DeepSeek-V4 Released: Chinese Media Says Its Biggest Significance is Domestic Computing Power Replacement
Chinese AI startup DeepSeek has recently released its latest model, DeepSeek-V4. Chinese media highlight that its greatest significance lies in utilizing domestic computing power for inference, driving the development of China's local AI industry chain, with multiple related companies seeing significant stock price increases.
📋 Article Processing Timeline
- 📰 Published: April 26, 2026 at 12:07
- 🔍 Collected: April 26, 2026 at 12:31 (23 min after Published)
- 🤖 AI Analyzed: April 26, 2026 at 16:05 (3h 33m after Collected)
(Taipei, April 26, Central News Agency) China's artificial intelligence (AI) model DeepSeek-V4 was recently released. Chinese media pointed out that its greatest significance is the use of domestic computing power for inference, driving the development of China's local AI industry chain, and the stock prices of many related companies have risen sharply.
Chinese AI startup DeepSeek officially released the latest version of its model on the 24th, divided into two versions by size: DeepSeek-v4-Pro and DeepSeek-v4-Flash.
According to public reports, DeepSeek V4 natively supports Huawei Ascend chips and completes inference entirely based on domestic computing power. DeepSeek-V4 has a super long context of millions of characters, achieving domestic leadership in Agent capabilities, world knowledge, and inference performance in China and the open-source field.
On the same day, China's AI industry chain collectively rose sharply, with Moore Threads, Biren Technology, Iluvatar CoreX, and Hygon Information all achieving increases. SMIC's H-shares rose by 10.01% in a single day, and Hua Hong Semiconductor's H-shares rose by more than 15% in a single day.
Daily Economic News commented on the 25th that the true value of DeepSeek V4 is not in the model itself, but in the supply chain logic behind it; it does not astound the world with its model capabilities, but allows Chinese AI users to call Tokens at low cost. DeepSeek's move this time is an important step to use domestic computing power for inference and stimulate China's AI industry chain.
Beijing News also commented on the 25th that China's most advanced open-source large models have achieved domestic substitution in the inference stage, which is a more noteworthy signal behind the release of DeepSeek V4.
The article said that this substitution has produced two-way value. On the one hand, the large-scale adoption of domestic large models provides real stress test scenarios and continuous demand pull for domestic computing power chips; the improvement of computing power supply capabilities, in turn, creates conditions for the large-scale deployment and price reduction and popularization of models.
DeepSeek clearly stated in its release that, limited by high-end computing power, the service throughput of V4-Pro is very limited, and it is expected that after the mass listing of Ascend 950 ultra-nodes in the second half of the year, the price of V4-Pro will be significantly reduced.
The commentary said that there is still a gap between the capabilities of DeepSeek V4 and international top closed-source models. The large-scale application of domestic computing power on the training side is still in its infancy, and there is still a considerable distance from individual cases to industry practice. (Editor: Zhang Shuling / Zhou Huiying) 1150426
Chinese AI startup DeepSeek officially released the latest version of its model on the 24th, divided into two versions by size: DeepSeek-v4-Pro and DeepSeek-v4-Flash.
According to public reports, DeepSeek V4 natively supports Huawei Ascend chips and completes inference entirely based on domestic computing power. DeepSeek-V4 has a super long context of millions of characters, achieving domestic leadership in Agent capabilities, world knowledge, and inference performance in China and the open-source field.
On the same day, China's AI industry chain collectively rose sharply, with Moore Threads, Biren Technology, Iluvatar CoreX, and Hygon Information all achieving increases. SMIC's H-shares rose by 10.01% in a single day, and Hua Hong Semiconductor's H-shares rose by more than 15% in a single day.
Daily Economic News commented on the 25th that the true value of DeepSeek V4 is not in the model itself, but in the supply chain logic behind it; it does not astound the world with its model capabilities, but allows Chinese AI users to call Tokens at low cost. DeepSeek's move this time is an important step to use domestic computing power for inference and stimulate China's AI industry chain.
Beijing News also commented on the 25th that China's most advanced open-source large models have achieved domestic substitution in the inference stage, which is a more noteworthy signal behind the release of DeepSeek V4.
The article said that this substitution has produced two-way value. On the one hand, the large-scale adoption of domestic large models provides real stress test scenarios and continuous demand pull for domestic computing power chips; the improvement of computing power supply capabilities, in turn, creates conditions for the large-scale deployment and price reduction and popularization of models.
DeepSeek clearly stated in its release that, limited by high-end computing power, the service throughput of V4-Pro is very limited, and it is expected that after the mass listing of Ascend 950 ultra-nodes in the second half of the year, the price of V4-Pro will be significantly reduced.
The commentary said that there is still a gap between the capabilities of DeepSeek V4 and international top closed-source models. The large-scale application of domestic computing power on the training side is still in its infancy, and there is still a considerable distance from individual cases to industry practice. (Editor: Zhang Shuling / Zhou Huiying) 1150426