Jack Ma-backed Ant Group has published a research paper this month claiming an artificial intelligence breakthrough. The company reports its models sometimes outperformed those from Meta Platforms Inc. on certain benchmarks, though Bloomberg News has not independently verified these claims.

Ant's approach could advance Chinese AI development by potentially reducing the cost of inferencing and supporting AI services. The company's work focuses on creating large language models (LLMs) that don't require the high-performing, expensive graphics processing units (GPUs) typically sold by Nvidia.

The company is utilizing Mixture of Experts (MoE) models, which have become popular among AI developers. This technique breaks tasks into smaller data sets, similar to having specialists focus on specific parts of a job. Companies including Google and Hangzhou startup DeepSeek have adopted this approach.

Traditional MoE model training has been costly, requiring premium GPUs from companies like Nvidia. This expense has limited adoption among smaller firms. Ant's research aims to change this by developing ways to train LLMs more efficiently without relying on top-tier hardware.

The title of Ant's paper clearly states this goal: to scale a model "without premium GPUs." This approach contrasts with Nvidia's strategy under CEO Jensen Huang, who has maintained that computational demands will continue growing even with more efficient models.

Huang has argued that companies will need better chips to generate more revenue rather than cheaper ones to cut costs. Nvidia's strategy involves:

• Building larger GPUs
• Adding more processing cores
• Increasing transistor count
• Expanding memory capacity

Meanwhile, OpenAI has asked the Trump administration to allow AI companies to use copyrighted materials for training, framing AI development as a national security issue. In a proposal submitted Thursday, the company argued this policy would help maintain American leadership in AI technology over competitors like China.