Ant Group Turns to Chinese-Made Semiconductors for Artificial Intelligence Development
Ant Group, a subsidiary of Alibaba, is relying on Chinese-made semiconductors to train artificial intelligence models, in a move to reduce costs and lessen dependence on restricted US technology.
Domestic Chip Suppliers
According to sources familiar with the matter, Ant has used chips from domestic suppliers, including those tied to its parent, Alibaba, and Huawei Technologies, to train large language models using the Mixture of Experts (MoE) method. The results were reportedly comparable to those produced with Nvidia’s H800 chips.
Cost-Effective Approach
The experimentation with domestic hardware reflects a broader effort among Chinese firms to work around export restrictions that block access to high-end chips like Nvidia’s H800. Ant’s approach is cost-effective, with training one trillion tokens costing around 5.1 million yuan (approximately $880,000) using lower-specification chips, compared to around 6.35 million yuan using conventional high-performance hardware.
MoE Models
MoE models divide tasks into smaller data sets handled by separate components, making the process of producing models more efficient. Ant has published a research paper describing its work, stating that its models, in some tests, performed better than those developed by Meta.
Open-Source Models
Ant has made its models open-source, with Ling-Lite having 16.8 billion parameters and Ling-Plus having 290 billion. For comparison, estimates suggest closed-source GPT-4.5 has around 1.8 trillion parameters.
Challenges
Despite progress, Ant’s paper noted that training models remains challenging. Small adjustments to hardware or model structure during model training sometimes resulted in unstable performance, including spikes in error rates.
Conclusion
Ant’s reliance on Chinese-made semiconductors for AI development signals a significant shift in the company’s approach to AI research and development. As the company continues to explore cost-effective ways to train models, its efforts may represent a step forward in China’s attempt to lower the cost of running AI applications and reduce the reliance on foreign hardware.
FAQs
Q: What is MoE (Mixture of Experts) model?
A: MoE models divide tasks into smaller data sets handled by separate components, making the process of producing models more efficient.
Q: Why is Ant Group using Chinese-made semiconductors?
A: Ant Group is using Chinese-made semiconductors to reduce costs and lessen dependence on restricted US technology.
Q: How does Ant Group’s approach compare to Nvidia’s approach?
A: Ant Group’s approach focuses on cost-effectiveness, using lower-specification chips, while Nvidia’s approach emphasizes building GPUs with more cores, transistors, and memory.
Q: What are the benefits of Ant Group’s open-source models?
A: Ant Group’s open-source models allow for greater collaboration and development within the AI community, potentially leading to more innovative applications of AI technology.
