CAN A COUNTRY CONTROL ITS AI IF IT DOES NOT CONTROL THE CLOUD?

1 September 2026

Governments are talking increasingly about AI sovereignty. They want national AI systems, stronger control over sensitive data, and less dependence on foreign technology companies. But AI depends on cloud infrastructure, data centres, chips, software, networks, and large amounts of computing power. This creates an important question: can a country really control its AI systems if the cloud infrastructure behind them is largely controlled by foreign companies?

On 25 August 2026, UNDP announced a partnership with the DFINITY Foundation to test sovereign cloud infrastructure and decentralized AI with governments, civil society organizations, and UNDP country offices. The programme is experimental, but the policy question is important.

Governments are using AI in taxation, health services, migration, education, administration, and communication with citizens. Many of these services involve sensitive information, so governments need to know where their data is stored, who can access it, which laws apply, and what happens if access to a cloud provider is interrupted.

Europe is already confronting this problem. Much of its cloud market is controlled by non-European companies, even as the EU invests in its own AI industry. A country may control an AI model while still depending on foreign servers, chips, software, and cloud services.

For smaller countries, complete technological independence is neither realistic nor necessary. They can still protect sensitive public data, avoid dependence on one provider, require that systems can be transferred, and build enough technical expertise to understand what they are buying.

The UNDP initiative is interesting because it places this issue directly on the international development agenda. It does not provide a final solution, but it recognizes that access to AI is only one part of the problem.

A country may have access to advanced AI systems while remaining highly dependent on infrastructure that it does not control. That dependence may be acceptable in some areas, but it can become a serious weakness when it affects sensitive data, essential public services, or important government decisions.

For that reason, the debate about AI sovereignty needs to include the cloud. Governments do not need to own every part of the system, but they do need to understand who controls it, what legal and technical dependence exists, and what options remain if the relationship with a provider changes.

The goal should not be technological isolation. The real issue is excessive dependence.