AI Regulation Enters Dual-Track Era: 'Computing Power Banks' and Data Governance Reshape Industry Valuation

“Flooding the Pond to Raise Fish” in Computing Infrastructure vs. “Drawing Lines and Building Walls” in Data Governance: Dual-Track Policies Reshaping the Foundational Logic of AI Industry Valuation

In March 2026, China’s AI regulatory landscape reached a landmark watershed moment: On one side, the Ministry of Industry and Information Technology (MIIT) publicly advanced its “Computing Power Bank” inclusive model—featuring token-based billing, deposit-and-withdrawal mechanisms for idle computing resources, and dedicated SME service zones—to lower technological entry barriers. On the other, the Cyberspace Administration of China (CAC), jointly with MIIT and the Ministry of Public Security (MPS), launched a full-chain special campaign on personal information protection; concurrently, the Actors Committee of the China Radio and Television Social Organization Federation issued the nation’s first industry-wide ban specifically targeting AI-powered face-swapping. These two initiatives are not isolated actions but rather two complementary facets of a unified governance philosophy—one that applies coordinated “dual-track” regulation to AI’s two core production factors: computing power and data—simultaneously promoting development and strengthening oversight. The direct market impact has been swift: major large-model-related stocks within the Hang Seng Tech Index plunged 10–15% in a single day—a sharp reflection of investors’ turbulent reassessment of commercialization pathways and compliance costs.

The “Computing Power Bank”: An Infrastructure Revolution from Resource Monopoly Toward Inclusive Sharing

The core breakthrough of MIIT’s Special Action Plan for Empowering SMEs through Inclusive Computing Power lies in transforming computing resources—long monopolized by major cloud providers and supercomputing centers—into standardized, quantifiable, tradable, and accumulable service units. Its innovation manifests in three institutional designs:
First, establishing a formalized operational architecture for the “Computing Power Bank,” enabling SMEs to deposit idle cycles from their in-house GPU clusters (e.g., overnight or during holidays) onto the platform in units of “core-hours,” earning transferable computing-power tokens in return.
Second, implementing a “computing power supermarket”-style supply-demand matching mechanism—within the SME zone of China’s National Computing Power Platform—that enables millisecond-level resource scheduling and supports highly flexible billing models, including per-GPU-hour, per-task granularity (e.g., training one LoRA fine-tuned model), and more.
Third, integrating national computing power internet service nodes to enable cross-domain coordination between edge computing resources (e.g., on-premises servers in smart factories) and cloud-based large-model inference capabilities.

At its essence, this model shifts computing power away from capital-intensive, heavy-asset infrastructure toward a lightweight, finance-enabled means of production. According to MIIT’s preliminary estimates, pilot regions have seen average computing power costs for SMEs drop by 42%, while model iteration cycles have shortened by 60%.

This transition carries profound industrial implications. It breaks the long-standing developmental bottleneck confronting AI application-layer enterprises—such as industrial quality inspection SaaS providers and developers of grassroots medical imaging assistance systems—whose growth has been constrained by high computing procurement thresholds and complex operations & maintenance requirements. As a result, the “small model + domain-specific computing power” paradigm is emerging as a scalable, replicable technical pathway. More critically, the token-based billing system lays the groundwork for future financial instruments—including computing-power futures and computing-power-backed financing—signaling China’s accelerated effort to build the world’s first national-level computing-power factor market infrastructure.

The Data Security Special Campaign: A Governance Upgrade from Gray Zones Toward Accountability Closure

In stark contrast to the openness on the computing-power front, data-side regulation has tightened with unprecedented rigidity. This special campaign—led by the CAC and jointly implemented by MIIT and MPS—marks the first time regulatory focus has shifted beyond traditional app-level violations (e.g., unauthorized data collection) to encompass the entire lifecycle of data generation and usage in the AI era. Its three key breakthroughs directly address industry pain points:
First, it explicitly classifies “using actors’ images and voice recordings—without explicit permission—for large-model training” as illegal. The Actors Committee’s statement—prohibiting “AI-powered face-swapping synthesis, voiceprint cloning and replication, and unauthorized alteration of film/TV materials”—effectively establishes a “personality rights-first” principle for audiovisual content data.
Second, it brings covert data-collection scenarios—including SDKs, ad-attribution algorithms, and AI-powered tutoring tools for education—under close scrutiny. All data-processing activities must now pass a triple verification: “minimum necessary” scope, “separate consent,” and “revocability.”
Third, it establishes an inter-departmental enforcement coordination mechanism: the CAC oversees content compliance review; MIIT regulates interface permissions; and the MPS traces the origins of data breaches—forming a closed-loop accountability chain.

Notably, this campaign avoids blunt, across-the-board restrictions. Policy documents emphasize “classified and tiered governance”: anonymized clinical data in healthcare and aggregated trajectory data in transport/logistics are explicitly endorsed for AI training—provided they operate within robust security frameworks. This “precision-demolition” approach reflects policymakers’ deep understanding of data’s economic value—the goal of governance is not to stifle innovation, but to rebuild a trustworthy data ecosystem through rights clarification (e.g., affirming facial images and voiceprints as personality rights), pricing (requiring licensing fees for data use), and traceability (mandating blockchain-based, full-lifecycle audit logs).

Reconstructing Industry Valuation Logic Through Dual-Track Policy Synergy

The simultaneous implementation of computing-power democratization and stringent data governance is fundamentally reshaping value distribution across the AI industry chain. Short-term market volatility is inevitable: the single-day valuation plunge among large-model companies stems from investors recalibrating two cost components—while computing costs decline under the “bank” model, data compliance costs (including copyright acquisition, manual auditing, and investments in privacy-enhancing hardware) are projected to rise by 30–50%. Yet the deeper structural shift lies in an irreversible business-model transformation. The historical path—relying on massive-scale web-scraping of publicly available internet data to train general-purpose large models—is now effectively obsolete. Model competitiveness will increasingly hinge on enterprises’ ability to acquire and govern high-quality, licensed, domain-specific data. For instance, a medical AI firm holding an exclusive, expert-annotated dataset from top-tier hospitals will see its valuation anchor shift from parameter count to its certified level of data compliance; similarly, film studios may evolve from pure content providers into “AI training-data operators,” generating new revenue streams via licensed digital avatars of actors.

This restructuring is also spawning novel industrial opportunities. The “Computing Power Bank” is driving surging demand for domestic GPU scheduling middleware and heterogeneous computing virtualization software; meanwhile, data governance is catalyzing explosive growth in niche sectors—including privacy-preserving computation (federated learning, trusted execution environments), AI content provenance (digital watermarking, blockchain-based evidence storage), and automated compliance-audit tools. According to CCID Consulting, China’s AI governance technology market is projected to reach RMB 8.7 billion in 2026—a 124% year-on-year increase.

Conclusion: Building Dynamic Governance Resilience Amidst Technological Acceleration

As computing-power tokens flow between SME servers—and AI face-swap detection algorithms instantly intercept noncompliant videos on the SARFT monitoring platform—China’s AI governance is moving decisively beyond binary, zero-sum dilemmas. The dual-track policy’s essence is to unleash innovative momentum through “soft” infrastructural provisioning, while safeguarding societal fundamentals via “hard” rule-based constraints. This balance-in-tension is neither technological suppression nor risk compromise—it is, rather, a deliberate rehearsal for AGI-era governance. Only when computing power becomes as universally accessible as electricity or water, and data attains the same clarity of ownership as land, can artificial intelligence truly serve as a stable engine for high-quality development. The market’s short-term turbulence is, in fact, the most authentic labor pang preceding the birth of a new order.