The ant group focuses on Chinese chips to strengthen its strategy in artificial intelligence.
In a global context, which is increasingly marked with technological competition between the US and China, Ant Group, controlled by AlibabaIt takes important steps to reduce its dependence on American chips and contains costs in the development of artificial intelligence (AI) models.
According to sources close to the company, Ant trusts Semiconduttori cesi Educating advanced language models using an approach that promises to revolutionize the way AI is produced in Asian country.
Strategic turning point in AI model training for ant group
In recent months, Ant Group has adopted chips by local companies, including organizations affiliated to Alibaba. Huawei technologiesTo train using AI models Mixture of Experts (MoeTechnical.
This approach, which is increasingly widespread among researchers, allows the tasks to effectively divide the tasks between the different “experts in within the model and to increase the efficiency of calculation.
Sources cannot be compared with only the results of these models. Nvidia H800 chips. However, in some tests, they even exceeded the performance of the models developed by. Meta.
Although it is Bloomberg News These performances have not independently confirm, data have made significant progress in China’s attempt to reduce operational costs and reduce technological dependence.
MOE technique is inspired by the principle of special delegation: each sub -module of the model is responsible for a certain part of the process that provides more scalability and efficiency than traditional approaches.
In addition to the ant group, Google And Chinese initiative from Hangzhou Deepseek They also use this methodology.
Ant emphasized his devotion to scientific spread by publishing an article emphasizing his goal. Scale Models Without Using High GPUs.
This approach becomes very important for companies that cannot use continuously due to high costs. High performance equipment.
Chinese vs United States: Local chips against American GPUs
The ant attempt fits a geopolitical context that Chinese technology companies are trying to get through. Restrictions on the export of USA developed chipsLike the NVIDIA H800.
Although not the most advanced chip on the market, the H800 is still one of the most powerful GPUs in China.
Although Ant Group is still maintaining some of the AI production based on NVIDIA chips, the company is increasingly changing to more economical and easy accessible alternatives. Like offered AMD and Chinese manufacturers.
This strategic choice leaves the vision of Nvidia’s CEO, Jensen Huangcompanies will continue to demand more calculation power.
According to Huang, customer investments will not decrease even with the emergence of more efficient models. Deepseek R1. Thus, oath.
One of the important points of Ant’s analysis is about a significant decrease in the cost of educational AI models.
According to the document, the basic units used to train and learn a model on a trillion coin are traditionally approximately costly costly 6.35 million yuan (about $ 880,000).
The cost has been reduced by using less performance chips optimized for MOE method. 5.1 million yuan.
A non -marginal savings that can revolutionize artificial intelligence, especially for beginners and developing industrial sectors.
Developed Models, Ling-lite And Ling-PlusDesigned for applications such as contexts. health care And financeTwo areas where AI power can offer concrete and urgent solutions.
In exactly the field of health, ant recently Haodf.comOne of China’s leading online medical platforms. Thus, to confirm its interest in expanding its presentation of artificial intelligence -based solutions.
Among the current services of the company ZhixiaobaoA Virtual Assistant and Financial Advisory Platform Maxiaocai.
“HTML
Opening and future of Chinese artificial intelligence
“”
Another distinguishing point of the ant strategy option to make models open source: Ling-lite 16.8 billion parametersWhile reaching ling-plus 290 billion.
To make a comparison, GPT-4.5Developed model developed by OpenAi, approximately 1.8 trillion parameter. Although it is closed and the public cannot be accessed. Research by Ant is not difficult.
The same work, during training, Small Variations The structure of the models or in the type of hardware Performance in PerformanceLike pointed ends in error rates.
Despite the progress, a structural difficulty emphasizing how it is Most Advanced Models It requires constant attention.
As observed Robin YuThe CTO of Beijing technology company Shengshang Tech, the concrete results obtained in the real world are really important:
“If you find a weak point to beat the world’s best Kung Fu master, you still won.”
It is an effective metaphor that emphasizes the value of practical practices only compared to theoretical comparison.
What is clearly revealed is that the ant group plays an important role in China’s attempt. Technologically more autonomous.
As a result, one more accessible AI, Less dependent on western equipmentand potentially more efficient For the strategic industrial sectors of the future.
The challenge to Western Ai Giants was initiated: not with brute force but Intelligence, productivity and strategic vision.
