Ant Group (Alibaba) achieves AI breakthrough: cheaper models through innovative chip combination

Published on: March 24, 2025 / update from: March 24, 2025 - Author: Konrad Wolfenstein

Ant Group (Alibaba) achieves AI breakthrough: cheaper models through innovative chip combination-Image: Xpert.digital

China's technological breakthrough: AI models are 20 % more cost-efficient

Trade conflict drives technological progress in AI chips

The Ant Group, a subsidiary of the Chinese technology giant Alibaba, has announced a significant breakthrough in the development of AI models. With an innovative combination of semiconductors from Chinese and US production, the company succeeds in training its AI models by 20 percent more cheaply. This development could have far-reaching consequences for the global AI chip market and strengthen China's position in the technology competition. It is particularly noteworthy that the ANT Group relies on performance weaker but efficient alternatives to the high-priced Nvidia chips without having to accept significant quality of quality. Against the background, this happens between the USA and China and represents an important step towards greater technological independence.

Suitable for:

Day 1 after the AI earthquake on the stock exchange: Nvidia-Crash, German tech shares in need-what does that mean for your investments?

The innovative chip strategy of Ant Group

The solution developed by Ant Group is based on a clever mix of different semiconductor types. Instead of relying exclusively on the powerful but expensive NVIDIA H800 chips, the company now increasingly uses more cost-effective alternatives from AMD and Chinese manufacturers such as Alibaba itself and Huawei. The development of this alternative chip strategy was significantly accelerated by the US export restrictions for high-performance chips to China.

For its AI training models, the Ant Group uses the “Mixture of Experts” (MOE) approach, a technique that enables large language models to train more efficiently. With this method, tasks are divided into smaller data records in order to process them more efficiently - “similar to the use of a team of experts, which focuses on a certain part of the problem in order to improve the overall efficiency”. This approach has already proven itself in other companies such as Google and the Chinese startup Deepseek.

Technical performance and comparison with Nvidia

Despite the switch to supposedly weaker chips, the Ant Group claims to achieve results that are comparable to those of Nvidia hardware. According to its own statements, the company was able to reduce the cost of training from a trillion token from 6.35 million Yuan (around 810,000 euros) to 5.1 million Yuan. This remarkable cost reduction with the same performance could increase the pressure on NVIDIA as a dominant provider of AI chips.

The Ant Group has recently published a research work in which it claims that its models in certain benchmarks can even exceed the performance of models from other large technology companies such as META. Especially in the area of the Chinese benchmarks, the own language models Ling-Lite and Ling-Plus are intended to do better than comparable models from Deepseek.

Economic effects and cost optimization

The cost optimization through alternative chip strategies is becoming increasingly important in the AI industry. While the prices for end customer services such as Chatgpt remain relatively stable, the costs in the B2B range decrease continuously. Alibaba itself had already reduced the prices for various AI models last year, including an 85 percent price reduction for her vision length of QWen-VL.

These costs are part of a larger trend in the AI industry, which is powered by various factors:

The high competition between different providers, both in China (Alibaba, Tencent, Baidu, JD, Huawei and bytedance) as well as global
The development of smaller, more cost -effective models that are becoming increasingly powerful
New methods such as model distillation, in which the knowledge of a large model is transferred into a smaller one
The cost pressure through numerous open source or freely available models such as Metas Llama family

The growing importance of cost-effective AI models

The breakthrough announced by the Ant Group comes at a time when the efficiency of AI models increasingly focuses on. In January, the Chinese start-up Deepseek caused a stir when it presented a AI model that was trained with significantly lower resources than previous models. This development even led to a dramatic decline in the Nvidia share price.

However, Nvidia CEO Jensen Huang argues that the actual need for arithmetic capacity is not created during training, but rather when generating the answers. Nevertheless, the Ant Group's approach shows that considerable cost savings are also possible in the area of AI training.

Suitable for:

Simply explained AI models: Understand the basics of AI, voice models and Reasoning

Geopolitical meaning and technological sovereignty

The development of the Ant Group has a significant geopolitical dimension. The strict US export restrictions on high-performance chips have increasingly prompted Chinese companies to seek alternatives to American technology. In December 2023, the US Minister of Trade Gina Raimondo made it clear that AI chips should not be delivered to China at the current state of the art and that Chinese industry should not be enabled to produce it itself.

In this context, the ANT Group's chip strategy is an important step towards greater technological independence. Robert Lea, Senior Analyst at Bloomberg Intelligence, noted that, if confirmed To master export controls on Nvidia chips ”.

Effects on global technology competition

China not only makes great efforts in the area of electric cars, but also at AI to get to the world market leader. The innovative approach of the Ant Group could serve as a model for other Chinese companies that have to operate in an increasingly complex geopolitical environment. The ability to access different technologies flexibly could prove to be a decisive competitive advantage.

The recent developments show that China made significant progress in the AI area despite the US restrictions. Similar to the Ant Group, Deepseek has also proven with its R1 model that you can achieve similar services such as Openais Chat with older hardware and only a fraction of the budget.

Wider implications for the global AI market

The breakthrough announced by the Ant Group could have far-reaching effects on the global AI market. A price war of AI providers has been raging in China for months. The trigger was the publication of “V2”, the predecessor of Deepseek's current AI “V3”. In response to this, Alibaba and other technology groups such as Baidu and Tencent had reduced prices for the use of their respective AI programs by up to 97 percent.

Price pressure and optimization in the global context

The price pressure in the AI area is not limited to China. Western companies like Google have also started to optimize their AI models and reduce prices. For example, Google reduced the prices for its Gemini 1.5 per model for input and output tokens by more than 50 percent in September 2024.

The developments at the Ant Group and other Chinese companies could increase this global trend. It is particularly interesting that the cheaper models do not necessarily have to be less efficient. Google reports, for example, of significant performance increases in its optimized models, especially in the areas of mathematics (20 percent improvement), long context and vision.

Another turning point in AI development from China?

The breakthrough announced by the Ant Group could mark an important turning point in AI development. The ability to train efficient AI models with more cost-effective chips could challenge Nvidia's dominance in the AI chip market and promote the democratization of AI technologies.

This development also shows how companies can strengthen their position in a dynamic and challenging market environment through innovative approaches and strategic partnerships. The use of a diversified chip basis could be a way to master the challenges of global supply chains and at the same time reduce the development costs.

For the future, it remains to be seen whether other companies will follow the example of the Ant Group and develop similar strategies. The ability to train AI models could be a decisive competitive advantage in an increasingly competitive market and ultimately enable access to AI technologies for a wider range of organizations and applications.

Suitable for: