At the AI Investment Summit held in Beijing on September 16, themed "Certainty Opportunities in the AI Infrastructure Era," Sun Guoliang, Chief Product Officer and Senior Vice President of Muxi, joined a panel discussion and shared his insights on the evolving landscape of AI infrastructure.

Sun noted that Muxi operates at the most foundational layer of the AI triad—computing power. He observed that infrastructure is shifting from single-card, single-chip systems toward multi-card cluster and super-node architectures, while AI infrastructure has evolved into what he terms "Agent Infrastructure." He forecasts that future model and business needs will fundamentally redefine how infrastructure and computing power are designed and deployed.

From a chip architecture perspective, Sun highlighted the accelerating pace of model iteration and the rapid penetration of various agents across industries. Chip manufacturers must adapt daily, working in tandem with system integrators like Lenovo to ensure seamless compatibility. He emphasized that whenever any company releases a new model, chip vendors are the first to respond—while model providers might wait three months before their next update, chip makers must continuously support and adapt to new models throughout that period.

Sun also stressed that the value of tokens varies significantly. Different industries, models, agents, and service domains command distinct pricing structures. For example, using tokens for casual chat inquiries versus using them to locate national mineral resources involves entirely different pricing models. Similarly, if tokens power agents aimed at breakthroughs in cancer or complex disease treatment, the value escalates dramatically.

On the infrastructure front, Sun reiterated Muxi's commitment to a fully domestic, homegrown technology path. He pointed out that infrastructure carries inherent sovereignty considerations, making the mission of building a reliable domestic computing power foundation both challenging and critical.