Jack Ma-Backed Ant Slashes AI Training Costs With Chinese Chips

Asian stocks traded mixed to flat overnight as investors digested the potential for a less severe tariff announcement on April 2—also known as President Trump’s “Liberation Day”—alongside news that Alibaba-affiliate Ant Group has been using Chinese-made chips to reduce the cost of training artificial intelligence models. 

Bloomberg reported that Jack Ma-backed Ant Group used both Chinese and U.S.-made semiconductors to train AI models that were more efficient and reduced costs by 20%. According to sources, the domestic chips included those from Alibaba Group Holding and Huawei Technologies. They noted that the results were similar to Nvidia H800 chips. The models relied on a technique known as the “Mixture of Experts” (MoE) machine learning approach, which allows AI training with significantly less computing power. 

Here’s more color: 

It got results similar to those from Nvidia Corp. chips like the H800, they said, asking not to be named as the information isn’t public.

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, one of the people said.

Two weeks ago, Ant published a paper on MoE titled “EVERY FLOP COUNTS: SCALING A 300B MIXTURE-OF-EXPERTS LING LLM WITHOUT PREMIUM GPUS,” focusing on overcoming cost inefficiencies and resource limitations in training AI models. 

Ant’s move to enter the AI race comes nearly two months after China launched DeepSeek – its answer to GPT, Llama 3.1, etc, that demonstrated significantly lower costs in training and computing power requirements versus models trained in the West (AI’s Sputnik Moment). This also comes as the US has restricted China’s AI development by limiting AI chip procurement. 

“As companies pour significant money into AI, MoE models have emerged as a popular option, gaining recognition for their use by Google and Hangzhou startup DeepSeek, among others,” Bloomberg noted, adding, “That technique divides tasks into smaller sets of data, very much like having a team of specialists who each focus on a segment of a job, making the process more efficient.” 

Bloomberg Intelligence analyst Robert Lea noted: “Ant Group’s paper highlights the rising innovation and accelerating pace of technological progress in China’s AI sector. The firm’s claim, if confirmed, highlights China is well on the way to becoming self-sufficient in AI as the country turns to lower-cost, computationally efficient models, to work around the export controls on Nvidia chips.” 

On Monday, Ant announced “major upgrades” to its healthcare AI offerings as it plans to leverage its low-cost AI model. 

China has shown that it can produce low-cost AI models far less than the tens of billions invested by OpenAI and Alphabet.

Let’s hope note…

deepseek better not be the real deal… pic.twitter.com/qv6vtINl2d

— zerohedge (@zerohedge) January 26, 2025

Refresh read: “Goldman Asks If China’s DeepSeek is AI’s Sputnik Moment.”