Information source:
https://wccftech.com/nvidia-orders-144-million-of-memory-chips-from-samsung-for-china-specific-blackwell-b40-ai-gpu-report/
Samsung Electronics is preparing to double its GDDR7 memory capacity to meet Nvidia's $144 million bulk order demand, behind which Nvidia specially designed Blackwell B40 AI processor project to re-enter the Chinese market. This major business cooperation not only marks the profound reconstruction of the global AI chip supply chain, but also reflects the strategic adaptation and innovation of technology giants under geopolitical pressure. With the continued evolution of the US chip export control policy to China, Nvidia is maintaining its dominance in the global AI market through technology route adjustments and supply chain diversification.
The news disclosed by South Korea's "ET News" shows that Nvidia CEO Huang Renxun has engaged in substantive negotiations with the Trump administration on the export license of a new generation of China's dedicated AI chips. The B40 chip uses GDDR7 memory configuration instead of traditional HBM memory, and this design choice is obviously to circumvent the export restrictions of the United States on high-bandwidth memory technology. The promotion of this project not only concerns Nvidia's business prospects in this key market in China, but also affects the competitive landscape and technological development direction of the global AI chip industry.
Deep logic of technical differentiation strategies
Nvidia's choice of GDDR7 memory as the core configuration of the B40 chip reflects the exquisite strategy of seeking a balance between technical compliance and performance requirements. Compared with HBM memory, GDDR7 has a certain gap in bandwidth performance, but its relatively mature technical characteristics and relatively loose export control environment make it an ideal choice for entering restrictive markets.
According to technical analysis by the Semiconductor Industry Association, the theoretical bandwidth peak of GDDR7 memory is about 1000GB/s, which is lower than the 1280GB/s of HBM3e, but it still provides sufficient data throughput for most AI inference and training tasks. More importantly, the manufacturing cost of GDDR7 is about 30% lower than that of HBM memory, which is of great significance to the price-sensitive Chinese market.
Samsung's technical strength in the field of GDDR7 has laid a solid foundation for this cooperation. The company's 24Gbps GDDR7 product has been mass-produced at scale, reaching the industry-leading level in power consumption control and signal integrity. Compared with its competitor Micron Technology, Samsung's GDDR7 exhibits better thermal management performance and long-term stability in high-frequency operating conditions, which are crucial for AI data center environments that require long-term high-load operation.
Nvidia's previously launched H20 processor has encountered sales difficulties in the Chinese market, mainly because its significantly weakened performance to meet export control requirements has led to its serious lack of cost-effectiveness. The design philosophy of the B40 chip obviously learns this lesson and tries to maximize product performance within the compliance framework, thereby maintaining an advantage in the fierce market competition.
Multiple impacts of supply chain strategic restructuring
This huge order has important strategic value to Samsung. Samsung's position in the AI processor memory supply chain has long been relatively marginalized, mainly because its HBM products lag behind South Korea's competitor SK Hynix in terms of technical specifications and manufacturing yields. With its first-mover advantage and continuous innovation in the field of HBM technology, SK Hynix has become one of Nvidia's most important memory suppliers.
However, the B40 project provides Samsung with an opportunity to re-enter Nvidia's core supply chain. GDDR7 memory is Samsung's technical strength, and the company has a complete R&D system and manufacturing capabilities in this field. According to industry analysts estimates, Samsung's current monthly production capacity of GDDR is about 1 million chips, and Nvidia's requirement to double its production capacity means that Samsung needs to increase its monthly production capacity to 2 million in a relatively short period of time.
This large-scale capacity expansion requires not only huge capital investment, but also complex supply chain coordination and technological upgrades. The manufacturing process of GDDR7 memory involves advanced EUV extreme ultraviolet lithography technology, while the procurement and use of related equipment are subject to strict international export control supervision. Samsung must ensure that its capacity expansion plans fully comply with various international regulations and avoid touching the regulatory redline of sensitive technology transfer and dual-purpose technologies.
From a more macro market perspective, the scale and growth potential of China's AI chip market are still huge. According to data from the China Semiconductor Industry Association, the size of China's AI chip market is expected to reach RMB 120 billion in 2024, with an annual growth rate of more than 25%. Despite various technological restrictions, the demand for high-performance AI processors in the Chinese market continues to grow, providing important business opportunities for international manufacturers such as Nvidia.
Company adaptation in geopolitical games
The Trump administration's re-examination of the export policy to China has brought a new policy window for Nvidia's Chinese business. At a recent White House press conference, President Trump publicly confirmed his communication with Jensen Huang on Blackwell's chip export licenses and said the relevant policies would be a "simplified version of major events." This statement has been widely interpreted by the industry as a positive signal of moderate policy loosening.
However, the two parties in the U.S. Congress remain highly vigilant on the issue of technology exports to China. Members of the Senate Foreign Relations Committee recently jointly spoke out, emphasizing that any export licensing involving advanced AI technology must undergo strict national security review procedures. This complex political environment puts Nvidia's B40 project at certain policy implementation risks.
At the same time, the rapid development of China's local AI chip industry has also added new variables to market competition. Products such as Huawei HiSilicon's Asteng 910B, Alibaba's Hanguang 800, and Baidu's Kunlun chips are rapidly iterating and upgrading in technical performance. Although these products still have a gap with Nvidia in terms of absolute performance indicators, their technological progress speed and localized service advantages cannot be ignored.
Political coordination of US allies such as the EU and Japan is also an important factor affecting the global AI chip trade pattern. The export restrictions on EUV lithography equipment of the Netherlands ASML company, Japan's export controls on key semiconductor materials, and the EU's AI technology export control framework are brewing together to form a complex multilateral technology control system. Whether Nvidia's B40 project can be successfully promoted depends largely on the policy direction and implementation of this multilateral coordination mechanism.
From the perspective of technological innovation, the implementation of the B40 chip project will promote an important evolution of the design ideas of AI processors. Traditional high-end AI processor design pursues the ultimate performance and usually uses the most advanced memory technology and process technology. However, geopolitical constraints force chip designers to rethink the balance between performance and availability.
GDDR7 memory adoption requires compromise on certain technical indicators, but also stimulates the potential for system-level optimization innovation. Nvidia's engineering team is exploring ways to compensate for bandwidth deficiencies by improving the memory controller architecture, optimizing data transmission protocols, and adopting more efficient data compression algorithms. These technological innovations may give birth to new AI processor architecture paradigms and open up new directions for the technological development of the entire industry.
