Week 20
EU Digital/AI Omnibus: when simplification benefits American Big Tech
On May 7th 2026 agreement on the EU Digital/AI Omnibus marks a turning point, but above all a victory for Big Tech against the data protection efforts for EU businesses and citizens. Behind the official narrative of “regulatory simplification” and the announced 150 billion euros in savings, lies a reality far more favorable to American tech giants than to European players.
Let’s start with the most explosive element: Article 88c.
This new provision introduces a “legitimate interest” legal basis allowing personal data to be used for training artificial intelligence models, without the explicit consent of European citizens. Meta didn’t wait: the company has already obtained the green light to train its LLMs on European users’ data. OpenAI and Google are rushing through the breach. What the GDPR had spent eight years building as a protective shield has just been pierced.
Next, the timelines.
The AI Act obligations for high-risk systems have been pushed back to December 2027 (Annex III) and August 2028 (Annex I), compared to August 2026 originally. Eighteen to twenty-four additional months granted to Big Tech to comply — a considerable competitive advantage over European startups that had already invested in early compliance.
How did we get here?
The Corporate Europe Observatory investigation provides a chilling insight. In 2025, Big Tech spent 151 million euros lobbying EU institutions. I have heard (but can no longer find the source) that there are more Big Tech lobbyists than members of parliament.
The Commission adopted the positions of the American tech industry in at least seven key instances of the final text. The alignment between the demands of Silicon Valley giants and the adopted text is, to say the least, troubling.
For European businesses, the outcome is ambiguous. Administrative simplification promises real savings, and extending SME exemptions to small mid-cap companies is welcome. But let’s be clear-eyed: the legitimate interest basis for AI training primarily benefits companies that already hold massive volumes of data — that is, American platforms, not our startups.
On the citizens’ side, Amnesty International is sounding the alarm over a rollback of fundamental rights. The notification deadline in the event of a data breach goes from 72 hours to 96 hours, with the threshold raised to “high risk” only.
Data collection for AI is expanding without proportionate safeguards. The only notable positive: the ban on non-consensual intimate content generated by AI, effective from December 2026.
Transparency obligations for general-purpose AI models (GPAI) are also being relaxed. OpenAI with GPT, Google with Gemini, and Meta with Llama directly benefit from this increased flexibility. The GPAI Code of Practice, which was supposed to govern these models, sees its scope reduced.
The fundamental question raised by this Omnibus is that of normative sovereignty. Europe had turned its technology regulation into a competitive advantage and a democratic shield. By yielding to American lobbying pressure, it risks losing both. Simplification was necessary, but not at any cost.
Next week, we will closely follow the publication of the EU Tech Sovereignty Package expected on May 27. It will tell us whether Europe intends to regain control — or whether the Omnibus was merely the beginning of a long series of concessions.
Nicolas Bombourg, Ai-Partner
Editorial Sources
1. Corporate Europe Observatory — How Big Tech shaped the EU’s roll back of digital rights
2. TechPolicy.Press — What the EU AI Omnibus Deal Changes
3. Amnesty International — EU Simplification Laws
4. Seresa.io — EU Digital Omnibus helps Google and Meta train AI models
5. EU Council — Official press release on the May 7 agreement
6. IAPP — EU Digital Omnibus Amendments to GDPR
7. Taylor Wessing — EU Digital Omnibus on AI
This Week’s Key Stories
Nscale raises $2 billion — the largest European AI infrastructure funding round
Nscale, a neocloud provider founded in May 2024 by Josh Payne (CEO) and Nathan Townsend in London — spun off from Arkon Energy, a specialist in crypto infrastructure powered by renewable energy — closed a $2 billion Series C led by Aker ASA and 8090 Industries, with participation from NVIDIA. Valued at $14.6 billion, the company will fund sovereign data centers in the UK, Norway, Iceland, and Portugal, positioned as a European alternative to American hyperscalers for hosting AI workloads.
GPUaaS: the illusion of European AI sovereignty?
An analysis published by The Next Web (TNW) in May 2026 dissects the GPU-as-a-Service model in Europe, drawing on Gartner market data and hyperscaler CapEx figures. The finding: European sovereign cloud spending reaches $12.6 billion in 2026 (+83% according to Gartner), but remains negligible compared to the $725 billion in combined CapEx from Google, Amazon, Meta, and Microsoft over the same period (+77% vs. 2025). The analysis highlights that GPUaaS broadens access to compute without altering the power structure: GPUs are still designed by NVIDIA (American) and manufactured by TSMC (Taiwanese), leaving Europe dependent for both hardware and pricing.
IBM Sovereign Core now available in EMEA — but Amazon has taken the lead
IBM launches the general availability of Sovereign Core, a software platform enabling AI and security workloads to be deployed within sovereign borders — on its own hardware, at local providers, or even on other clouds. The first partners are Cegeka (Belgium/Netherlands) and Computacenter (Germany). IBM’s approach is purely software-based, which sets it apart from the competition. But Amazon has already taken the lead: AWS launched its European Sovereign Cloud in Brandenburg (Germany) in January 2026, with a €7.8 billion investment, physically separated infrastructure operated by EU residents, and planned expansions in Belgium, the Netherlands, and Portugal. However, AWS’s sovereignty remains questionable: only ~90 services available (vs. 240+ in standard regions), no GPU instances, a ~15% surcharge, and the persistent legal risk associated with the American CLOUD Act despite German-law entities.
Amp raises $1.3 billion to democratize access to AI compute
Amp, a startup founded by Anjney Midha — a former partner at Andreessen Horowitz — and based in Menlo Park (California), has raised $1.3 billion to build an “AI compute grid” redistributing computing capacity to startups and universities. The model: acquire surplus capacity from data center operators (US and Europe) and make it accessible to those who cannot afford to contract directly with hyperscalers. Target: 1.9 GW of capacity within five years, including 200 MW operational by end of 2026. Mistral, ElevenLabs, and Black Forest Labs are among the project’s supporters.
Iliad bets €3 billion on AI — from data center to subscriber
The Iliad Group (parent company of Free) is investing €3 billion in AI, of which €2.5 billion is earmarked for data center infrastructure through its subsidiary OpCore. OpCore operates 15 data centers across Europe (Paris, Marseille, Poland) and plans, in partnership with Infravia, to deploy several hundred megawatts of capacity in the short term with an ambition of several gigawatts in the long term. The AI campus at Bruyères-le-Châtel is entering its operational phase with 13,800 NVIDIA GB300 GPUs. The stated objective: to position Iliad as the European leader in sovereign hyperscale, from the infrastructure stack all the way to subscriber services.
Study in Focus
GITEX AI Europe — Blueprint for European Digital Sovereignty
The whitepaper published on the occasion of GITEX AI Europe provides an uncompromising assessment of European digital sovereignty. The analysis is structured around four strategic pillars: sovereign compute, cloud, open source, and capital.
First alarming figure: non-European hyperscalers — primarily AWS, Microsoft Azure, and Google Cloud — control approximately 70% of the cloud market on the continent, creating a massive structural dependency in the hosting of European critical data and services. This concentration means that the majority of European AI workloads are executed on infrastructure subject to foreign jurisdictions, notably the American CLOUD Act.
On the funding side, the imbalance is equally striking: only 5% of global venture capital flows are directed toward the European tech ecosystem. By comparison, the United States and China each capture $650 billion in annual AI infrastructure investment, versus roughly $200 million for Europe. The United Kingdom leads with over 18 billion euros in cumulative AI investment over 2020–2025, followed by Germany (8 billion) and France.
The study identifies neocloud providers as an essential lever for reconquest. These players offer sovereign GPU resources at costs significantly lower than US hyperscalers, while strictly complying with European data residency laws. Nscale, with its record-breaking $2 billion raise and $14.6 billion valuation, illustrates the emergence of this credible alternative.
Third pillar: AI Factories are multiplying across Eastern Europe. The Czech Republic has inaugurated CZAI in Ostrava, Poland is launching its second AI Factory in Krakow, and Romania is investing over 70 million euros in two facilities planned for 2026. The EuroHPC Joint Undertaking framework now operates 14 supercomputers and 19 AI Factories, supported by approximately 10 billion euros in combined funding over 2021–2027.
Finally, open source is positioned as the fourth strategic pillar. The study highlights that European open-source AI labs have attracted over 300 million euros in investment, confirming that transparency and technological independence go through open models. The launch of the European multilingual model Domyn, trained on CINECA’s Leonardo supercomputer, is a concrete example. An action plan that contrasts sharply with the regulatory concessions of the Omnibus.