5 steps to ensure the EU doesn’t lose the second leg of the AI race

Listen to the full episode here: https://eutechloop.com/bart-de-witte-on-open-source-the-ai-race-eu-simplification-eu-inc/

In her State of the Union speech last month, President of the European Commission Ursula von der Leyen stated that the second leg of the AI race will focus on AI applications, even if the EU doesn’t have its own frontier AI models. 

The EU does have vast amounts of valuable data and capacity to build AI applications. However, due to overlapping rules, regulatory fragmentation, and data that are often untrainable and inaccessible for smaller enterprises, EU policymakers must accelerate actions to make building in Europe easier. Otherwise, the EU risks repeating the scenario from the first leg of the AI race, in which Europe lags behind the US and China.

#1 Clearly identify layers that the EU needs to own

While some celebrated von der Leyen’s frontier AI statement as sober - acknowledging that the EU does not seem likely to catch up with the United States or China in terms of AI frontier models anytime soon - Bart de Witte commented to EU Tech Loop that the EU “should not repeat the mistakes of the past” that led to Europe losing out on the first leg of the AI race. He disagrees that the EU should now own the frontier AI models, and thinks that the EU should  clearly identify which layers it needs to own to move forward and make deliberate, strategic choices - in some sectors it should be the weights, in others the hardware, in others - the openness. 

„I disagree that we should not own the first leg. I think it is important to name the layer you need to own, because if we don't own these models, we are going to make that same mistake again: the value will simply be shifted to the API holders.”

Bart de Witte also added that the EU should also make a deliberate choice when it comes to large language models (LLMs), otherwise Europeans will not only lose the ability to encode their own languages and culture, but also the way of thinking into the systems that European societies will run on. In his view, LLMs are not just technology, but also important repositories of language, culture and knowledge. 

#2 Make European data trainable

Building AI applications requires access to trainable data, yet, according to Bart de Witte, European data often remains untrainable. Simultaneously, debate over the Digital Omnibus simplification exercise is hindered by fears that relaxed rules will primarily benefit non-European firms. This resistance, coupled with other EU rules, leaves EU innovators struggling to access foreign frontier AI models available to US competitors, creating a competitive disadvantage:

“And making data trainable doesn't mean that we make it trainable only for European companies. Don't bring geopolitics into it, let these ecosystems flourish. The counterside is that we saw Meta delivering models that were not accessible in Europe, or that you could not implement in Europe because the data was not trainable and we had GDPR issues with that. We see similar tendencies with Apple and Siri”

#3 Cut the deployment time to the length of the product cycle - and revisit the AI Omnibus

European founders increasingly face situations in which they are subject to multiple overlapping rules or to double certification. In May 2026, EU policymakers tried to address this via the AI Omnibus, however, the impact was limited. The overlaps were removed only for machinery products, even though the German proposal also called for removing overlaps between the AI Act and sectoral regulations for medical devices, toys, lifts, and watercraft. Bart de Witte commented to EU Tech Loop that the failure to remove these overlaps will decrease European competitiveness in areas where the EU could otherwise be very competitive:

“I was a big proponent of the [AI] Omnibus, but in healthcare they didn't accept the Omnibus simplification, and we now have these double certification standards. I think that's going to decrease our European competitiveness in a field where we can actually be very competitive. Because, and here is where I agree with Ursula von der Leyen, we have really good data, but make the data trainable, and make these product cycles short.”

Bart de Witte further argues that deployment time must be cut to match the length of a product cycle because, in his words, while new AI models are launched on a weekly basis, navigating certification across overlapping regulations can take up to 3 years:

“Imagine you need to wait until you certify a general-purpose AI model, which probably takes 1,5 years to get certified, and then you go to the MDR certification. It takes you three years to get your product certified in a space where, on a weekly basis, we see new models. That doesn't fit. That is not a working model. That is not how we're going to be able to win this game. So I think the Commission can already switch off that duplication of AI requirements where the device rules are doing the same job, like we have in MDR.”

#4 Don’t overfocus on trade secret argument and let new innovations flourish

As reported by MLex last week, negotiations on the Digital Omnibus have stalled due to Germany's request for stronger trade secret protections in the new Data Act, which merges the Open Data Directive, Data Act, and Data Governance Act, among other frameworks. 

In Bart de Witte’s opinion, Europe must strike a delicate balance here, because with overly stringent trade secret protections, the EU risks missing out on the value and opportunities that new layers could create:

“We will not have companies that are truly innovative, and we will not see innovation coming from Europe because we prefer protecting older structures, older hierarchies, and older organizational powers, instead of allowing anybody to use all these great innovations out there and extract that value in very different ways through new business models so we can start flourishing based on that amazing data that we have.”

In practice, having overly strong trade secret protections, in Bart de Witte’s words, can slow Europe down and make every business case non-viable. In his example on longitudinal patient records, data from millions of patients could be used to predict the onset of breast cancer; however, if that data is protected by patents or trade secrets, the possibilities of doing so become minimal:

“Now, of course, these companies are trying to protect that sort of IP or patent that they have. But then again, if we start protecting every single invented asset before we are allowed to use machine learning on data, it's going to slow us down and make every business case non-viable. It doesn't make any data available.” 

De Witte also added that while insight generation has moved to consumer-grade hardware, Europe still keeps paying the scarcity price set two decades ago and protecting data as if every discovery required a moonshot. He added that while the first human genome cost 100 million dollars, today sequencing costs a few hundred and can be done on a desk-sized machine. 

#5 When protecting personal data, focus on the application layer, not the model 

EU policymakers are currently discussing proposals for the Digital Omnibus, in which the Commission initially proposed relaxing General Data Protection Regulation (GDPR) rules so that personal data would no longer be considered personal after sufficient pseudonymization. 

Since then, the proposal on pseudonymization has drawn criticism from privacy advocates and has subsequently been diluted. However, recognizing AI development as a legitimate interest legal basis appears to remain in place for now.

In Bart de Witte’s opinion, when trying to protect personal data, the difference between the model and application layer has to be recognized:

“So I find data protection important, but then shift that to the application layer. You don't need to block training—if you have these large language models, you can use that data to train, but you will always have an application layer that filters out certain outputs. <...> I have nothing against these models being trained on my data, as long as it is protected on the application layer, because that's where the liabilities lie. I think we need to talk way more about liabilities on the application layer instead of trying to forbid data innovation from happening.”

Furthermore, theoretical arguments against de-identification - claiming that even de-identified data can be re-identified - are, in Bart de Witte’s words, far removed from actual practice and are often used as a narrative to support regulatory capture: 

“I've read so many research papers about de-identification. If you look at the theoretical analysis of how, when you have access to additional information, you can re-identify data, it is so far away from actual practice that [the argument against pseudonymization] is used as a sort of regulatory capture, especially in [healthcare] domain.”