Archive — CN — technode
MiniMax CEO predicts another 10x drop in AI inference costs within two years
MiniMax CEO Yan Junjie predicts that AI inference costs could decrease by another tenfold within 1-2 years due to advancements in optimization and algorithms. Speaking at the 2025 AI Conference in Shanghai, he noted a dramatic price reduction in custom AI projects and emphasized the impact of open-source models and multi-agent systems on market dynamics.
Key points
- CEO Yan Junjie predicts a further 10x drop in inference costs for AI models in 1-2 years.
- This follows a previous substantial drop over the last year.
- Advancements in system optimization and algorithms are key drivers of this cost reduction.
- There is an ongoing price war in China's large model market, with prices for custom projects dropping from over RMB 10 million to as low as RMB 200,000.
- Yan believes that the capabilities of AI will increase while becoming more affordable.
- AI training precision is being enhanced using reinforcement learning in reward-defined environments.
- The emergence of open-source models and multi-agent systems is changing the competitive landscape, reducing the likelihood of a single company dominating.
Why it matters
- The prediction aligns with the broader trend of decreasing AI costs, reflecting ongoing competition in the tech industry.
- Advancements in AI efficiency are consistent with 2025 IT trends prioritizing cost-effective and scalable solutions.
- The rise of open-source models is part of a shift seen in many tech industries where collaboration accelerates innovation.
The anticipated drop in AI inference costs signifies a pivotal shift in the industry, fostering innovation and competition while ensuring that advanced AI capabilities become accessible to more companies.