“If You Don’t Own Your Model, the Model Owns You!” Sovereign AI: What Does It Mean for a Model to Be Sovereign and Is AI Sovereignty a CIO Pipe Dream? – Part 1

If you are the CIO of a multi-national company – and we are thrilled that there are actually quite a lot of you here! – or if you are leading an AI Center of Excellence, or, indeed, if you are leading a country and your boss is at the UN Security Council this week, attending tomorrow’s AI briefing by Sam Altman, then stay tuned: this Beyond Beta briefing is particularly relevant for you! It will start structuring for you the topic that we got by far the most questions on over the past two weeks: sovereign AI.

But first: THANK YOU from Frankfurt, where I am making a brief stopover this week! Two weeks ago we started the Beyond Beta-Briefings in Oxford. Today we cracked the threshold of 1100 subscribers and counting! And the sheer number of readers is not the only thing the team and I are excited about: the briefings are only the publicly visible part of this research project. At the heart of the project is our market analysis of the sovereign tech market, our network analysis, and the comprehensive database and map of AI and tech sovereignty companies, investments, innovators, and products across Europe and – by popular demand – now also selected other parts of the world (to the many Middle East and Asian readers and contributors: we are caving in to you!). And for that heart of the project, I received tremendous input from you behind the scenes – thanks a lot!

Our guiding questions for the project are still: Who are the enterprise-scale suppliers of sovereign tech? Who really buys sovereign tech, how do multi-national companies implement it, to what effect, and if so – how much will they spend and how sovereign is sovereign enough for them? If you would like to contribute – any time!

In the past two weeks, one thing has become unquestionably clear to me: the tech sovereignty market, both the supply and the demand side, is booming.

That was not necessarily what I expected. Tech sovereignty – a perfect panel discussion and policy bubble topic: for sure! But I was not so certain if there was actually movement in the practical, multi-national, and shareholder-focused realm of CEOs and CIOs. In the past two weeks, the exponential growth of our market data master spreadsheet and the many contributors that have reached out, agreed to give interviews, help with data, product names, and/or real-life experiences, has astonished me. If we as a team have not got back to you – we will!

But now on to this week’s topic du jour: sovereign AI. And let me make two things clear upfront: first, if I wanted to spend the rest of my research journey only on this topic, it would be easy to fill the next eight weeks with nothing else.

Second, there is yet again a huge tension between what policymakers mean with the term “Sovereign AI” and what guides their actions, and what business leaders mean with the term and what they are hoping for. While Ursula von der Leyen only last week gave out AI sovereignty as the goal for Europe in her State of the Union speech, we spoke also last week with two CIOs of European companies who both proclaimed AI sovereignty was a complete “pipe dream”, totally unachievable. At the same time, both belong to the 93% of companies that would prefer a German (sovereign!) AI provider, if possible, according to bitkom. What do others here think?

So while it is not possible to tackle the topic in one briefing, let’s use this week to map the issue a bit better. That alone is complex. As with all layers of sovereignty, the deeper I dug into the topic, the easier it was to get completely lost: first, there is indeed no real definition of sovereign AI – policymakers around the world use it in a dozen different ways. The best summary of this comes from Stanford HAI: “‘Harder’ notions of AI sovereignty emphasize achieving self-sufficiency by ensuring that your country’s AI stack is made up entirely of domestic components;”, say Caroline Meinhardt and her co-authors, “‘softer’ notions frame sovereignty as strategic autonomy, where governments retain some limited degree of strategic control and regulatory leverage over their AI dependencies. The former is costly and, for the most part, unfeasible for most countries; the latter preserves flexibility but may lead to a false sense of security as AI vendors ultimately retain strategic leverage.”

On the tech and business leader side, definitions are equally fuzzy. But what is pretty clear are the fears that CIOs and CEOs have. Most of them hope, sovereign AI will resolve these fears for them: in Beyond Beta interviews, some CIOs shared with me their fears of data and IP leakage (for 45% of companies this is still a main blocker to AI usage overall, according to bitkom). Other CIOs and CEOs complain about what they consider meaningless legal texts that are supposed to protect them from vendor lock-in, from the “kill switch” of the US or Chinese governments, or from model providers changing the underlying model weights from one day to the next. Other business leaders point to complex dependencies of all layers of the sovereignty model. For example, does AI sovereignty start on the model, the data, or the application layer or even before? What about the physical hardware, like Chips and submarine cables? So the set of worries on the business side is broad.

In addition, there is an East – West divide in Europe. Western European companies see sovereignty mainly as a compliance exercise or an exercise of business continuity. Eastern European companies go beyond that: more than 30% of them see sovereignty as a business opportunity. For example, to get ready to become a supplier for defense companies.

“If you don’t own your model, the model owns you”, states Jakob Foerster, Oxford Professor, Senior Tech Lead at Meta AI, and Troublemaker in Chief from the British Open-ended Learning & Discovery Lab. We will hear from Jakob again in more detail in a forthcoming Beyond Beta Briefing.

For now, we would love to keep collecting more questions on AI sovereignty, especially from the business side. What would you like us to dig deeper into? What have we overlooked so far?

As always, ping us! Follow us along on www.beyondbeta.eu and sign up for our newsletter here or – for free and more in-depth content – directly on our website!