Ant Group’s Jack Ma Develops Cost-Effective AI Models with Huawei Chips, Matching Meta’s Llama Performance

The position of NVIDIA in the Chinese artificial intelligence sector faces potential challenges as the Ant Group, supported by Jack Ma, has reportedly crafted AI models utilizing local Chinese chips at reduced costs.

The Rise of Homegrown AI Models in China

China’s AI industry is undergoing a significant transformation, driven by U. S.export restrictions and the rise of domestic powerhouses like Huawei. The country’s once-dependable revenue streams for NVIDIA are now being threatened by the emergence of local AI hardware. Recent reports by Bloomberg indicate that these domestically produced chips are challenging NVIDIA’s stronghold in the market. Ant Group, primarily known for its financial technology solutions, appears to have developed models that rival those from industry giants like Meta.

Cost-effective Advancements in AI Training Techniques

The report reveals that Ant Group has introduced AI models named “Ling-Lite”and “Ling-Plus, ”which are reported to be 20% cheaper than current industry standards. This cost efficiency stems not only from the use of Chinese AI chips but also from innovative training techniques developed by Ant Group. The company employs chips from Huawei alongside proprietary solutions, utilizing the “Mixture of Experts”method that greatly improves the efficiency and scalability of large models.

Huawei Preps Ascend 910C To Tackle NVIDIA's H100 In China's Domestic AI Market 1

Benchmark Achievements and Future Prospects

Ant Group is reportedly achieving performance levels from its hardware stack that are comparable to NVIDIA’s H800 AI GPUs, with its models reportedly outperforming some Meta benchmarks. However, as this data comes from internal testing within the company, independent verification remains necessary. Ant Group’s innovative approach allowed them to train one trillion tokens at a substantially lowered cost of around 5.1 million yuan, signifying a 20% reduction compared to traditional methods.

While Ant Group has yet to confirm any plans for public access to these models, the developments indicate a notable shift in China’s AI landscape. The nation appears poised to redefine the underlying premise that extensive financial investment is indispensable for successful AI training.

Source & Images