"What exactly is AI, a field that has drawn so much attention? And how does it truly connect with society?"

This is the question Yan Junjie, Founder and CEO of MiniMax, has pondered over the past 15 years as he wrote code, read papers, and conducted experiments daily.

On July 26, the World Artificial Intelligence Conference 2025 (WAIC 2025) and High-Level Meeting on Global AI Governance officially opened in Shanghai. As a special guest at the opening ceremony, Yan delivered a keynote speech on the main forum titled "Everyone's AI".

As one of the most prominent global events in the AI field, this year's WAIC adopts the theme "Global Solidarity in the AI Era". In Yan's view, AI is becoming a fundamental productivity force in society as models improve. He believes that AI capabilities will continue to advance, with no foreseeable limits to their growth.

"With AI becoming so powerful and influential, will it eventually be monopolized? Will it fall into the hands of a single organization, or remain distributed among many?" Yan posed this question and firmly argued that multiple players will coexist in the AI landscape.

Yan emphasized that building an AI company is not about replicating an internet company model:  "AI is a more fundamental productivity force—it continuously amplifies personal and societal capabilities. That's why two things matter most: first, AI is a capability; second, AI is sustainable."

Photo provided by Minimax

"There Will Be Multiple Players in AI for the Long Run"

Founded in 2021, MiniMax gained notable attention on July 16 when NVIDIA CEO Jensen Huang praised several Chinese AI companies—including DeepSeek, Alibaba, Tencent Hunyuan, MiniMax, and Baidu Ernie Bot—during his speech at the opening ceremony of the 3rd Chain Expo in Beijing. Huang's remarks significantly elevated MiniMax's visibility as a rising AI unicorn.

On the same day, MiniMax launched its MiniMax Agent full-stack development platform, claimed to be the world's first high-delivery AI agent for complex full-stack web applications. A month earlier, MiniMax also released its MiniMax-M1 series—billed as the world's first open-source large-scale hybrid-architecture reasoning model.

Yan offered three reasons why multiple players will remain in the AI space:

"Every model today relies on model alignment, and different companies have different alignment goals," he explained. "Some models aim to act like competent programmers, making them excellent at agent tasks; others focus on human interaction and have high emotional intelligence; some are highly imaginative. These alignment goals reflect the value systems of their creators, and that diversity ensures long-term coexistence."

Second, Yan noted that today's AI systems are no longer single models, but multi-agent systems involving various models using different tools to solve increasingly complex problems. "This weakens the dominance of any single model," he said.

Third, the most intelligent systems of the past six months have not necessarily come from big tech. "Thanks to the surge in open-source models, smaller players are gaining ground," said Yan. He showed a visual of recent AI rankings, observing, "While closed models still lead, open-source models are quickly catching up."

Innovation Is Making AI R&D Less Costly

In just six months, MiniMax's video model Hailuo has generated over 300 million videos globally.

"With high-quality AI models, content and creativity on the internet become far more accessible," Yan said. "Lower barriers mean more people can unleash their creativity."

He argued that AI is already being used in ways that far exceed its original design and expectations—and this trend will continue. AI usage will become more universal and affordable.

On the cost side, Yan believes the training gap between firms with vast compute power and those with limited resources will narrow. "Startups can compensate by improving experimental design, thinking capacity, and organizational efficiency."

He added that through advances in distributed computing and optimization algorithms, model inference costs could drop by another order of magnitude within the next year or two.

"To sum up, the cost of training a single model is unlikely to rise significantly," Yan stated. "Innovations will make AI development less of a money-burning endeavor, though compute demands will still grow."

In closing, Yan reaffirmed his belief in the future of general artificial intelligence (AGI): "We firmly believe AGI will become a reality—and it will be accessible to all."

Disclaimer: The content and data presented in this article are for informational purposes only and do not constitute investment advice. Please verify all information before making financial decisions. Use at your own risk.