AI’s biggest overlooked vulnerability is geography. The infrastructure that powers the technology is becoming a strategic asset that governments increasingly want to curb through export controls and other means.
A growing strategic mistake companies are making is choosing an AI model without thinking about who ultimately controls it. That’s why some senior executives are now starting to ask whether critical AI workloads should run on-premise or with providers closer to home, rather than in the cloud. The logic is that companies no longer want to rely on infrastructure they don’t control.
The greatest risk here is dependence. Once companies become locked into foreign AI platforms, decisions taken by overseas governments can cut off access to the AI tools those companies depend on.
That risk is real. Earlier this year, President Donald Trump's administration forced Anthropic to suspend its most advanced AI models, instantly blocking access for users outside the US. Anthropic had only 90 minutes to obey the demand, underscoring how quickly companies can be cut off from what has become critical business infrastructure.
Anthropic eventually restored the models. But the disruption was enough to prompt an urgent UK government review into the economic and security risks of losing access to frontier AI.
Governments, as well as companies, are now responding. Policymakers around the globe are increasingly investing in what Nvidia chief executive Jensen Huang calls “sovereign AI” – control over the data centres, chips, models and applications that increasingly underpin economic competitiveness.
The problem is that many countries simply don’t have credible home-grown alternatives. Nearly 90 per cent of the world’s compute, AI investment and leading foundation models are concentrated in the US and China, leaving many businesses across the world dependent on foreign AI.
In Europe, for instance, more than 80 per cent of its digital infrastructure and services are supplied by overseas companies, mostly American ones.
Much of the unease among companies stems from the US. Tariffs, export controls and other restrictions have shown how quickly AI can become entangled in geopolitics. For companies relying on the most advanced models and compute, that is becoming a genuine strategic risk.
Data security and privacy only deepen those concerns. The issue isn’t where sensitive corporate data is stored, but whose laws govern the company storing it. A US cloud provider may store European data in Europe. But in some circumstances, American authorities can still compel access to it under the US Cloud Act.
That changes how companies should buy AI. Most don’t even need the frontier model. They need the one that solves the problem in front of them and makes the business more productive. So they shouldn’t necessarily buy into the hype around the frontier.
What’s more, the technology is moving too fast to lock yourself into one AI vendor or regulatory framework. Companies need to stay flexible, which means building optionality into your AI stack, just as companies increasingly do with their supply chains by mixing up suppliers. Or in cyber security by skirting single points of failure.
Run a simple test: if your AI vendor vanished tomorrow, how much of your business would stop running? If the answer is too much, you need alternatives. The companies best placed to cope with the geopolitics of AI will be the ones with the most options.
Many companies are already moving in that direction. Research firm Gartner expects one in five cloud workloads to shift from global to local providers by 2027 because of geopolitical risk.
Companies are also becoming much more selective about where they put different workloads. Industrial group Airbus, for example, is moving its most critical applications to a French cloud provider while continuing to use US platforms for less sensitive workloads. The takeaway for other firms is that your most sensitive data deserves your most secure infrastructure.
Companies are also starting to build strategic buffers of key parts of the AI stack. Before US export controls came into force, Chinese tech companies ByteDance, Alibaba and Tencent spent billions of dollars last year stockpiling Nvidia AI chips to protect themselves against being cut off. The lesson is that if access to critical infrastructure looks uncertain tomorrow, secure it today.
Others are giving companies more control over where and how they run AI. Europe’s AI champion Mistral, for instance, is building open models that companies can run on their own infrastructure and clouds instead of locking themselves into a single provider.
The final piece is policy. Companies can redesign supply chains, rethink investment plans and switch technologies when the direction of travel is clear. But it’s the sudden reversals that create the biggest strategic risk. So, as AI becomes core business infrastructure, policy certainty may become just as important as access to the technology itself.
Naomi Haefner is professor of artificial intelligence and innovation at IMD
