Summary of Key Points
DeepSeek, a large-scale model company known for its "low cost and high performance," announced a significant increase in API prices following the release of the V4 Flash version. Behind this price hike are both strategic considerations to "give competitors a chance to survive" within the industry and practical pressures: the parent company, Huanfang Quantization, experienced a substantial decline in performance in July, leading to tighter funding; coupled with a global surge in AI usage, which has caused a shortage of computing power and increased costs. This event marks a shift in China's large-scale model industry from a phase of "burning money to gain market share" towards commercial exploration. Companies like DeepSeek are beginning to seek financing and expand their use cases in an effort to find sustainable profit models.
1. Is the Price Hike Still Worth It? DeepSeek’s Cost-Effectiveness Remains Uncompromised
How competitive were DeepSeek’s prices before? For example, the cost of processing the same tasks with the latest V4 Flash is one-fortieth that of GPT-5.6sol and one-hundredth that of Claude Fable5. Even with this significant price increase, industry insiders believe its cost-effectiveness remains attractive. A research report from Zheshang Securities states that even if prices rise by 150%, they are still cheaper than foreign models and could help boost the overall pricing power of domestic open-source models.
Why aren’t companies concerned about the price hike? Many companies use a combination of models: simple, high-volume tasks (such as customer service responses and data organization) are handled by DeepSeek, while more complex tasks (like in-depth analysis and creative writing) use GPT or Claude. After the price increase, DeepSeek continues to play a role in providing cost-effective solutions, with minimal impact on corporate costs.
2. The Financial Pressures Behind the Price Hike: Huanfang’s Losses and Insufficient Computing Power
DeepSeek has been able to maintain low prices thanks to funding from its parent company, Huanfang Quantization, which owns a supercomputing center with thousands of GPUs, providing low-cost computing power. However, in July, Huanfang’s performance plummeted, with nine of its main products experiencing a decline of over 20% in monthly revenue, and its management scale shrunk from 80 billion to 50 billion yuan, leaving it unable to support DeepSeek as before.
More importantly, the cost of computing power is rising sharply: global AI usage is increasing by 20% weekly, and China has surpassed the United States for 15 consecutive weeks. After the launch of DeepSeek V4 Flash, demand for computing power increased by 570%. The supply of chips is struggling to keep up with demand, with the U.S. restricting the export of high-end NVIDIA GPUs. In China, the growth in computing power supply (128%) laggs far behind demand (417%), and the rental price of NVIDIA H100 has increased by 40%, while that of B200 has doubled. DeepSeek must raise prices to cover its costs and maintain service quality.
3. Is the Price Hike a Strategic Move or a Compromise with Reality? Both Possibilities Make Sense
DeepSeek’s actual controller, Liang Wenfeng, is nicknamed "Liang Sheng" for his commitment to low prices and open-source technology, stating that AI should be shared with others. Some speculate that the price hike is a deliberate attempt to give competitors a chance to survive, as DeepSeek’s previous low prices made it difficult for them to compete. Others believe it’s due to practical constraints: Huanfang’s financial setbacks and rising computing power costs have forced DeepSeek to raise prices to meet its goals.
However, with costs increasing significantly and the need for more affordable computing power, this move is likely a necessity rather than a strategic choice.
4. Not Just DeepSeek: Chinese Large-Scale Models Are All Facing Challenges and Seeking Self-Sufficiency
DeepSeek is not alone in these changes. Other companies like Kimi and Zhipu are also taking similar actions:
- Accelerated Financing: DeepSeek has launched two rounds of financing this year ( totaling 10 billion yuan), while Kimi has raised four rounds in half a year and is preparing to go public. They now need to prove their profitability by seeking external capital.
- Business Expansion: Kimi has released an office collaboration tool called Kimi Work, and DeepSeek has developed an enterprise AI platform and invested in avatar intelligence companies, transitioning from simply selling model APIs to becoming comprehensive AI service providers.
- Localizing Computing Power: Companies are trying to use domestic chips (such as Huawei’s Ascend), but pre-training still relies on NVIDIA. Although domestic chips offer slightly lower performance and less mature software ecosystems, this is a long-term necessity.
These developments indicate that China’s large-scale model industry has moved beyond the stage of "burning money for technology" and must now face the realities of limited computing power, the need for capital returns, and fierce competition. They must find profitable business models to survive.
In Conclusion
DeepSeek’s price hike is not an isolated incident; it signals a shift in China’s AI industry from idealism to practicality. No matter how competitive or passionate the industry was in the past, companies must adapt to the current reality and learn to generate revenue to sustain themselves.