Will AI tear down the wall or build another?

From Silicon Valley to Shenzhen, a new digital curtain is descending across the world.

The next Cold War may not be fought over nuclear arsenals. Its strategic assets are energy, rare minerals, algorithms and talent. Washington now openly speaks of "winning the AI race"; Beijing argues that AI should be developed as a global public good. Between them sits most of the world, facing a question more consequential than which chatbot to use: whose technological ecosystem will shape its economic sovereignty?

This rivalry is real, but the Cold War analogy is incomplete. The US and China are competing for technological advantage and diplomatic influence. Yet they remain embedded in a global semiconductor system that runs through Taiwan, South Korea, Japan and the Netherlands, while private companies exercise power once reserved for states. This is less a return to the 20th century than the birth of a new form of techno-geopolitics.

The contrast between the two governance models is increasingly clear. The American approach is market-led, investment-heavy and deliberately deregulatory. President Donald Trump's AI Action Plan prioritises faster innovation, domestic infrastructure and the export of "full-stack" American AI to allies and partners. The rationale behind this is that technological leadership produces economic and military power, and global adoption of American systems extends that leadership.

China follows a more state-coordinated model. Its rules place greater emphasis on data security, content control, national sovereignty and government oversight, while Beijing simultaneously promotes open-source models and international AI cooperation. At the 2026 World AI Conference, President Xi Jinping called for openness, open-source collaboration and assistance to the Global South, while stressing that AI must remain "secure and controllable". China also announced 5,000 AI training opportunities for developing countries over five years.

Which model lowers barriers for developing economies? For now, China may have an advantage at the entry point. Cost-efficient open models such as DeepSeek and Qwen can reduce dependence on expensive proprietary systems and allow countries to adapt AI to local languages and sectors. The World Bank similarly argues that open-source technologies can help developing economies localise AI without rebuilding foundation models from scratch. America, however, offers the deeper capital markets, leading cloud platforms and a powerful innovation ecosystem. Its weakness is that access can become entangled with export controls, strategic alignment and vendor dependence. China's weakness is different: cheaper access can still carry governance, cybersecurity and political-dependence risks. Albeit, neither model is charity.

The numbers reveal why smaller states should pay attention. The World Bank says high-income countries account for 87% of notable AI models, 86% of AI start-ups and 91% of venture-capital funding, despite representing only 17% of the global population. They also host 77% of global co-location data-center capacity; low-income countries hold less than 0.1%. UNCTAD warns that 118 countries remain unrepresented in major AI-governance discussions. This is an emerging digital-sovereignty divide.

Yet infrastructure does not translate mechanically into dominance. Stanford's 2026 AI Index reports that the US hosts 5,427 data centers, which is more than ten times that of any other country, and attracts roughly 23 times China's private AI investment. Still, the performance gap between leading American and Chinese models has effectively closed.

Nvidia chief Jensen Huang has argued that every country needs to "own the production of their own intelligence". Political scientist Ian Bremmer goes further. He warns that technology companies increasingly exercise a form of sovereignty in digital space. Together, those observations capture the new hierarchy: power is shifting not only between states, but also from states to the firms controlling models, infrastructure and platforms.

This is why neutrality will become harder. A government may remain diplomatically non-aligned, but once its institutions depend on one country's cloud, chips, models and cybersecurity architecture, neutrality becomes largely rhetorical. Technology stacks create switching costs; standards create path dependence; data localisation creates jurisdictional consequences.

The sensible response for smaller states is therefore not neutrality but strategic multi-alignment. They should avoid exclusive dependence on either AI-bloc, require interoperability and data portability in public procurement, invest jointly in regional compute, strengthen domestic data-protection and cybersecurity regimes, train local AI talent, and preserve the legal right to switch vendors and models. Open-source systems should be treated as strategic infrastructure, not merely an inexpensive software.

UN Secretary-General António Guterres recently warned that the compute, data and talent behind advanced AI are concentrated in only a handful of companies and countries, turning the AI divide into a development, security and sovereignty gap. That is the central geopolitical danger.

The countries that succeed will be countries that retain the capacity to choose their technology, control their data, develop their people and switch between technological ecosystems without surrendering their sovereignty. In the twentieth century, non-alignment meant avoiding military blocs. In the twenty-first century, technological non-alignment will mean avoiding technological dependence. And that is where the real geopolitical debate on artificial intelligence should begin.