Ant Group Stacks Chips to Reduce Development Costs for AI

China’s Ant Group is using local semiconductors to train AI at a cost that is 20 percent less than companies typically spend, according to reports. Ant used domestic chips — from companies including Alibaba, an investor in Ant, and Huawei — to launch a unique Mixture of Experts (MoE) training approach that produced results commensurate to training with Nvidia H800 chips. Ant is the latest Chinese company to focus on low cost training, joining a competition triggered by DeepSeek, which in January announced it could build AI comparable to the models released by U.S. companies like OpenAI, Anthropic and Google for billions less.

“Hangzhou-based Ant is still using Nvidia for AI development but is now relying mostly on alternatives including from Advanced Micro Devices Inc. and Chinese chips for its latest models,” Bloomberg reports, noting “the company published a research paper this month that claimed its models at times outperformed Meta Platforms Inc. in certain benchmarks.”

Conceding it couldn’t verify that claim, Bloomberg said “if they work as advertised, Ant’s platforms could mark another step forward for Chinese artificial intelligence development by slashing the cost of inferencing or supporting AI services.”

MoE has emerged as an efficient training method, assigning smaller but more focused sets of data to specific tasks. TechRepublic writes it is becoming “a preferred approach in AI training,” though some mix it with a transformer method.

OpenAI and Google have reportedly used the approach, as does DeepSeek-R1, an LLM from the Chinese startup. Bloomberg describes MoE as “very much like having a team of specialists who each focus on a segment of a job, making the process more efficient.”

Ant “on Monday announced ‘major upgrades’ to its AI solutions for healthcare, which it said were being used by seven major hospitals and healthcare institutions in Beijing, Shanghai, Hangzhou and Ningbo,” CNBC reports, adding that “the healthcare AI model is built on DeepSeek’s R1 and V3 models, Alibaba’s Qwen and Ant’s own BaiLing.”

BaiLing can answer medical questions and provide insight on patient services, Ant said in an announcement, noting that it has “medical reasoning and multimodal interaction capabilities.”

Since the U.S. imposed sanctions limiting China’s access to the most advanced processors used for AI development the country has been forced to be more industrious with the tools it has available, including indigenous and lower-end Nvidia chips.

TechRepublic explains the wave of seemingly capable LLMs from China  — from DeepSeek, Alibaba and now Ant — “illustrate how Chinese developers can create models without exclusive dependence on U.S.-based companies,” calling the move “significant” in light of the fact that the Nvidia H800 GPU, while not among the company’s newest, “is currently restricted under U.S. export controls.”

No Comments Yet

You can be the first to comment!

Leave a comment

You must be logged in to post a comment.