Jack Ma’s Ant Group Pioneers Low-Cost AI Training

By turning to domestic semiconductors—including those from Alibaba and Huawei—the company says it can reduce training costs by about 20% while still achieving performance levels similar to those made possible by high-end chips.

Firms traditionally relied on powerful chips from a leading US tech company to build large AI systems. But this kind of setup can be too costly for smaller players.

Did you know?

Want to get smarter & wealthier with crypto?

Subscribe - We publish new crypto explainer videos every week!

What is Blockchain? (Animated Examples + Explanation)

SUBSCRIBE

ON YOUTUBE

Ant Group’s approach, which blends chips from an American semiconductor firm with Chinese-made processors, aims to change that. Though Ant still uses high-performance hardware from global suppliers, its focus is shifting toward lower-cost solutions for its latest AI development projects.

The company released a research paper on March 11 suggesting its models outperformed those from larger international tech companies in some tests. These results have not yet been confirmed, but if accurate, they highlight growing progress in China’s AI field.

One of the biggest advantages of Ant’s method is that it can open doors for smaller businesses. With reduced expenses, they can experiment with advanced tools that were once out of reach.

Ant Group also shared that it used to cost 6.35 million yuan to train one trillion data units with high-end hardware. Their updated method brings that figure down to about 5.1 million yuan, which offers a more budget-friendly way to scale AI systems.

Recently, Manus, a China-based AI platform, claimed it had outperformed OpenAI's $200-a-month agent. How did the AI community respond? Read the full story.