America builds walls, China paves roads: The global impact of two divergent AI development paths

Editor's note: Warwick Powell is an adjunct professor at Queensland University of Technology. The article reflects the author's opinions and not necessarily the views of CGTN.

Artificial intelligence (AI) can become an international public good that benefits all of humanity. This proposition, advanced at the 2026 World Artificial Intelligence Conference, crystallizes a fundamental choice now confronting the world. 

We face a crossroads. One path seeks to enclose capability behind regulatory barriers, export controls and proprietary moats. The other prioritizes open cooperation, low-cost deployment and the deliberate diffusion of opportunity. 

The United States is building walls. China is paving roads. The consequences will shape not only technological trajectories but the structure of global order itself.

The American approach has hardened into a strategy of technological containment dressed in the language of open innovation. Successive administrations have layered export restrictions on advanced semiconductors, tightened controls on model weights and training data, and promoted frameworks such as the US AI action plan. Concepts like "adversarial distillation" have been invoked to police knowledge flows, while initiatives such as the Silicon Peace Declaration attempt to rally allies around exclusionary standards. 

The rhetoric celebrates openness and democratic values. The practice constructs digital fortresses. Hypocrisy defines the modus operandi.

These measures rest on a proprietary business model that treats frontier capability as a scarce rent-generating asset. Massive capital is poured into closed systems whose economic rationale depends on maintaining technological distance from competitors. Yet the model is already confronting insurmountable contradictions. Training and inference at scale demand enormous energy and capital intensity. Returns remain elusive for many commercial applications. 

Meanwhile, open-source alternatives of rapidly improving quality erode the scarcity premium that underwrites the entire edifice. When capable models can be obtained, adapted, and deployed at a fraction of the cost, the logic of closed proprietary systems begins to unravel.

The root problem, in fact, is thermodynamic and systemic. American AI development has become entangled with an energy system under strain, aging grid infrastructure and a financial architecture oriented toward speculative valuation rather than durable material productivity. The result is a paradox: a technology celebrated as a generator of informational order that, in practice, accelerates energetic and systemic entropy. High energy return on energy invested (EROEI) is the foundation of any sustained expansion of computational capacity. Where that foundation weakens, claims of inevitable dominance rest on fragile ground.

China's path is different in both design and effect. It treats AI not primarily as a vehicle for rent extraction or geopolitical leverage but as infrastructure for enabling capability across societies. 

Open-source models released by Chinese developers have already demonstrated that frontier performance need not be locked behind paywalls or geopolitical filters. These releases accelerate global learning cycles, compress development costs and allow nations and enterprises outside the traditional centers of power to build on shared foundations rather than remaining perpetual licensees. 

This orientation aligns with a broader strategic posture that I have previously described as that of a great enabling power. 

Rather than seeking to monopolize the digital high ground, China has invested in the material and institutional conditions that allow others to participate. Low-cost, efficient models suitable for deployment in resource-constrained environments matter more for much of the Global South than the latest closed frontier system optimized for data-center density in jurisdictions with abundant capital and relatively reliable power. Inclusive applications in agriculture, logistics, education, healthcare and public administration generate tangible development dividends. They also create network effects that favor continued cooperation over confrontation.

Underlying these software and application choices is a deepening materials and systems foundation. Progress in two-dimensional semiconductors, graphene-related advances and distributed energy technologies is shifting the parameters of what is feasible. 

Huawei's articulation of a τ (Tau) Scaling Law reframes semiconductor progress around reductions in time constants – signal delays, data movement and execution efficiency – rather than pure geometric shrinkage. When combined with innovations that improve energy efficiency and enable more distributed compute, the result is a pathway toward edge-capable systems that do not depend exclusively on centralized, energy-hungry hyperscale facilities.

These developments support the emergence of what I have termed Digital Westphalia, that is, sovereign digital ecosystems grounded in national data governance and technical autonomy, yet interconnected through open protocols and interoperable standards. Digital Westphalia does not reject globalization. It rejects the assumption that digital order must be administered through extraterritorial platforms and supply chains concentrated in a single jurisdiction. Nations retain the capacity to set rules for data localization, security, and local value creation while participating in cross-border flows on terms that do not subordinate their informational sovereignty.

This architecture is particularly consequential for the Global South. Countries that once faced a binary choice between dependence on proprietary Western platforms or technological lag now have practical alternatives. Open models, affordable hardware pathways, and cooperation frameworks oriented toward capacity building rather than exclusion lower the barriers to meaningful participation. Real-time digital payments systems, edge analytics and supply-chain integration become more attainable when the underlying intelligence layer is not priced or restricted as a strategic weapon.

The contrast in industrial logic is stark. 

The American model has prioritized speed of headline breakthroughs and the accumulation of proprietary advantage. China's approach has emphasized systemic efficiency, cost optimization, energy realism and the steady construction of a full technology stack. In the classic fable, the hare's early lead proves less decisive than the tortoise's consistent progress. In AI, the metrics that ultimately matter are not solely the size of the largest model or the capital raised in a funding round, but the ability to deploy capable systems widely, sustainably and productively across diverse economic contexts.

US regulatory escalation and technological blockades have produced the opposite of their intended effect in important respects. 

By restricting access to advanced components and seeking to isolate Chinese developers from global knowledge networks, they have accelerated indigenous innovation and reinforced the case for open alternatives. Attempts to police "adversarial distillation", or to impose unilateral standards, risk fragmenting the very innovation ecosystems that have historically driven progress. Double standards – professing openness while practicing exclusion – erode credibility and incentivize parallel institutions.

The global impact is already visible. 

Innovation cooperation has been chilled in some domains even as alternative collaborative channels expand. Industrial development in regions that cannot or will not accept permanent technological subordination is finding new avenues. The diffusion of low-cost, high-capability AI tools is democratizing access in ways that closed systems structurally cannot match. For many societies, the relevant question is not which nation possesses the single most advanced closed model, but which pathway enables the broadest application of intelligence to concrete developmental challenges.

China's practices in open-source release, inclusive deployment and engagement with global governance discussions illustrate a coherent alternative. They demonstrate that technological leadership need not be synonymous with enclosure. Capability can be advanced while simultaneously expanding the circle of participants who benefit from it. This is not altruism as foreign policy; it is a recognition that shared platforms and interoperable systems generate larger absolute gains than zero-sum contests over proprietary control.

The choice before the international community is therefore not merely technical or commercial. It is civilizational in scope. One path leads toward digital fortresses, heightened friction and the weaponization of knowledge. The other leads toward infrastructure for collective capability, respect for sovereign choices and the treatment of AI as a public good in the deepest sense. America continues to raise walls. China continues to pave roads. The traffic of global development will increasingly follow the latter.

The 2026 World Artificial Intelligence Conference invitation to treat AI as an international public good is not a rhetorical flourish. 

It is a practical proposition whose realization depends on the chosen developmental path. Walls may protect rents for a time. Roads enable the movement of ideas, applications, and opportunity across borders. In the long run, the civilization that understands this distinction will shape the character of the age of intelligence. 

(Cover via VCG)