Ant Group, the Chinese financial technology giant backed by Jack Ma, has developed new artificial intelligence training methods using homegrown semiconductors that cut costs by 20% while matching the performance of Nvidia’s H800 chips.
The breakthrough comes as China seeks technological self-sufficiency amid U.S. export restrictions on advanced semiconductors, potentially reshaping the global AI landscape.
Finding domestic solutions under global constraints
Ant Group’s achievement represents a significant step in China’s push toward technological independence. By using chips from Chinese companies like Alibaba and Huawei, Ant has demonstrated viable alternatives to restricted foreign components.
“I think we’re reaching the stage where we can confidently say that China defeated the U.S. semiconductor sanctions,” said entrepreneur Arnaud Bertrand.
The development marks a turning point in China’s response to international trade restrictions. While U.S. export controls were designed to slow China’s advancement in cutting-edge technologies, Ant Group’s innovation suggests these measures may have accelerated domestic research and development efforts instead.
Industry analysts note that necessity has become the mother of invention in China’s semiconductor sector. What began as a defensive strategy has evolved into an aggressive push for technological sovereignty, with companies like Ant Group leading the charge toward self-reliance.
The timing is particularly significant as global competition in AI development intensifies. By reducing dependence on foreign technology, Chinese companies can pursue innovation pathways tailored to their specific needs and market conditions without fear of supply chain disruptions.
A smarter approach to AI
At the heart of Ant Group’s success is the Mixture of Experts (MoE) machine learning technique. This approach breaks down tasks into smaller, specialized models, improving efficiency and reducing computing needs.
By implementing MoE, Ant Group achieves results similar to those from high-performance Nvidia chips but at significantly lower cost. This technique represents a fundamental shift in how AI systems can be designed and optimized.
Traditional AI models attempt to solve all problems with a single large network, which requires enormous computational resources. The MoE approach, by contrast, activates only the parts of the network needed for specific tasks, dramatically improving efficiency.
“It’s like having a team of specialists rather than a single generalist,” explained one AI researcher familiar with the technology. “You don’t need everyone working on every problem.”
This methodology allows Ant Group to distribute workloads across domestic chips in ways that maximize their collective capabilities while minimizing resource usage.
Cost savings drive competitive advantage
The financial benefits are substantial. Training AI models on 1 trillion tokens — basic units in machine learning — traditionally costs about 6.35 million yuan ($880,000) using premium hardware. Ant Group’s method reduces this to 5.1 million yuan.
These savings strengthen Ant Group’s market position while making advanced AI more accessible within China’s tech ecosystem. The cost reduction of 1.25 million yuan per trillion tokens translates to significant operational advantages when scaled across multiple AI projects.
For perspective, modern large language models often train on several trillion tokens, meaning Ant Group’s approach could save tens of millions of yuan during development cycles.
These efficiencies create a virtuous cycle: lower training costs enable more experimentation, which leads to better models, which attract more users and generate more data, further improving the models.
The economic implications extend beyond Ant Group itself. As this technology becomes available to other Chinese companies, it could accelerate AI adoption across sectors from manufacturing to healthcare, creating new competitive advantages for China’s economy.
AI applications expand into healthcare and finance
Ant Group is now deploying its AI technology across multiple sectors. The company recently acquired the online health platform Haodf.com to enhance AI-driven medical services.
The company has created an AI Doctor Assistant to support Haodf’s network of 290,000 physicians with tasks like managing medical records, analyzing patient symptoms, and suggesting potential diagnoses. This system aims to reduce administrative burdens while improving diagnostic accuracy.
“Medical documentation can consume up to 30% of a doctor’s time,” noted a healthcare technology expert. “AI assistants like Ant Group’s could return valuable hours to patient care.”
Beyond healthcare, Ant Group is launching Zhixiaobao for everyday assistance and Maxiaocai for financial advisory services. Zhixiaobao helps users with daily tasks from scheduling to information retrieval, while Maxiaocai provides personalized investment advice based on individual financial situations and goals.
These applications represent just the beginning of what analysts expect will be a wave of AI-powered services leveraging Ant Group’s technology. The company’s strong position in financial services provides natural opportunities to implement AI for credit scoring, fraud detection, and investment analysis.
Technical challenges remain
Despite its progress, Ant Group faces ongoing challenges with AI training stability. Minor changes in hardware or model design have sometimes caused increased error rates, highlighting the need for continued refinement.
While Ant Group’s results are promising, independent verification remains necessary to fully validate their performance claims against international competitors. The company has been somewhat guarded about specific benchmarks and testing methodologies, making direct comparisons difficult.
Technical experts point out that achieving consistent performance across diverse AI tasks presents a particular challenge. While MoE techniques excel at certain applications, they may require additional optimization for others.
“The real test will come when these systems are deployed at scale in production environments,” said one AI systems specialist. “Laboratory results don’t always translate to real-world performance.”
Looking ahead
Ant Group’s innovation in AI development through Chinese semiconductors and techniques like Mixture of Experts marks a turning point in China’s quest for tech self-reliance. This achievement showcases domestic innovation potential and sets the foundation for wider applications across industries, strengthening China’s position in global AI development.
As the technology matures, industry watchers expect Ant Group to expand partnerships with other Chinese tech companies, creating a more robust domestic AI ecosystem. Government backing for such initiatives remains strong, with AI development featuring prominently in China’s latest five-year plan.
The global implications are significant. As Chinese companies demonstrate the ability to innovate despite trade restrictions, the leverage provided by such measures diminishes. This could potentially influence future international technology policy and trade negotiations.
For Ant Group specifically, success in AI represents an important diversification beyond its core financial technology business, which has faced increased regulatory scrutiny in recent years. By establishing leadership in AI, the company positions itself for growth regardless of regulatory headwinds in any single sector.
As China achieves AI breakthroughs with domestic chips despite U.S. sanctions, what do you think this means for the future of global technology competition and innovation? Please share your views below.