The Ministry of Industry and Information Technology’s new directive to establish “computing power banks” marks a sophisticated shift in how digital infrastructure is commoditized. By treating floating-point operations and token generation as storable, tradable assets, the government aims to rectify the chronic supply-demand mismatch that currently inflates operational costs for 95% of small and medium-sized enterprises (SMEs).
This initiative targets a comprehensive service system by 20 end-2028, aiming to cover at least 10 out of 15 major industrial categories. For SMEs striving to reach “little giant” status—a designation for high-growth, tech-focused firms—the ability to access high-performance computing (HPC) without the heavy capital expenditure of private server clusters is a critical competitive advantage.
The “computing power bank” model allows organizations to deposit idle capacity during off-peak cycles and withdraw it when processing demands spike. This creates a dynamic ecosystem where resource utilization rates, which often hover below 30% in fragmented private centers, can be optimized toward a 70% to 80% efficiency threshold through national pooling.
According to reports from People’s Daily, the introduction of “computing power supermarkets” will further lower barriers by providing a marketplace for on-demand access. This platform-based approach utilizes token-based billing—similar to the consumption models used by global LLM providers—allowing firms to pay exactly for the inference or training cycles they consume, rather than flat-rate monthly leases.
To bridge the initial affordability gap, the government is deploying targeted subsidies through computing power vouchers. These digital coupons, distributed via the SME service network, effectively lower the entry price for AI adoption, allowing a startup to run complex multimodal architectures at a fraction of standard commercial cloud rates.
The plan also calls for an expansion of the China Computing Power Platform to enable real-time, precise allocation across regional borders. By integrating fragmented supply from underutilized nodes, the system can redirect GigaFLOPS of power to high-demand clusters, ensuring that a surge in AI development in one province doesn’t lead to a localized “computing drought.”
Strategically, this move aligns with China’s broader goal of building a resilient AI supply chain that is not solely dependent on high-end hardware imports. By maximizing the efficiency of existing domestic silicon through smarter distribution, the country can sustain a high innovation rate even under external procurement constraints.
For the 10 targeted industry categories, the reduction in “complexity of deployment” is as vital as the cost reduction. Standardizing the interfaces for these power banks ensures that an SME doesn’t need a dedicated team of systems engineers to tap into the national grid, thereby reducing the “talent friction” typically associated with high-tech scaling.
The success of this pilot will likely be measured by the growth rate of AI-integrated products within the SME sector over the next 24 to 36 months. If the 2028 targets are met, the cost of AI inference for a typical mid-sized manufacturing firm could drop by an estimated 40% to 60%, fundamentally altering the ROI calculations for industrial automation.
In summary, the transition toward a “tradable, storable asset” model for computing power reflects an advanced understanding of the digital economy’s requirements. By treats computing cycles with the same liquidity as financial capital, China is constructing a foundation for the next generation of “little giant” enterprises to scale with unprecedented speed.
News source:https://peoplesdaily.pdnews.cn/business/er/30051806182