AI is advancing at remarkable speed, driven by a self-reinforcing cycle of expanding datasets, growing computational capacity, and evolving algorithms. During his keynote speech at China Unicom's 2026 Partner Conference on September 10, Cheng Weiliang, founder of Moxin, underscored this rapid trajectory while highlighting a twofold strain facing the sector.

On the demand side, Cheng pointed out that a severe shortage of computing power persists. As large language models scale to trillions or even hundreds of trillions of parameters, the strain on computing resources for both training and inferencing is expanding exponentially. AI applications have moved out of labs and into commercial use at scale, causing token consumption to surge and leaving a continuous shortfall in compute supply. On the supply side, he identified significant fragility due to export restrictions on advanced chips and unpredictability around leading-edge process availability, both of which substantially hinder large model development and the practical rollout of AI initiatives, making high-caliber domestic substitution indispensable.

According to Cheng, the path forward for compute lies in constructing an autonomous and dependable infrastructure. Accelerating progress in domestic computing chips is essential to dismantle technical barriers, transforming the industry's status from dependent reliance to independent digital foundation.

He went on to explain that as intelligent computing centers expand their scale, attention must move beyond raw GPU single-chip specifications toward a comprehensive total cost of ownership (TCO) perspective. Beyond performance, TCO now emerges as the key criterion for customers when making choices, covering both visible expenses like hardware procurement and energy consumption during operation, alongside hidden costs including migration and maintenance. Central to reducing these concealed expenses is a consistent degree of standardisation, which helps streamline AI training and inferencing, prevents redundant investment and idle resources, and optimizes hardware utilization to trim cost setbacks. He also stressed the importance of ecosystem openness, as a broadly compatible ecosystem requires fewer code rewrites and lowers migration expenses, thereby accelerating service rollouts. Furthermore, supply chain self-sufficiency proves essential, as dependable local sourcing and mastery of core technologies sustain long-term supply stability and fortify the safety of the entire computing infrastructure.