL72 (Dual-die) and HGXB300NVL16 (Single-die). Notably, NVIDIA has branded the 300-series Dual-die and Single-die designs as Ultradual-die and Ultrasingle-die respectively. Kuo believes this is merely a marketing distinction and lacks substantive technical differences.

Based on this architectural adjustment, Kuo explains recent market rumors about NVIDIA cutting CoWoS-S capacity. He predicts significantly decreased demand for CoWoS-S for at least the next year. This conclusion rests on several key points: firstly, the 200-series completely eliminates the original B200A (Single-die, CoWoS-S manufactured), directly removing CoWoS-S demand. Secondly, starting in Q1 2025, NVIDIA will aggressively promote the 200-series while reducing the supply of H-series products (CoWoS-S manufactured), further decreasing CoWoS-S demand. Lastly, although system products using the B300 series are expected to ship in large volumes in 2026, including those with Single-die/CoWoS-S designs, NVIDIA and its CSPs currently strongly favor the GB300NVL72 (CoWoS-L production). This means GB300NVL72 will be prioritized, creating a more urgent demand for CoWoS-L and delaying CoWoS-S demand.

This product line shift will inevitably impact NVIDIA and its supply chain's performance. Some suppliers will face significant challenges, which might explain the recent notable adjustments in the stock prices of some related companies. However, Kuo emphasizes that NVIDIA's slowing or cancellation of CoWoS-S expansion plans stems primarily from its own product line adjustments, not declining market demand. This change also aligns with TSMC's strategy of promoting CoWoS-L as the mainstream packaging solution.

So, how will TSMC respond to this change? Kuo predicts that investors will likely inquire about the slowdown in CoWoS-S expansion at TSMC's upcoming earnings call. While he cannot predict TSMC's specific response, he offers analytical perspectives. He points out that while CoWoS-S expansion is slowing, CoWoS-R (inferring CoWoS capacity changes through interposers might have errors) capacity is increasing. Kuo suggests that TSMC might respond by addressing overall CoWoS expansion, potentially stating that it's "on track." Considering TSMC internally still views AI/HPC as a significant growth driver this year and maintains a positive outlook on this trend, a downbeat statement about the AI outlook from official TSMC channels is unlikely.

From TSMC's perspective, the transition from B200 to B300 keeps the front-end-of-line (FEoL) process unchanged, while back-end-of-line (BEoL) changes can be addressed through engineering change orders (ECOs). Therefore, these two products can be considered the same, meaning the impact on TSMC is negligible regardless of the timing of this transition.

In summary, NVIDIA's Blackwell architecture adjustment, particularly the significant reduction in CoWoS-S packaging demand, will have complex effects on the entire supply chain. While for NVIDIA, this is more of a strategic product line adjustment than a market contraction, it still poses challenges for some suppliers. For TSMC, the impact might be relatively limited due to its comprehensive advanced packaging technology layout and optimistic outlook on the AI/HPC market. However, market dynamics are unpredictable, and future developments require continuous monitoring. The change in CoWoS-S market share and TSMC's response will be key areas to watch. Further research and data analysis will be necessary for a more comprehensive understanding of this architectural adjustment's long-term impact on the semiconductor industry. NVIDIA and TSMC's subsequent strategies, and the market's reaction to this change, will be crucial factors in shaping the future landscape of the semiconductor industry. This event also highlights the importance of rapid technological updates and agile supply chain responsiveness in the fast-paced technology sector. Continuous monitoring of industry dynamics and in-depth analysis of technological trends will help companies better navigate challenges and seize opportunities. This event provides a valuable case study for other chip design and manufacturing companies, providing insights into the role of advanced packaging technology in product strategy and effective supply chain risk management. Ultimately, only those companies that can adapt quickly to market changes and continuously innovate will achieve lasting success in this highly competitive industry. In the future, with the continued growth of the AI and HPC markets and the continued development of advanced packaging technologies, we can expect to see even more efficient and powerful chip products that will bring more innovation and progress to human society. However, at the same time, we also need to closely monitor potential risks and actively seek solutions to ensure the stable and healthy development of the industry chain. The decrease in demand for CoWoS-S is just one specific case in the development of the semiconductor industry, reflecting deeper technological and market trends that deserve in-depth consideration and discussion.