On EdTech logo

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

On November 3, the European Commission approved a request from eight member states to pool resources for EdTech, including AI tools for schools. One explicit goal was to reduce their dependence on suppliers outside the EU. That same day, American voters favored limits on data-center development in 46 of 47 local ballot measures. In Grove City, Ohio, residents passed the Community Consent Amendment, reserving the right to vote on projects that exceed certain thresholds for land, electricity, or water use.

Why put those decisions in the same EdTech post? A month earlier, at MoodleMoot Europe in Barcelona, I kept hearing the term digital sovereignty. In one conversation it meant a university having a choice about who hosts its LMS, but in another it meant Europe having AI companies of its own. Back home, I rarely hear Americans use the term, and I doubt many Ohio voters would describe their vote that way. But community consent gets at a question I heard throughout MoodleMoot. When a technology decision matters, who actually gets to make it?

I’ve been writing about versions of that question since April, when I described some of the educator pushback I heard at ASU+GSV as anti-abdication. Those educators wanted a say in how AI would change their teaching. By June I was asking what happens when a college builds a service around an AI model and the company behind it changes the terms. The college made the original decision. How much say does it have over the next one?

I think digital sovereignty will help explain which AI tools schools buy—both direct contracts with frontier LLMs and embedded AI features in EdTech platforms—which they push back on, and what they ask vendors to change. The trouble is that few people using the term stop to define it, and the definitions I’m hearing don’t always line up. It’s hard to analyze what you can’t define. Before getting into what Europe is building or where I think AI in EdTech is headed, I want to lay out where those meanings overlap, where they diverge, and the working definition I’ll use in the posts that follow.

I expect that digital sovereignty will be a big issue and will shape what and how colleges and universities can buy EdTech and AI tools in the coming years.

A Working Definition

As I pull together the various definitions (often partial or implied), I would summarize them as follows: Digital sovereignty is the degree to which a government, community, institution, or individual has both the authority and the practical ability to make and revise decisions about the digital technologies on which it depends.

I should acknowledge that the word revise does a lot of work here. Moodle CEO Scott Anderberg traced how sovereignty evolved from the authority of monarchs to states, communities, and individuals, then brought the discussion down to a question people could use: “Who gets to decide?” A university may have good reasons to pay a company to host its LMS, but the telling moment might come years later when the company changes its terms. Can the university take its courses, records, and customizations to another provider and carry on teaching? Two universities can sign the same contract and have quite different answers, depending on their staff, budgets, and alternatives.

Cleaned up slide photo from MoodleMoot

I would also have the definition do one more job. Whose sovereignty are we talking about? A university that disables AI feedback across all courses has exercised institutional authority. The instructor who designed an assignment around that feedback has lost some discretion. Whether that was a sound educational decision is a separate question.

What Changed Since the Open Source LMS Debates?

Those of us who lived through the LMS debates of the 2000s know the familiar questions. Could we inspect the source code? Change it? Hire someone else to support it? Those questions still have fairly intelligible answers. At MoodleMoot, however, I asked a lawyer involved in the European policy discussions what open would mean for AI. Access to the application code? Access to the model weights? Controlling which model is used? There was no clear answer, and none that I can find in relevant policy documents.

Then there’s a complication the old LMS comparison misses entirely. There are many EdTech products on the market and a much smaller set of model providers behind them. Sometimes the buyer can’t even establish which model did the work. A university could have the source code for an advising tool and still lose its AI features if the model provider withdrew access. Access to the model weights—the numerical settings learned during training—could let the university or another host run that model, provided it had the license, hardware, and expertise. MagicSchool, for example, tells schools that it routes requests among models from OpenAI, Anthropic, and Google depending on the task. That tells a teacher who might have generated the feedback on a student’s essay. It doesn’t identify which model generated that feedback. An April study of public disclosures for 20 educational AI tools found that nine reported using multiple models, often without naming them. The ability of an EdTech company to switch models is one thing. Whether its customer can see or influence that decision is another.

In Europe, Sovereignty Is an Explicit Political Project

France and Germany laid out their thinking in a June joint paper intended to shape EU legislation. Their definition of digital sovereignty is close to the one I’ve offered here—it connects the practical capacity to develop, adapt, and control technology with the ability to act independently and retain “final decision-making authority.” Where the two governments go further is in the political goal they attach to that definition. Under the heading Aspiration, they put it plainly: “Europe needs to strengthen its own digital ecosystem.”

What difference does that make in EdTech? Suppose a university can switch the AI model behind its advising service from Anthropic to OpenAI and keep the service running. By my working definition, that university has gained some sovereignty—it can change its mind about a vendor and act on that decision. For France and Germany, there is a further question, however. If both of the university’s realistic choices are American, what has Europe gained? The university has gained leverage over its supplier, but Europe presumably still needs companies capable of supplying the technology itself.

Not all European countries or constituents support the Franco-German position, but the overall political agenda has found a largely receptive audience in the European higher ed circles I’ve encountered. Universities brought their own concerns about vendor lock-in, control of research and student data, and the influence of commercial platforms on academic work, and the European University Association was documenting those concerns well before this year’s proposals.

I’ll leave analysis of the details to a future post, keeping this one about the definition of digital sovereignty, but I will note that in June the European Commission put forward a package of measures to reduce Europe’s dependencies on non-EU technology suppliers. The proposed Cloud and AI Development Act would make control a factor in public purchasing, including whether a foreign government could gain access to data or disrupt a service.

Under the proposed act, Article 41 would encourage public bodies to use open standards and openly licensed components. A separate provision would require a public buyer to justify purchasing a cloud service outside the applicable EU assurance level. A proprietary service could qualify.

None of these has made it through the lengthy legislative processes.

In the US, the Data Center Is Where People Voted

Americans rarely get to vote on whether an AI company should release a more capable model or whether a software vendor should build it into a product they already use. They sometimes do get to vote on whether a facility drawing large amounts of power and water should be built down the road. That makes the data center a proxy for a much wider argument about AI.

The US movement has plenty of top-down politics of its own. Food & Water Watch describes helping residents organize for local moratoria across seven states. And according to AdImpact figures reported by AP, campaigns spent more than $380 million on television ads that mentioned data centers. At the national level, Anthropic’s Dario Amodei has argued for deliberately pacing frontier-model development to give safety evaluation time to catch up. Nvidia’s former CEO Jensen Huang has publicly opposed a slowdown, while David Sacks has argued that a coordinated slowdown could protect the leaders from competition. These arguments give Americans competing ways to interpret the same proposed facility, but a resident can also reach a view simply by looking at the electric bill.

The EdTech Angle

One detail from Barcelona has stayed with me. For all the discussion of European AI sovereignty, I didn’t encounter a session focused on Mistral, the French AI company building LLMs. That same week as MoodleMoot, Mistral announced a model it had trained in its own European data centers using 3,800 Nvidia chips. Here was a European company developing a model independently of Anthropic and OpenAI, but using American hardware. This opens up interesting questions about different ways to apply a digital sovereignty concept, and it was not discussed. At Online Educa Berlin (OEB) in early December, I will be looking for these conversations.

In June I wrote about what changed when Amazon opened its Frankfurt cloud region. Suddenly a university that wanted its data hosted in Germany had an option it could put in a contract. What would the equivalent look like for AI? Suppose a university wants to keep its advising service but switch away from Claude. Somewhere there has to be another model that can do the work, someone able to run and support it, and enough computing capacity to get through registration week at a price the university can pay. Who gets to decide is the first question. What could they realistically choose is the second.

On EdTech+ Team access

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.