Particle.news

Download on the App Store

Ant Group Cuts AI Training Costs by 20% Using Chinese and AMD Chips

The Alibaba-backed fintech giant leverages innovative techniques and diversified hardware to advance AI healthcare solutions and reduce reliance on Nvidia.

Ant Group headquarters
The results attained were even similar to those from Nvidia Corp. chips like the H800.
Image
Image

Overview

  • Ant Group achieved a 20% reduction in AI training costs by combining Chinese-made chips from Alibaba and Huawei with U.S.-made AMD chips.
  • The company unveiled two new AI models, Ling-Plus and Ling-Lite, which were trained using the cost-efficient Mixture of Experts (MoE) technique.
  • These AI models reportedly perform on par with Nvidia's H800 GPUs and outperform Meta's models in certain internal benchmarks, according to Ant Group.
  • Ant's AI healthcare solutions, built on proprietary and collaborative models, are now operational in seven major hospitals across China, improving medical services.
  • The shift to domestic and AMD chips reflects China's broader push for technological independence in response to U.S. export restrictions on advanced semiconductors.