China – Ant Group, the fintech affiliate of Alibaba and a leader in enterprise digital transformation, has revealed a breakthrough in AI cost efficiency through the deployment of Chinese-manufactured semiconductors. The initiative, developed in collaboration with local chip partners including Huawei and Alibaba Cloud, reduced training costs for large language models (LLMs) by an estimated 20%.
The strategic shift is part of Ant’s broader plan to expand industrial AI capabilities, especially in high-value sectors such as healthcare and finance. According to internal sources, Ant’s latest models—Ling-Plus and Ling-Lite—were trained using a “mixture of experts” (MoE) architecture, a method that partitions tasks into specialized units to improve efficiency and reduce computing load.
This innovation aligns with Ant Group’s commitment to scalable and sustainable enterprise AI infrastructure. The company reported that training one trillion tokens with optimized hardware cost just 5.1 million yuan (approx. USD 720,000), down from 6.35 million yuan using conventional high-end GPUs.
While Ant continues to utilize some Nvidia hardware, its increasing reliance on alternative chipsets from local and international partners signals a push toward ecosystem diversification. The approach not only mitigates supply chain risks in light of global semiconductor restrictions but also opens up broader access to AI adoption across industries.
The models are being positioned for B2B deployment across key verticals. Ling-Plus, with 290 billion parameters, and Ling-Lite, a lighter and more agile version, are already being integrated into Ant’s AI solutions, including its financial advisory engine Maxiaocai and digital life assistant Zhixiaobao. The company’s 2025 acquisition of Haodf.com further amplifies its footprint in AI-driven healthcare.
Ant’s research was released in March 2025, with benchmark data suggesting strong comparative performance—especially in Chinese-language tasks—against leading models from Meta and DeepSeek. The company has since open-sourced both models to encourage enterprise adoption and collaborative improvement.
“Ant’s work illustrates a clear commitment to cost-effective innovation,” said Robert Lea, Senior Analyst at Bloomberg Intelligence. “This is not just a technology success—it’s a scalable business case for AI integration in complex environments.”
For business leaders and government stakeholders alike, Ant Group’s efforts offer a viable model for enterprise-grade AI transformation, enabled by resilient hardware partnerships and a focus on measurable ROI.
