The author is the founding partner of VC in Hong Kong. capital
Deepseek always changes the global competition in technology. In China, founder Liang Wenfeng has become a local champion. For an overseas degree, especially in the United States, it is still considered to be famous than their families. Students and parents are stunned and found that the research team of his artificial intelligence startup has been educated in China.
Beijing is more confident than ever. Deepseek's success destroys obstacles caused by the US-China Science and Technology War.
This company headquartered in Hangzhou decided to release low cost, open source AI model, and detailed disclosure of their training methods, which means that from St. Paulo to Stockholm's startups and Nairobi doctors, everyone can visit State-ART AI has almost no cost.
In the Chinese start -up department, there is a chain reaction. Create a new AI application. Competition will become more intense. The risk demand for early risk investment is increasing. DeepSeek's decision to pursue open source AI models inspires people's hearts and put pressure on others. The first response was Alibaba's Qwen team. The team released Qwen2.5 as open source on the eve of the Lunar New Year last month.
This is a great change. At the end of 2022, after the US OpenAI released the AI model ChatGPT it generated at the end of 2022, the global digital economy was gradually controlled by a few technical giants. These participants are pursuing more than efficiency-to establish more and more models. These models need amazing calculations, energy and capital, while guarding their own training methods as business secrets.
The centralized closed model creates a dangerous feedback circuit. The more data they have accumulated, the stronger they become, and the more marginalized anyone. For consumers, this means a lot of costs, surrender data and watch the future of AI without meaningful participation.
The promise of DeepSeek reasoning R1 model lies in its adaptability. Open sources can be tailored according to local needs. It avoids the use of so -called sparse neural network training to avoid redundant computing, which means that its efficiency can reduce computing and energy requirements through order order.
This means that advanced AI can benefit from the masses, not just a few people. It proves that this technology is a product. There is no need to waste the competition between billions of dollars in technology giants and closed models. The value of artificial intelligence should not be in the proprietary model, but that all of us can use it.
As an investor, I worry that the prominentness of DeepSeek may lead to stricter sanctions in the United States. In China, the export restrictions of GPUs such as NVIDIA's powerful H100 and other graphics processing units (GPU) have hindered the start of the start. Due to compliance risk problems, foreign investors' funds are restricted. However, the real danger is to limit the opportunity to gain global education and research cooperation, which stifle the global knowledge flow that is vital to maintaining progress. Talents can avoid shortage of chips, but have established a long -term stagnation of obstacles to learning risks.
However, even other restrictions on the United States, conspiracy theories and coating campaigns for Deepseek cannot change the reality of Chinese start -ups in the hands of human beings.
In all noise, let us see it as a historical moment. In 1440, Johannes Gutenberg brought a printing machine to Europe, which broke the monopoly of previous elite knowledge. DeepSeek's achievements have added this tradition, even if information is easier to access. Its low -cost reasoning model proves that artificial intelligence can belong to everyone, not just the HO accumulation regulations, chips and capital.