English
The storm created by DeepSeek in the field of innovative technology has raised a lot of questions, including how it is possible, based on an investment of only $ 6 million, to create a program aimed at developing Artificial General Intelligence (AGI) and what will be its sustainability.
According to the analysis that we conducted with the help of Google's Notebook LM, it appears that DeepSeek manages to have a low cost compared to the technological giants in the IT market, through the following strengths:
- Efficiency of training linguistic models: DeepSeek launched the R1 linguistic model, which is said to have been developed with a fraction of the money spent by large American companies. This suggests that training linguistic models can be done with fewer video cards. This efficiency reduces the costs required for hardware.
- Smart initial investments: Although DeepSeek launched in 2023, founder Liang Wenfeng began purchasing thousands of Nvidia GPUs as early as 2021, preparing the infrastructure needed for his project. This early strategic approach could have reduced the cost of acquiring the necessary equipment.
- Focus on research: In 2023, Liang's fund announced that it would focus on creating a "new independent research group to explore the essence of AGI" (Artificial General Intelligence). This focus on research, rather than massive marketing or expensive infrastructure spending, could have helped reduce costs.
- Competitive salaries for specialists: Liang uses the proceeds from his investment fund's transactions to pay attractive salaries to the best local AI specialists. This strategy attracts top talent without exceeding the allocated budget.
• Weaknesses of DeepSeek
With this in mind, we asked OpenAI to tell us how DeepSeek can be sustainable. According to Open AI, DeepSeek's sustainability, despite its relatively low declared costs, can be explained by several strategies applied simultaneously. These include DeepSeek's use of agile development methods, small and well-specialized teams, and affordable infrastructures, collaboration with universities or organizations that offer access to infrastructure and resources at low costs, use of existing algorithms or open-source resources, which means low research costs, focus on a Minimum Viable Product (MVP) - DeepSeek could quickly develop a working version of AGI, saving resources for further development - and access to China's vast AI talent pool, which equates to low labor costs for a high level of expertise. Furthermore, Open AI took into account the fact that most companies in this field in China collaborate with the authorities in Beijing and argues that, if DeepSeek has ties to the Chinese government or universities, it is possible that part of the costs were covered indirectly, either in the form of grants or through infrastructure made available for free, especially since in-kind support (access to laboratories, advanced equipment or human resources) is not always reflected in the declared budget.
In light of the latest information above, Google's Notebook LM identified several weaknesses regarding DeepSeek. The first refers to a high risk of cyberattacks and Notebook LM recalls that the Chinese start-up was the target of a "large-scale" cyberattack immediately after launch, which led to the temporary unavailability of the service, indicating a potential vulnerability in its security.
The second weakness is the lack of public data on investments. Although it is known that DeepSeek was founded by High-Flyer - Liang Wenfeng's hedge fund, there is no precise data available on how much this fund has invested in DeepSeek. This lack of transparency can be a disadvantage in terms of trust and credibility.
The last weak point is related to the origin of the company. Although not a technological weakness, DeepSeek's Chinese origin could raise geopolitical and security concerns for some users, especially in the US.
Euro
Dolar SUA
Franc elveţian
Liră sterlină
Gram de aur
