87% of AI Models Come From Two Countries, It’s Time the Rest of the World Responds

As artificial intelligence reshapes geopolitics, economies and democratic processes, one reality is becoming impossible to ignore: AI is no longer just a technology conversation. It is a power conversation.

In politically sensitive environments, even subtle algorithmic changes can have outsized real-world consequences. Elections, public opinion and digital discourse are increasingly mediated by AI-driven platforms. The influence is immediate, the diffusion is rapid, and the implications are structural.

Yet beneath the acceleration lies a deep imbalance. Today, nearly 87% of notable AI models originate from just two countries, the United States and China. Venture capital concentration mirrors this divide, as does data centre dominance. From compute and infrastructure to models and deployment platforms, the architecture of intelligence remains tightly concentrated in a handful of global power centres.

For emerging economies across India, Africa, Latin America, Central Asia and Southeast Asia, this raises a defining question: will they shape AI, or merely consume it?

The Digital Capacity Paradox

Digital adoption across the Global South is accelerating at an unprecedented scale. Mobile-first populations are leapfrogging legacy systems, while AI tools are already influencing healthcare decisions, education access and financial behaviour.

But diffusion without deep digital capacity creates vulnerability. When users lack the tools to critically assess AI-generated outputs, they become more susceptible to misinformation, cultural misalignment and systemic bias embedded in models trained on foreign datasets. The paradox is stark: high adoption but low governance leverage.

Left unchecked, this imbalance risks widening the digital divide rather than closing it.

Safety as Strategy, Not Compliance

AI safety is often framed as regulatory friction. For emerging economies, that framing is misleading. Safety is not a brake, it is a strategic lever.

By setting standards around multilingual capability, data representation, model transparency and evaluation protocols, developing nations can shift from passive recipients to active demand-shapers.

Key questions must move to the foreground. What datasets trained the models being deployed locally? Do they reflect linguistic, cultural and social realities? What testing frameworks are required before AI systems are integrated into public digital infrastructure? And who audits the logic behind automated decisions that increasingly affect livelihoods and rights?

Safety frameworks create negotiating power. They allow nations to demand AI systems that reflect their populations, rather than importing intelligence optimised for entirely different contexts.

Beyond Today’s AI Systems

This debate extends far beyond current generative AI models. It reaches into the trajectory of advanced AI and, eventually, superintelligent systems.

If the next wave of AI evolves without meaningful representation from emerging economies, governance norms will become even more centralised. The race towards advanced AI cannot remain a two-country sprint.

The Global South represents the majority of the world’s population and, by extension, its largest future AI user base. That demographic weight translates into market power, and market power can influence standards if deployed strategically.

The shift required is psychological as much as technological. Emerging markets must stop asking how to use AI and start asking what kind of AI they are willing to accept.

From Borrowed Intelligence to Owned Direction

The future of AI will not be defined by compute alone. It will be shaped by standards, accountability frameworks, and the diversity of voices involved in governance. Countries that define evaluation norms, cultural safeguards, and auditability mechanisms will ultimately influence how intelligence itself is built, deployed, and controlled.

In this context, safety is not defensive; it is developmental. As AI becomes foundational infrastructure, comparable to electricity or the internet, the stakes escalate rapidly. This is no longer a debate about tools or applications. It is about sovereignty, standards, and who shapes the intelligence layer of the global economy.

The Global South now stands at a crossroads. It can remain a consumer of borrowed intelligence, or it can become an architect of its direction. The next phase of AI will not just test innovation capacity; it will test governance imagination.

That distinction may prove to be the decisive advantage.