At the recent AI Investment Summit held in Beijing under the theme "Deterministic Opportunities in the AI Infrastructure Era," Sun Guoliang, Chief Product Officer and Senior Vice President of MUXI Co., Ltd. (also known as Muxi Shares), delivered a keynote address. He emphasized that with rapid model iteration and the expansion of Agent applications across industries, computing infrastructure is shifting from single-card, single-chip architectures toward multi-card cluster and super-node designs. He argued that future model capabilities and business requirements will ultimately redefine the entire computing landscape.
Describing the intense pressure on chip suppliers, Sun highlighted the shift in industry dialogue from "AI infrastructure" to "Agent infrastructure." He noted that as models compete for speed and agents proliferate across verticals, his company must achieve day-zero adaptation for every new model released and every industry-specific agent, often collaborating with leading system integrators like Lenovo to ensure seamless compatibility. He painted a vivid picture of his team's reality, stating that whenever any new model is launched, his team is the first to mobilize. Even if the model developer waits three months to observe market reception, his engineers are busy adapting to a constant stream of new models during that entire period.
Sun also introduced a crucial perspective on token valuation, arguing that not all tokens hold equal worth and that pricing logic varies drastically across industries. He explained that using a token for casual chat or simple inquiries represents one pricing model, while deploying tokens to search for mineral resources on a national scale follows an entirely different valuation framework. He further noted that if tokens evolve into agents assisting breakthroughs in cancer research or intractable diseases, the pricing would be incomparable to other uses. He suggested that this differentiated valuation logic is essential for comprehending the commercial potential of the computing layer.
On the infrastructure front, Sun reiterated the company's unwavering commitment to a fully domestic, self-developed technology route. He acknowledged that while numerous chip options once existed, the semiconductor sector has now entered an intensely competitive phase. At this year's World Artificial Intelligence Conference, the company unveiled its super-node solution and will continue operating under a three-generation parallel model involving mass production, ongoing research, and future planning to support its foundational compute infrastructure. He emphasized that they consider this infrastructure to carry sovereign significance and that building a reliable domestic computing foundation is a formidable long-term mission.
Concluding the address, Sun returned to the summit's central investment theme, candidly admitting that many current models and chips may not be world-class, and existing facilities might not be the best globally. He framed the core question as identifying which industries can use current equipment, mature models, agents, and systems to create value through a closed loop. He suggested that perfection is not a prerequisite for success; achieving a passing grade is sufficient to attract buyers. The real question facing China's AI industry today is who will pay, how much they will pay, and for what specific solutions. If an 80% solution can genuinely deliver tangible results and practical applications within its sector, Sun believes it could resolve the majority of investment uncertainties currently plaguing the market.