In the previous chapter, we explored how the intelligent computing center plays the role of "City Brain" on the grand stage of urban governance. Yet the vitality of a nation's digital economy depends not only on the wisdom of the "brain," but also on the health and vigor of the billions of "cells" that make up its body -- the thousands upon thousands of small and medium enterprises.
According to the standing formulation of China's Ministry of Industry and Information Technology, SMEs contribute more than 50% of tax revenue, over 60% of GDP, over 70% of technological innovation, and over 80% of urban employment (see MIIT's SME bureau materials). They are the cornerstone of the national economy, the indispensable "capillaries" of the industrial chain, and the most active wellspring of innovation.
However, in the face of the surging wave of artificial intelligence, these "capillaries" are confronting unprecedented challenges, even a crisis of "generational gap" that threatens their very survival.
As we analyzed in Chapter 3, the research, development, and application of AI technology carry exceptionally high barriers. It demands three core elements: computing power, data, and talent. And these three are precisely the resources that SMEs most acutely lack.
- The computing power barrier: A top-tier AI training chip often costs from tens of thousands to over a hundred thousand yuan (an order-of-magnitude illustration that shifts with chip generation and supply), and building a small training cluster typically costs in the millions. For the vast majority of SMEs, this is an unaffordable fixed-asset investment.
- The data barrier: High-quality, large-scale, labeled industry data is the "fuel" for training good models. The giants naturally possess vast data thanks to their platform advantages; SMEs' data, by contrast, is often scattered, non-standard, and insufficient in scale -- a "low-grade ore."
- The talent barrier: The compensation of a senior AI algorithm engineer is far beyond what most SMEs can afford. Talent is heavily concentrated in leading enterprises, forming a "talent black hole."
These three barriers together constitute a deep and wide "giant chasm." On one side of the chasm are the deep-pocketed tech giants with vast computing power, data, and talent, galloping forward on the AI track. On the other side are the vast multitude of SMEs, watching helplessly as this technological revolution's train roars past, unable to board because they lack a "ticket" -- anxious and powerless.
If this "Matthew effect" is allowed to run its course, the consequences would be disastrous. It would lead to the ossification and suffocation of industrial innovation vitality, ultimately resulting in a situation where a handful of giants monopolize innovation.
The core mission of the national intelligent computing center at the level of industrial empowerment is to play the role of a "bridge builder" and "empowerer." Its responsibility is not to gild the lily, helping giants run faster, but to deliver timely aid -- to issue an "entry ticket" to the intelligent era for the tens of millions of SMEs shut out at the door.
This "entry ticket" is no empty slogan. It is composed of two specific, operable institutional innovations that must be jointly created by the government and the intelligent computing center:
- The computing voucher system: solving the problem of "cannot afford" computing power.
- The data sandbox: solving the problem of "dare not use" and "cannot use" data.
6.1 The Computing Voucher System: The Wisdom of Investing in "Possibility"
A "computing voucher" is a "computing subsidy certificate" funded by government finances and issued to eligible tech SMEs, startup teams, and university research institutions.
After receiving a "computing voucher," an enterprise or team can use it like a coupon to offset fees for using AI computing power at the local national intelligent computing center.
This seemingly simple institutional design embodies profound industrial development logic and public policy wisdom. It is not a simple "welfare" or "relief," but a strategic investment in the "future possibilities" of the regional innovation ecosystem.
6.1.1 Why Must It Be a "Voucher" Instead of Direct Cash Subsidies?
Traditional government support policies often involve direct cash subsidies to enterprises. But this approach has clear drawbacks in the AI era:
- Risk of fund misappropriation: After receiving cash, it is difficult to effectively monitor whether the enterprise truly uses it to purchase computing power and invest in R&D. The funds may be diverted to other purposes such as salary payments or marketing, deviating from the policy's original intent.
- Market distortion effect: Direct cash subsidies can easily breed "rent-seeking" and "fraud." Some enterprises may fabricate R&D projects to obtain subsidies, disrupting fair competition in the market.
- Low support efficiency: For startups, what they often lack most is not hundreds of thousands in cash, but professional, flexible computing services that allow them to quickly validate their technical ideas. Giving money directly is not as effective as giving resources directly.
The "computing voucher" system, however, perfectly avoids these problems:
- Use-locked, precision drip irrigation: Computing vouchers are earmarked for specific purposes and can only be spent at the designated intelligent computing center, ensuring that every penny of fiscal funds is precisely transformed into real computing power, irrigating the very roots of innovation.
- Results-oriented, stimulating vitality: Only when an enterprise is genuinely engaged in AI R&D and generates actual computing consumption can it use the voucher. This mechanism inherently filters for innovation entities that are "truly doing the work," preventing "bad money from driving out good."
- Leverage effect, amplifying value: One million yuan of government investment in computing vouchers might leverage millions in the enterprise's own R&D funds and human resources, ultimately incubating an innovation project worth hundreds of millions. The "leverage effect" of fiscal funds is greatly amplified.
6.1.2 Designing and Operating the Computing Voucher System: A Complete Closed Loop
For the computing voucher system to truly take root and deliver results, the government, the intelligent computing center, and enterprises must work together to design a scientific, fair, and efficient operational closed loop.
Step 1: Qualification Determination and Quota Allocation (Who Gets Vouchers?)
This is the entry point of the system and must be strictly controlled to ensure fairness.
- Establish an "Innovation Score" model: Led by departments of science and technology, industry and information technology, and market regulation, establish an "innovation score" evaluation model for SMEs. Evaluation dimensions may include:
- Technological advancement: Does the enterprise have independent intellectual property (patents, software copyrights)? Is its technical path in a frontier area encouraged by the state?
- Team background: Do core team members have relevant research or industry backgrounds?
- Commercial potential: Does its product or service have a clear market positioning and business model?
- Local contribution: Does the enterprise pay taxes and provide employment locally?
- Tiered quota allocation: Based on the total innovation score, classify enterprises into different tiers (e.g., seed stage, growth stage, gazelle enterprise) and grant different amounts of vouchers. For example, a newly established AI startup might receive 100,000 yuan worth of vouchers annually, while a "specialized and sophisticated" enterprise with mature products and rapid growth might receive 500,000 yuan.
- Dynamic adjustment and exit mechanism: Quotas are not set in stone. Annual reviews should be conducted. For enterprises with rapid R&D progress and significant results, quotas can be increased; for those that "receive vouchers but never use them" or abuse resources, quotas should be reduced or eligibility revoked.
Step 2: Voucher Issuance and Use (How to Use Vouchers?)
- Platform-based management: The intelligent computing center needs to develop an online "computing voucher" management platform. After passing qualification determination, the enterprise account will automatically receive electronic vouchers of the corresponding amount.
- Integration with billing system: When an enterprise rents computing power at the center, the incurred fees will be deducted from the voucher quota first. After the quota is exhausted, the excess is paid by the enterprise itself.
- Scope of use: It may be stipulated that vouchers are mainly used to offset fees for core computing tasks like AI training and inference, and not for general-purpose fees like storage and bandwidth, ensuring that "good steel is used for the blade."
Step 3: Performance Evaluation and Outcome Tracking (Are Vouchers Used Well?)
This is the closed loop of the system and the key to its self-optimization.
- Establish an "Outcome Archive": Every enterprise using vouchers must regularly submit "R&D progress reports" on the platform. The report should explain what milestones have been achieved using this computing power. For example:
- How much has model accuracy improved?
- What new patents have been applied for?
- What product features have been iterated?
- What new customers or orders have been obtained?
- Performance linked to quotas: These "outcome archives" will serve as the most important basis for the next year's innovation score review and quota adjustment, forming a virtuous cycle of "input-output-reinvestment."
- Benchmarking and demonstration: The intelligent computing center and government should regularly select and publicize excellent cases of voucher usage, and vigorously promote and recognize them. By establishing these "everyday heroes," more SMEs can be inspired to engage in AI innovation.
Through such a complete closed-loop design, the "computing voucher" transforms from a simple subsidy tool into a powerful "industrial innovation incubator." It systematically solves the most difficult "cold start" problem for SMEs in their journey from zero to one.
6.1.3 The Manager's Role: From "Rent Collector" to "Angel Investor"
Implementing the computing voucher system places higher demands on the intelligent computing center's manager. Our mindset must shift from a "rent collector" who quibbles over every kilowatt-hour and every hour of machine time, to an "angel investor" with long-term vision and the courage to invest in the future.
- We invest not in certain returns, but in "possibility": Most startups using vouchers will inevitably fail. This is a natural law of the innovation process. We cannot, for fear of failure, only invest resources in projects that look the "safest" and most "reliable." Our mission is precisely to provide a tolerant soil for trial and error for the boldest, most cutting-edge, and most uncertain "whimsical ideas." A seemingly fanciful student project today could be the next great company tomorrow.
- What we harvest is not just financial returns, but "ecosystem dividends": Even if these enterprises never directly generate much revenue for the intelligent computing center, as long as some of them grow, they will create local jobs, pay taxes, attract talent, and improve the industrial chain. A prosperous, vibrant local digital economy ecosystem is the foundation for the long-term survival and development of the intelligent computing center. This indirect, long-term "ecosystem dividend" is far more valuable than short-term machine-hour revenue.
The "computing voucher" is a warm hand extended from the intelligent computing center to SMEs. It conveys not just precious computing resources, but also a message of support and confidence from the national level. It tells every innovator with a dream: you are not fighting alone; the nation's powerful computing power is your strongest backing.
6.2 The Data Sandbox: Building a "Revolving Door" Between Security and Openness
If the "computing voucher" solves the problem of "cannot afford," then the "data sandbox" addresses a more complex and sensitive issue -- the "dare not use" and "cannot use" of data.
Data is the "fuel" for AI, but this "fuel" is highly sensitive.
- For the state and government: Government data, medical data, and financial data involve national security and citizen privacy and are absolutely "red line data" that cannot be directly opened to enterprises.
- For enterprises: Production data, supply chain data, and customer data are their most core trade secrets and are equally not to be easily shared.
This phenomenon of "data silos" greatly hinders the industrial implementation of AI technology. On one hand, enterprises have algorithms and computing power but cannot access high-quality industry data to train truly useful models. On the other hand, governments and leading enterprises hold massive "data gold mines," but due to security and compliance concerns, these mines lie dormant in servers, unable to be transformed into productivity.
How to break this deadlock? The "data sandbox" is the ingenious "revolving door" we design in this seemingly insurmountable wall.
The essence of a "data sandbox" is a secure and controllable data computing environment, built and operated by the intelligent computing center, strictly isolated from the outside world. In this environment, the "ownership" and "usage rights" of data are cleverly separated.
6.2.1 The Core Principle of the Data Sandbox: "Algorithms Move, Data Stays"
The traditional model of data cooperation is "algorithms stay, data moves" -- Enterprise A needs data from Enterprise B, so Enterprise B copies the data to Enterprise A via an interface or file transfer. In this model, once data flows out, control is lost, posing a huge leakage risk.
The "data sandbox" reverses this, following the core principle of "algorithms move, data stays."
- Data provider (e.g., government department, hospital): Securely "injects" anonymized and authorized data into the sandbox environment. This data never leaves the sandbox's physical boundary from start to finish.
- Data user (e.g., AI startup): "Submits" their developed AI model or algorithm code into the sandbox environment.
- Sandbox environment: Internally, uses the data to "feed" the algorithm, completing the model training or validation process.
- Result output: The sandbox only allows the trained model file or statistical analysis results to be "taken out"; the raw data itself can never be exported.
In this way, data users may obtain training capability inside a controlled environment while providers reduce the risk of direct raw-data exfiltration. It does not provide absolute security: model memorization, gradient or output leakage, access-control mistakes, side channels, insiders, and supply-chain failures can still expose information. A security conclusion must correspond to a threat model, isolation boundary, privacy tests, audit records, and incident response; it cannot follow from "the data never leaves the domain" alone.
The design goal of this "revolving door" is to let people (algorithms) enter and exit (with models) while raw data stays behind the door -- but the door's value depends on the continued effectiveness of the technical and administrative measures above, not on a physical impossibility of extraction.
6.2.2 Technical Architecture and Governance System of the Data Sandbox
To build a trustworthy "data sandbox," both powerful technology and a strict governance system are needed.
Technical Architecture:
- Secure and isolated environment: The sandbox must be deployed on physically isolated or strictly logically isolated server clusters, with independent networks and storage, completely disconnected from the public internet.
- Trusted execution environment: Hardware-encrypted confidential computing technologies (such as Intel SGX, AMD SEV) can be used to ensure that even the center's operations personnel cannot peer into the data and code running inside the sandbox.
- Privacy computing technologies: Within the sandbox, technologies like federated learning, secure multi-party computation, and differential privacy can be further applied. For example, multiple enterprises could jointly train a more powerful industry model without exposing their respective raw data.
- Strict code audit and behavior monitoring: All code submitted to the sandbox must undergo rigorous automated and manual auditing to prevent malicious logic designed to steal data. Every move of the algorithm while running in the sandbox (such as file access, network requests) will be meticulously recorded and monitored.
Governance System:
- Establish a "Data Ethics and Security Committee": Composed of government legal experts, industry experts, technical experts, citizen representatives, etc., responsible for formulating the sandbox's "game rules."
- Data access standards: What kind of data can enter the sandbox? What anonymization process must it undergo?
- User qualification review: What enterprises and individuals are eligible to apply to use the sandbox?
- Application scenario approval: Applicants must specify what specific problem they hope to solve using the data. Does the application align with public interest and ethical norms? (For example, it must never be used to develop discriminatory or privacy-infringing applications.)
- Establish a clear data classification system: Data entering the sandbox should be strictly graded based on sensitivity. Different levels of users can only access data within their authorized scope.
- Transparent audit and traceability mechanism: All operations within the sandbox must leave an immutable log record. In the event of a security incident, the responsible party can be accurately traced.
6.2.3 The Manager's Role: From "Data Gatekeeper" to "Data Exchange Director"
Operating a "data sandbox" requires managers to possess a new capability -- finding the delicate balance between security and development. Our role is no longer a passive "data gatekeeper" who only knows how to say "no," but an active "data exchange director" who knows how to create value.
- Our core product is no longer data itself, but the "availability of data." We do not sell data; we sell "the ability to generate insights from data under secure and compliant conditions." This is a higher-level, more valuable service.
- Our core capability is building "trust." We need to make data providers believe that placing their data with us is safer than keeping it on their own servers; we need to make data users believe that the rules we set are fair and their intellectual property will be protected. This "trust" is the sole foundation upon which the "data sandbox" market can operate.
- Our ultimate goal is to promote the "circulation and fusion of data elements." Through the data sandbox, we can break down departmental and industry walls, bring dormant data to life, and allow data from different sources to safely "meet," creating unprecedented sparks of innovation. For example, fusing and analyzing traffic data, weather data, and commercial footfall data can provide invaluable decision-making support for urban commercial planning.
Conclusion: Paving the "Last Mile" for the Future of Industry
Chapter 6 focuses on how the intelligent computing center empowers the "nerve endings" of industry. This may not be the most glamorous or technically impressive area, but it is the most critical area that determines the "depth" and "resilience" of a region's digital economy.
- The "computing voucher" is the "precision drip irrigation" of fiscal funds, solving the "starting fuel" problem for SME innovation. It is an innovation in inclusive finance and an "angel investment" in future industry leaders.
- The "data sandbox" is the "ingenious design" of technology and institutions, solving the "trust deficit" problem in the flow of data elements. It builds a solid bridge between the cliffs of security and openness.
These two "entry tickets" together form the core lever for the national intelligent computing center to empower SMEs. They transform the intelligent computing center from a lofty "technological temple" into a "public service platform" accessible to all innovators.
As managers of this platform, our job is to pave this crucial "last mile" for local industrial innovation. Our success should not be measured merely by rack rental rates, but by how many "specialized and sophisticated" enterprises we have incubated, how many "hidden champions" in niche markets we have cultivated, and how many high-value-added jobs we have created locally.
When one day we see a local agricultural tech startup, once unknown, using our "computing vouchers" and "data sandbox" to develop an AI model that can accurately predict pests and diseases, helping thousands of farmers increase production and income -- at that moment, we can truly say with pride: we have "used well" this powerful "convergence engine." We have not only converged data, but also converged people's aspirations, converging the surging power of a region's industries moving vigorously upward.