Ten thousand more than the price a year ago! H20 has become the favorite of "big factories" NVIDIA Corporation (NVDA.US) is the biggest winner of DeepSeek's hot trend?

From the sudden drop in the stock price of NVIDIA Corporation to the surge in sales of its H20 chip, the connection between NVIDIA Corporation (NVDA.US) and the top player in the tech world, DeepSeek, is becoming increasingly closer. "DeepSeek is a great innovation, reasoning brings huge computational demand." At NVIDIA Corporation's latest quarterly earnings call, Huang Renjun praised DeepSeek. Indeed, this is the case. Thanks to its data center business, NVIDIA Corporation delivered impressive results for the fourth quarter of 2025. The head of the computing resource information sharing platform Berlin Cloud told Caixin reporters that there has been a significant change in demand for H20 recently, with inquiries being three times higher than before, mainly due to the high cost-effectiveness of deploying the full version of DeepSeek. At the same time, this person showed Caixin reporters nearly 10 WeChat groups set up for connecting to H20 demand in recent weeks. According to interviews with various sources, due to cost-effectiveness and compliance considerations, the H20 GPU, which was once not favored by the market, has become "popular among major manufacturers." The price of an 8-card H20 machine has increased by 100,000 yuan compared to last year, and some industry insiders even predict that "the price should not go down." TrendForce analyst Gong Mingde, in an interview with Caixin, reported that from the observation of the AI server market, the main GPU suppliers in the Chinese market are still NVIDIA (H20) and others. In fact, last week, Caixin reporters learned from multiple interviews about the supply-demand situation of the AI computing power market under the fervor of DeepSeek. Since the start of the year, the consultation volume for renting and purchasing computing power equipment has increased significantly, with demand still focused on NVIDIA Corporation's H-Series GPUs, along with the 4090 and A100 models. In terms of the choice of these popular GPUs, a distributor of NVIDIA Corporation's GPUs told Caixin reporters, "H20 is the mainstream, followed by A100, with cost-effectiveness coming next, and the 4090 is for 'modified cards' (consumer cards used for professional computing), considering all factors, it is not very suitable for enterprises." Another distributor bluntly stated, "After DeepSeek became popular, the model with the most increase in demand was H20." In terms of pricing, Caixin reporters learned from a distributor that the H20 series has experienced the largest increase in prices among the computing cards. "Currently, an 8-card machine is priced at around 1.1 million yuan, which is more than 100,000 yuan higher than before." "The price should not go down, because large manufacturers are adopting DeepSeek reasoning, ordering a large number of H20 cards, with companies like Tencent ordering 100,000 to 200,000 cards." This person said. He believes that due to cost-effectiveness and compliance considerations, H20 is now "the favorite of major manufacturers." Just this week, there have been reports citing insiders that orders for H20 from Tencent, Alibaba Group Holding Limited Sponsored ADR, and ByteDance have significantly increased. The distributor mentioned earlier also confirmed to Caixin reporters that the demand for H20 from major manufacturers has indeed been strong under the fervor of DeepSeek. While the increase in demand for H200 may not be as significant as that of H20, it is still a key product for NVIDIA Corporation. The distributor mentioned earlier stated that the demand for H200 has not changed much in the past few months, and the current spot price is around 2.3 million yuan, with a futures delivery time of about a month, similar to before the new year. A professional from a server company admitted to reporters, "After DeepSeek came out, large models have transitioned from training to reasoning, and the cost of computing power has significantly decreased. There are even reports that many computing power projects that were previously planned have been put on hold because they no longer need such strong computing power after equipping DeepSeek." However, according to a client demand form provided by the head of Berlin Cloud to Caixin reporters, the clients they have been in touch with are mainly focused on NVIDIA Corporation's H200. A professional from a server company admitted, "After DeepSeek came out, large models have transitioned from training to reasoning, and the cost of computing power has significantly decreased. There are even reports that many computing power projects that were previously planned have been put on hold because they no longer need such strong computing power after equipping DeepSeek." However, when asked whether the demand comes from major manufacturers, the head of Berlin Cloud admitted to Caixin reporters that while the demand for H20 is strong, the demand side they are in touch with generally does not disclose specific information such as company names, "Smart computing projects are private, and there are too many middlemen. The demand side also hopes to find a source of goods more efficiently, so there are more inquiries and comparisons." The head of Berlin Cloud further explained that matching supply and demand is crucial, for example: "Many demand sides require machines to be deployed in designated rooms, but the rental side already has stable rooms and operation and maintenance support, so it is difficult to move machines to other designated rooms for deployment." Compared to the current popularity of NVIDIA Corporation, an industry insider said that domestic cards have not reached a level where they are in high demand. However, NVIDIA Corporation is not without pressure. Wu Quan, founding partner of Huaxin Jintong Semiconductor Capital, told Caixin reporters, "DeepSeek greatly enhances the confidence of local technology, upgrades the underlying elements of domestic economic and social development, greatly promotes the efficiency of new quality production factors, and various industries are showing a burst of momentum in large models and AI applications." "It has also brought opportunities and play for the entire AI computing industry, including domestic chips. As a key component and link, we believe that domestic cards have the confidence and ability to make new breakthroughs." "In terms of AI chips, in the future market, GPUs and ASICs will coexist. Cloud service operators will not only adopt NVIDIA GPUs but also actively develop their own ASIC chips, which can reduce their reliance on NVIDIA chips and lower costs," said TrendForce analyst Qiu Peiwen. Currently, many domestic computing power manufacturers such as Huawei Ascend, ... (the text continues)Kunlun core, Cambrian, etc., all use ASIC solutions.Gong Mingde also observed that some large Internet companies and chip design companies are actively promoting self-developed AI chip solutions. He predicts that in the next three to five years, with the continuous impact of international changes and China's encouragement of domestic chip production, the trend towards the self-reliance of AI chips will gradually strengthen, especially in the field of Edge AI applications in various industries. This article is reproduced from "Cailian News", edited by GMTEight: Xu Wenqiang.