NVIDIA Corporation (NVDA.US) is unwilling to give up the Chinese market! Planning to launch another "China-exclusive version" of AI chips.

According to the latest reports from the media, global "AI chip leader" NVIDIA Corporation has informed its most important clients in the Chinese market, including ByteDance, Alibaba Group Holding Limited Sponsored ADR, and Tencent, among other tech companies, that the company is redesigning its AI chip design architecture to comply with the latest export restrictions imposed by the US government, while still continuing to supply AI chips to Chinese companies. According to the latest report by "The Information," NVIDIA Corporation CEO Jensen Huang revealed this latest China-specific AI chip plan to customers during his recent high-profile visit to the Chinese market. Huang's visit came after the US government notified NVIDIA Corporation that it will require special government approval to sell AI chips to customers in the Chinese market - that is, the US government AI restriction list for AI chips has expanded to include H20, which was previously exported to China with reduced performance compared to H100/H200. H20's performance compared to H100/H200 has been significantly reduced, and it is an exclusive customized version of the AI chip approved by NVIDIA Corporation for export to the Chinese market. However, the latest restrictions by the Trump administration effectively prohibit the sale of H20 to Chinese customers. As a result, NVIDIA Corporation is expected to incur additional expenses of up to $5.5 billion in the upcoming financial quarter, which the company disclosed in its Form 8-K filing, leading to a significant drop in its stock price by nearly 7% thereafter. Despite escalating tensions between the US and China, NVIDIA Corporation remains committed to developing new AI chips for the Chinese market that comply with regulatory restrictions, highlighting the importance of the eastern market for the performance of the California-based semiconductor giant NVIDIA Corporation. "The Information" reported that Chinese tech giants such as ByteDance, Alibaba Group Holding Limited Sponsored ADR, and Tencent had ordered AI chips worth over $16 billion in the first three months of this year, and it is currently unclear how the latest US government ban will affect these AI chip orders. The report also stated that NVIDIA Corporation informed some customers in the Chinese market that it would need regulatory approval from the US Department of Commerce before launching any new versions of AI chips specifically for the Chinese market. As of the 2025 fiscal year ending January 26, NVIDIA Corporation achieved sales of up to $17.1 billion in the Chinese market, accounting for approximately 13% of the total revenue of the semiconductor giant of $130.5 billion. Will the upcoming China-specific version of AI chips follow the ASIC route rather than the general GPU route? Following the latest report by "The Information," some industry analysts have suggested that NVIDIA Corporation may shift its AI chip technology roadmap from general GPU to AI ASIC specifically tailored for AI training/inference in order to meet the US government's export restrictions for the Chinese market. According to these analysts, the specific architecture of the GPU makes it almost impossible for NVIDIA Corporation to launch AI chips that comply with US export restrictions without significant performance reduction in the short term. However, significant performance reduction may make NVIDIA Corporation's AI chips less cost-effective compared to domestic AI chips. Nevertheless, some analysts suggest that NVIDIA Corporation's China-specific AI chip strategy may focus on "quick and moderate downgrades on AI GPU architecture to avoid regulatory red lines," such as reducing the NVLink interconnect rate, bandwidth, or tensor computing capability. In the medium to long term, the company may continue to evaluate the possibility of launching dedicated AI ASICs for AI inference. AI ASICs, also known as "customized AI chips," "dedicated AI chips," or "AI application-specific integrated circuits" within the industry, are designed to efficiently execute specific AI tasks (such as deep learning, artificial intelligence inference and training systems, etc.) using proprietary hardware architectures to enhance AI computing efficiency, reduce power consumption, and improve performance, especially in large-scale AI parallel computing scenarios. For example, the TPU (Tensor Processing Unit) developed by Alphabet Inc. Class C together with Broadcom Inc. is a typical AI ASIC used for deep learning inference and training, optimizing key calculation operations such as matrix multiplication to improve AI computing efficiency. Broadcom Inc. and Marvell Technology, Inc. are currently dominant in the AI ASIC field. The low-cost DeepSeek paradigm indicates that AI inference can be optimized through algorithm engineering to reduce inference costs, making it more convenient and affordable to deploy large models. This also means that the advantage of AI ASIC in AI inference will be even greater in the future. Although the general AI GPU by NVIDIA Corporation is powerful, its power consumption, enterprise purchase costs, and computing power rental costs are much greater in large-scale inference computing scenarios. Companies such as Microsoft Corporation, Amazon.com, Inc., Alphabet Inc. Class C, and MetA, without exception, are collaborating with Broadcom Inc. or Marvell to develop AI ASIC chips for massive inference edge computing deployment. For example, Alphabet Inc. Class C partnering with Broadcom Inc. to create the TPU (Tensor Processing Unit) is one of the most typical AI ASICs.Looking ahead, the future prospects of computing power, NVIDIA Corporation's AI GPU may focus more on large-scale cutting-edge exploratory training, rapidly changing multimodal or new structure rapid testing, as well as general computing power for HPC, graphics rendering, visual analysis, etc. AI ASICs, on the other hand, will focus on extreme optimization of deep learning specific operators/data flows, specializing in stable structural inference, high-throughput batching, and high energy efficiency. In the long term, both will coexist harmoniously, with the possibility of a significant expansion in the AI ASIC market share in the short to medium term. NVIDIA Corporation's general GPU will focus on complex and varied scenarios, while Research Frontiers Incorporated will focus on high-frequency stable, large-scale AI inference workloads as well as a portion of mature stable fixed training processes.