The emergence of China Depseek in the artificial intelligence (AI) sector with its low -cost model has triggered a wave of services of large Chinese technology, threatening the supremacy of leaders such as Openai and forcing a reconfiguration of the sector.
Some of these services have shown capacities similar to Openai chatgpt, according to specialized portals, at a smaller price, which has intensified competition in the industry.
Constant news
In March, the Baidu Technology, known as the ‘Chinese Google’, presented its ‘chatbots’ Ernie 4.5 and Ernie X1, with approximately 50% lower prices than those established by the DEPEEEK R1, which had already set very small rates compared to those of Chatgpt.
For its part, the Digital Giant Tencent recently presented its new reasoning model, Hunyuan T1, which competes in capacity and cost with the Deepseek R1.
Tencent stressed that the service offers “reliable results” and a “low frequency of hallucinations”, in reference to the information that the ‘chatbots’ invent, a problem that has brought the sector head.
Other prominent technological companies such as Bytedance, owner of Tiktok, or Alibaba have also presented new editions of their ‘chatbots’ in recent weeks.
Open source model
Most of these companies have opted for open source models, an applauded strategy by Chinese authorities.
In March, the spokesman for the Chinese legislature, Lou Qinjian, praised Deepseek’s open model, comparing its evolution with that of the telecommunications industry, mentioning how companies with restrictive strategies have lost relevance.
“Symbian used to dominate with a closed model and declined, while Android, through open source, obtained market recognition,” he exemplified.
Thus, Alibaba, for example, has released part of his Qwen model, while Bytedance has done the same with his Doubao platform, following the carest of Deepseek.
The analyst Grace Shao, cited by the Hongkoniest newspaper South China Morning Post, explains that “Chinese entrepreneurs born in the 80s and 90s wish to demonstrate to the world that Chinese companies can innovate, instead of simply ‘copy'”, which has “spurred the open source approach.”
According to Shao, for these entrepreneurs “it is more exciting to be summoned or used by developers and companies outside China than to make money with a single project.”
This approach could influence Openai’s recent decision to launch an open source language model, which is a change of course for the company, which had traditionally opted for restricted code technologies, limiting the ability of third parties to modify its base structure and adapt it to specific needs.
Reduced development cost
According to Deepseek, its R-1 model was trained for 55 days with a budget of 5.57 million dollars (5.33 million euros), using NVIDIA H800 graphic processors, a reduced capacity version.
This cost represents less than a tenth of the expenditure in the training of the 4O of OpenAi model, according to the Chinese press.
Some Chinese companies, such as Tencent with their Hunyuan model, have also highlighted the optimization of memory use, one of the biggest challenges faced by the sector for the high energy demand that these services entail.
This optimization could question the business model of companies like NVIDIA, which fell in the stock market up to 12% shortly after Depseek’s emergence.
Analysts of the Assets Manager Janus Henderson explained that “a potentially more efficient approach to AI processing questions a multimillion -dollar investment in infrastructure and intellectual property.”
They added that “the high growth of the expected profits has been used to justify very high valuations, which leaves investors very exposed to any disappointment.”
Uncertainties persist
However, some representatives of the sector have been skeptical of the emergence of so many language models.
The CEO of the Minimax company, Yan Junjie, predicted to the local press that “only five companies worldwide, and I do not know how many they will be in China, they will develop models in the future”, while urging the industry to “improve their precision before playing a role in more traditional sectors.”
In addition, there are doubts about the real global expansion capacity of the Chinese ‘chatbots’ for the censorship exerted by the authorities.
China regulated them in 2023 to respect “the fundamental socialist values”, which are prohibited “to generate content that threaten national security, territorial unity and social stability.”
Thus, many Internet users have been able to see how these services refuse to answer questions about the 1989 Tiananmen massacre or if Taiwan is part of China.