Chinese AI companies are often in the headlines due to their low-cost AI models. Now, a new Chinese company, Ant Group, backed by Alibaba founder Jack Ma, claims to cut AI training costs by almost 20%. A recent Bloomberg report states that the company achieves this by using local chips from Alibaba and Huawei. So, in this story, I will talk about how these Chinese AI companies are able to keep their costs low.
Strategic Use of Domestic Semiconductors
In the recent example, the Ant Group used domestic semiconductors and employed the Mixture of Experts (MoE) machine learning approach. This divides tasks into specialized subsets, optimizing computational efficiency and reducing expenses. As a result, the company achieved similar outcomes to those who use Nvidia’s H800 chips, but at an almost 20% cheaper price point.
Other Chinese AI companies are adopting creative strategies to lower costs. For example, companies like 01.ai and DeepSeek focus on training models with smaller, high-quality datasets and employ the “model-of-expert” approach.
This technique uses multiple neural networks trained on industry-specific data to achieve intelligence levels comparable to larger models but with low computing power.
Impact of US Export Restrictions
US export controls limiting access to advanced chips have prompted Chinese companies to seek domestic alternatives. This shift has spurred innovation, leading to the development of cost-effective solutions that maintain performance standards.
The Chinese government is also reportedly supporting these AI companies. With this support and collaboration among tech firms, costs have come down. For example, Manus developed an AI assistant using Alibaba’s Qwen AI models.
So, by using domestic resources, innovative development approaches, and supportive policies, these AI companies are able to reduce costs without compromising on efficiency.
