Resilient economies require robust artificial intelligence (AI) ecosystems. Questions of AI sovereignty matter to national policymakers seeking to drive economic growth, push at the frontiers of scientific discovery, mitigate geopolitical risks, and sustain inclusive and safe AI for all. And they matter to enterprise organisations, too, and the senior IT and business professionals who lead them.
Specifically, senior leaders need to balance sovereignty requirements with access to leading AI capabilities from around the world. They need to consider where AI models, data, and infrastructure are developed, hosted, and governed. And they need to confront the implications for regulatory compliance, cybersecurity, resilience, and control over intellectual property.
It was in this context that Tech Monitor—in partnership with AMD—brought together a group of senior IT and business leaders for a candid, peer-level discussion on the consequences of sovereign AI.
Here are some of the key takeaways from the executive roundtable that took place in early October.
What is AI sovereignty, anyway?
The conversation began with the basics or, perhaps more accurately, a baseline understanding of what AI sovereignty means in practice. To that end, AI sovereignty is the ability of an organisation (or country) to control where its AI systems, data, models, and computing infrastructure are located, how they are operated, and who can access or influence them. It is about retaining control where possible while reducing unnecessary dependence on external providers.
The key principles of AI sovereignty are, therefore, access and control. And one way to ensure both, suggested one speaker, is to embrace open standards. Open, rather than closed, models are more easily auditable and allow for greater freedom.
Offering a counter view, one attendee noted that open models introduce “an element of liability you don’t have with closed models … With freedom comes responsibility.” Another argued that there is a trend away from open models and that the market leaders of today are likely to be the de facto choice of tomorrow. “First mover advantage matters,” he said.
In response, speakers pointed to activity that suggests there are options beyond the big frontier models. These include OpenEuroLLM—a series of foundation models promising “transparent” AI in Europe—and a €10 billion initiative designed to fund seven European AI gigafactories. One speaker drew comparisons with Airbus, the aerospace business set up explicitly to compete with Boeing.
Getting the balance right
An enterprise organisation’s relationship with AI is one of compromise, argued one attendee. It is about finding the necessary equilibrium, one which will provide resilience but not at the expense of innovation; access to the best systems, models, and infrastructure but not at the expense of control; and an opportunity to build bespoke models but not at the expense of profitability.
In the search for a balanced approach, some principles can be discerned. For example, AI sovereignty does not necessitate that an organisation own and operate everything itself. Rather, it means understanding key dependencies, maintaining meaningful control over critical data and workloads, and having credible alternatives when relying on external providers.
Portability? No problem?
Of the many issues thrown up by discussions of sovereignty, portability nears the top of the priority list. Portability prompts a number of questions that are similar to those raised by the move from on-premises data centres to cloud computing. For example, can workloads be moved between clouds or models if a supplier becomes unavailable or unacceptable? Who owns and controls the underlying AI models? And can organisations customise these models and infrastructure, or move to another provider?
In considering those questions, the conversation inevitably returns to issues of openness and the merits of open source (see above). Broadly, however, guests appeared relatively relaxed about the ability to move from one system to the next, and the ability to swap out one processor in favour of another. “It’s getting easier and more standardised,” noted one speaker. Another said: “Everything that goes out of your system can be abstracted away.” Meanwhile, a dissenting voice observed that “model retrofitting is a non-trivial task”. It is, he said, a form of “technical debt”.
The governance challenge
The search for control asks another question of IT and security leadership: namely, how do they satisfy the desire of the business to embrace the AI opportunity at speed while putting the necessary guardrails in place? As one speaker representing a leading retailer noted: “Our business is asking us to speed ahead, and we are trying to keep up.”
In an effort to manage expectations (and to manage upwards), another speaker recommended a “simple framework” that attempts to demonstrate the range of options available to enterprise organisations. Providing a scale from zero to four—where those on zero depend entirely on third-party providers and those on four exclusively use their own solutions—senior leadership can begin to plot how reliant or otherwise they are willing to be on external provision.
The same number on the scale needn’t be applied to every function, noted another speaker. Rather, organisations should choose the degree to which they want to control their technology stack based on use case. For example, applications that draw on personally identifiable information and/or intellectual property are likely to necessitate more control than those that don’t.
In search of best practice
The evening ended with attendees sharing advice, guidance, and some lessons learned. One urged organisations to avoid starting with “a blank piece of paper” when creating AI governance guidelines. Yes, the “velocity” of AI may be different but “there are core frameworks on risk that already exist. Use them.”
Picking up on the theme of balance, discussed throughout the night, one speaker argued that we should not let “perfection be the enemy of the good”. This advice plays to the risk appetite organisations are willing to accept, as well as the means of calibrating control and resilience.
Another guest said that enterprises shouldn’t “underestimate the importance of the network”, noting that from the centre to the edge AI is making new demands on our network infrastructure. In a similar spirit, another said that “data and security are key” to successful AI implementation. On the former, she urged organisations to get their data in shape before applying it to AI models. The goal, she said, is “a single source of truth”.
Finally, the search for control of AI was compared to our relationship with the seas that surround us. Leaning into the metaphor, the speaker said: “We can’t control the tides but we can navigate the oceans.”
‘What Sovereign AI Means to the Modern Enterprise’— a Tech Monitor / AMD executive roundtable discussion—took place on Thursday 1 October 2026 at Pavyllon, Four Seasons Hotel, London.
