Domestic GPU, the bigger the storm, the more expensive the fish

Giant Tide WAVE

2026.11.19 18:58

· From Beijing

Full text 4511 words

Text | Juchao WAVE, author | Degree, editor | Yang Xuran

Jen-Hsun Huang, a leader in the Chinese technology and business community who has been in charge of Nvidia for 33 years and promoted it to become the world's largest company by market capitalization, resigned from his positions as president, CEO and director due to unexpected events.

On NVIDIA's official website, Huang Jenxun has been removed from the management sequence and placed in the "Founder" column together with Chris Malakovsky. Even on the X platform, his Blue V certification and CEO title have disappeared.

The era when Huang Renxun controlled this global AI computing power hegemon has ended in a substantial sense. Even if it finally returns for some reasons, Nvidia's strategy at the global computing power level will inevitably undergo significant changes.

On November 12, after the news was announced, Nvidia's stock price fell by more than 13% intraday, and the A-share Nvidia industry chain immediately came under pressure. However, on the other side of the coin, Nvidia's old rival AMD, as well as domestic GPU companies such as Cambrian, Suiyuan Technology, and Muxi Technology, have begun to receive more market attention.

As a global Chinese technology leader, Huang Renxun is in charge of the most critical company in the world in the AI ​​era, but he can only resign from all management positions in such a short period of time. At the same time, it coincides with the US mid-term elections, and the political and economic game has become extremely complicated.

The departure of a founder will not make the technical moat that Nvidia has accumulated for many years disappear. In other words, there will be no obvious problems in its production and supply. The problems will arise with those customers who have clear needs for computing power and are in the international trade game - that is, Chinese AI companies.

Text | Juchao WAVE

Huang Renxun's personal withdrawal has posed a more stringent torture to China's domestic GPU, because supply risks have been completely exposed. China's AI industry will increasingly realize that "autonomous and controllable" has a long road ahead, but it is already a top priority that must be paid attention to.

reverse

GPU is the underlying hardware for large model training and reasoning and computing clusters, which is equivalent to the "heart" of the AI ​​industry.

Since October 2022, the United States has directly included high-end chips such as NVIDIA A100, H100 and AMD MI300 in the control list. Since then, from A800, H800 to H20 to H200, NVIDIA's "special edition" chips have actually been restricted from being exported to the Chinese market.

By the first half of this year, the Chinese market's contribution to NVIDIA has been significantly reduced, and its share of the Chinese market has dropped to 37%, surpassed by its domestic peers. Of course, this does not represent NVIDIA's share of all the computing power available to Chinese AI companies, because there are also institutions and equipment that provide computing power to Chinese companies outside the Chinese market.

At present, domestic GPUs have passed the "can we make products" stage and have begun to enter the large-scale delivery period. According to the third quarterly report of 2026, the revenue of core manufacturers has generally increased several times, and the quality of profits has improved significantly.

Cambrian and Haiguang Information are the first two companies to enter a positive profit cycle and lead in scale.

In particular, Cambrian's net profit margin in the first three quarters was close to 39%, with a net profit of 3.7 billion yuan, indicating that it has passed the critical point of life and death and has real profitability. Haiguang Information is the largest, with revenue in the first three quarters approaching 15 billion yuan and net profit of 2.8 billion yuan.

"The No. 1 domestic GPU stock" Moore Thread's revenue has increased rapidly, and its profits have passed the critical point of profit and loss. In the third quarter, revenue reached 1.552 billion yuan, a new single-quarter high, and non-net profit after deduction turned positive for the first time to 20.07 million yuan. However, in the first three quarters, it still suffered a cumulative loss of 131 million yuan after non-compliance deductions, with a net operating cash outflow of 3.309 billion yuan.

Moore Thread takes the route of full-featured GPUs. Its products cover AI computing, graphics rendering, scientific computing and video encoding and decoding. It has a lot of room for imagination, but R&D and ecological construction also eat up cash.

Muxi's net profit attributable to the parent company in the first three quarters was 639 million yuan. After deducting significant changes in fair value, non-net profit was 259 million yuan. Moreover, the company has achieved non-profit for two consecutive quarters, indicating that revenue has begun to cover costs such as research and development.

Suiyuan Technology, which was listed later, saw revenue surge 3.3 times in the first three quarters, but it is still in a high-investment stage. The single-quarter loss in the third quarter has narrowed to 122 million yuan, and there is still a way to achieve sustained profitability. However, due to the continuous surge in stock prices, its market value has surpassed Moore Thread and Muxi shares, which are more advanced in the industry.

Hong Kong-listed Biren Technology's revenue increased nearly 20 times in the first half of the year, its gross profit margin increased to 42.7%, and its losses narrowed significantly. However, the stock price has retraced nearly 70% from its high point. To achieve sustained profitability, the company needs to cross more hurdles such as customer repurchase, software adaptation, and delivery.

Tianshu Zhixin, another Hong Kong-listed company, achieved profitability in the first half of the year, with revenue of 946 million yuan, nearly three times that of the same period last year. Its adjusted net profit was 246 million yuan, turning losses into profits for the first time since its listing. However, its profit mainly comes from an equity investment income of 760 million yuan. After excluding it, the main business still loses money.

It can be seen that domestic GPUs have entered a period of rapid growth, the income of domestic players has increased significantly, and profits have begun to improve. Some leading companies have even entered sustained profitability, and the operating status of domestic AI chip companies is much better than before.

Game

Huang Jen-Hsun's departure comes during the sensitive period after the U.S. midterm elections, further disrupting the stability of the global AI supply chain.

In the past few years, Huang Renxun has long advocated relaxing chip exports to China based on Nvidia's business interests. After the A100 and H100 were restricted from being exported to China, NVIDIA immediately launched the "reduced version" of the A800‌ and ‌H800‌. After the A800 and H800 were included in the banned list, the H20 was quickly released.

He also used his personal reputation, status and influence to mediate between Congress and the Department of Commerce. In the U.S. discussion on chip export controls to China, Huang Renxun is also one of the clearest and most influential opponents.

Anthropic founder Dario Amodei is a representative of the "blockade faction" in Silicon Valley. He has long advocated strict restrictions on China's access to advanced American AI chips in a tough stance. He even stated that selling H200 chips to China was a "huge mistake" and "as crazy as selling nuclear weapons to North Korea."

Later, Huang Renxun responded in a podcast that directly comparing AI chips to nuclear weapons was "too crazy" and the logic of export controls was "extremely stupid" and a "loser mentality."

To a certain extent, Huang Renxun is a "cushion" in the chip game between China and the United States. His departure is equivalent to the disappearance of the most influential voice opposing U.S. chip export controls. In the internal debate in the United States, there is one less heavyweight checking force.

After Huang Renxun resigned, US House Speaker Johnson said that the consensus within Congress on restricting the export of semiconductor chips to China is "continuously expanding." At the same time, the U.S. Department of Commerce is also studying new chip export control rules and plans to include Southeast Asia and other countries in stricter export licensing management to block transshipment routes.

Some reports also show that when Huang Renxun was in office, Nvidia would still have some fuzzy room for operation when selling to China, and some Chinese data center projects could obtain certain Nvidia chips. Contracts for these projects have generally been signed many years ago, and some have already started construction.

Once the control measures are tightened again after Huang Renxun's departure, the "twilight zone" where these projects were originally operable will be completely interrupted, the projects may be shelved, or they may be forced to redesign the system at a higher cost.

The drastic changes in the external environment have directly blocked many large domestic models from classifying Nvidia chips as "high-performance chips." Domestic GPUs serve as a supplementary strategy, forcing the entire industry to turn to endogenous alternatives.

If the use of NVIDIA chips is subject to more restrictions, domestic chips will become the "primary choice" by default, and the "replacement window period" for domestic computing power will accelerate.

‌For example, DeepSeek's next-generation large model V5 will rely on Huawei Ascend for the first time to complete pre-training. Previously, Ascend was only responsible for inference or post-training. DeepSeek V3 and V4 both completed training on NVIDIA chips and ecosystem.

If successful, it will mark the shift of domestic chips from the inference side to the more difficult training side, and have the ability to support the full-process development of top large models.

Go ahead

At the Huawei Full Connectivity Conference on September 17, Xu Zhijun, Huawei’s rotating chairman, mentioned: “It is difficult to measure NVIDIA’s market share in China, and from the data we can collect, Ascend should have surpassed NVIDIA.”

When NVIDIA was caught in the double attack of a management vacuum and congressional investigations, domestic GPUs developed freely at the " stall " where AI demand exploded and international giants' exports were blocked.

Domestic GPUs have now formed a pattern of "one super (Huawei Shengteng) + multiple strong ones (Pingtou Ge, Cambrian, Kunlun Core, etc.)". According to IDC statistics, in the first half of this year, Nvidia shipped 678,000 units in China, a 44% year-on-year decrease, while Huawei's shipments surged to 572,000 units. The market share difference between the two parties was only 5.3%.

Alibaba Pingtou Ge shipped 133,000 yuan, ranking third, followed by Cambrian, Baidu Kunlun Core, Haiguang Information and Moore Thread, with shipments ranging from 50,000 to 100,000.

Some institutional analysts believe that Huawei's Ascend share in the Chinese market has reached half, while Nvidia may only have about 8% left. This is a huge reversal in four years compared to 2022, when Nvidia alone accounts for 95% and Ascend only accounts for 3%.

Morgan Stanley estimates that China's AI chip self-sufficiency rate will be 35% in 2025, 45% in 2026, and will jump to 68% in 2027, 77% in 2028, and 95% in 2030, basically achieving complete independent replacement.

The turmoil in Nvidia’s management has further accelerated the localization process. Just five days after Huang Renxun left office, Nvidia established a new National Security and Export Control Oversight Committee, which is responsible for supervising the company's national security policies, export control compliance and global trade at the board level.

The United States also quickly began investigating Singaporean company Megaspeed, accusing it of illegally transferring Nvidia chips to China. Foreign media also reported that the former senior executive of the Singaporean company may have a personal relationship with Huang Jen-Hsun.

Judging from various signs, the window for purchasing overseas high-end chips has become smaller and smaller, and further localization of GPUs is imperative.

Compared with NVIDIA, domestic manufacturers still have gaps in advanced processes, single card performance, HBM supply, ultra-large-scale training stability, and system ecology.

In particular, after long-term accumulation of CUDA, the ecological rainforest is extremely prosperous, and domestic chips have to face huge migration costs. The reason why Nvidia has become the world's AI chip overlord is that the core foundation lies in the moat of CUDA. Developers are accustomed to CUDA, which has become the biggest threshold in front of new chips such as Ascend, but the gap in the hardware itself is not the first.

However, domestic manufacturers are already looking for new solutions.

Liao Heng, chief scientist of Huawei Semiconductor, said frankly that the first obstacle encountered by Shengteng was not hardware, but mainly ecological barriers. Today, with the emergence of new compiler languages, the barriers to CUDA software are gradually dissolving, and cutting-edge AI laboratories no longer attach as much importance to CUDA as they did two years ago.

In terms of performance, since a single card cannot beat it, thousands or even tens of thousands of cards can be assembled into a supercomputer to improve performance in the form of a cluster. This is the "super node" route that Huawei and Alibaba are currently betting on.

As Xu Zhijun pointed out, Huawei is leading the world in overall computing power through the systematic innovation of "super node + cluster".

Huawei has released the world's first super node Ascend 960 based on NPO technology, which can be expanded to 4096 cards; Alibaba launched the first SNPO optical interconnection super node Panjiu AL64 at the Yunqi Conference. Sugon, Ziguang, Muxi, etc. also joined in succession.

Just like DeepSeek's new approach, this kind of innovation in engineering and system makes it possible for Chinese chips to catch up with Nvidia in terms of performance.

write at the end

"The leather jacket remains, but the crown is lost."

Huang Renxun was once the "buffer zone" for Nvidia's trade, quotas, and even gray re-export channels with China. He completed the coordination of the conflict between China and the United States in computing power in a relatively quaint and somewhat gray way.

However, after the change of leadership, Nvidia and even more American technology companies will most likely continue to tighten their control over China. This will inevitably push China's AI industry chain to completely abandon the fantasy of "taking a detour to get cards" and fully invest in its independent route.

From a longer time perspective, Nvidia's coaching change is not the starting point, but an accelerator. As Huang Renxun has emphasized many times, restricting U.S. chip exports can only force China to achieve more independent innovation.

Today, the consensus in China's AI chip industry has been further strengthened, and it is only a matter of time before full localization.

Respect for the original work, if you are qualified, you should be appreciated.

According to the real-name requirement of the Cybersecurity Law, please bind your mobile phone number before posting a comment.

Express

2026-11-19 23:11

Director Zhu Lidan sues Wahaha Commercial Co., Ltd.

2026-11-19 23:02

Domestic commodity futures markets mostly closed higher at night, with black products leading the gains.

2026-11-19 23:01

International Energy Agency: Middle East conflict may reshape global energy landscape

2026-11-19 22:59

SpaceX plans to issue at least $32 billion in AI chip debt on Thursday

2026-11-19 22:55

Anthropic is expected to open at $99.60 on its first day of trading

2026-11-19 22:52

Livermore’s leading index of Chinese concept stocks expanded to 2%, while Pinduoduo rose nearly 13%.

2026-11-19 22:51

About 15% of shares in Anthropic IPO sold to retail investors

2026-11-19 22:50

APEC Connectivity Blueprint

2026-11-19 22:46

Anthropic's IPO doesn't place shares to majority institutional buyers

2026-11-19 22:45

Wang Yi: The meeting reached the Shenzhen Outlook, which is the most important outcome of the APEC Year in China

JPMorgan Chase says some customer information leaked in cyberattack on supplier

2026-11-19 22:39

The Nasdaq's gains quickly expanded to 2%, the Philadelphia Semiconductor Index rose 4%, and NVIDIA rose more than 4%.

2026-11-19 22:34

U.S. stocks opened: The Nasdaq opened 1.6% higher, storage, optical communications and other sectors rose, and Nvidia rose nearly 2%

2026-11-19 22:29

Yuanji Food passed the Hong Kong Stock Exchange listing hearing

Three land parcels in Beijing received 9.247 billion yuan in land auctions: China Construction Intelligent Land and Swire Properties jointly won the Chaoyang Jiangtai residential land for more than 4.9 billion yuan

2026-11-19 22:23

DeepSeek’s second round of financing lands

2026-11-19 22:17

Wu Yongming: The target of building global data centers exceeding 20GW may be lowered

2026-11-19 22:16

20 quick facts about News Network on November 19

2026-11-19 22:06

BASF announces termination of acquisition of Evonik

2026-11-19 21:58

Chongqing Luqiao: Shareholder Hangzhou Linxin plans to increase its holdings of company shares by 15 million yuan to 30 million yuan

Scan to download App