Europe Simplifies Its AI Act: Is the Expected Turning Point Here?

On 7 May 2026, the European Parliament and the Council of the EU reached an agreement on the Digital Omnibus on AI — a substantial revision of the AI Act, adopted barely two years earlier. The original Act was intended as a major step in European AI governance, establishing a framework based on risk and the protection of fundamental rights. Yet recent debate on regulatory burden and compliance costs, notably reflected in the Draghi Report, has highlighted how this framework can become a significant obstacle to the competitiveness of European AI companies.

Rewriting a major law before it has even fully entered into force is an admission. Not of failure — but of a mismatch between the pace of technology and the pace of legislation.

This agreement is good news. But to understand what it actually changes — and what it does not — it must be placed in context: the coordinated industrial pressure that preceded it, and the structural issues that remain unresolved.

What the agreement resolves: operational frictions

The measures adopted are concrete and respond point by point to demands expressed for months by European industry.

Deadlines have been significantly pushed back. Stand-alone high-risk AI systems — biometrics, recruitment, education, border control and law enforcement — will have until 2 December 2027 to comply, an additional 16 months. For AI systems embedded in regulated products such as lifts, toys, medical devices and machinery, the date moves to 2 August 2028.

SME relief is extended to mid-sized companies. Businesses with fewer than 750 employees will now benefit from simplified documentation and adjusted penalties previously reserved for SMEs. An intermediate category — small mid-caps (SMCs) — absent from the original AI Act, enters the text.

The most difficult regulatory overlaps are addressed. This was probably the most technical and tangible friction for manufacturers: an AI system incorporated into machinery, a medical device or a lift could be subjected to duplicate obligations. The agreement creates a mechanism to resolve such conflicts and fully exempts machinery from the AI Act’s direct applicability. Industrial systems will be governed by sectoral safety rules, with safeguards designed to ensure an equivalent level of health and safety protection.

The AI Office sees its remit clarified and strengthened. Oversight of general-purpose AI models is better coordinated with national authorities, reducing fragmentation between Member States. The EU AI Office will provide central supervision of general-purpose AI systems, while national authorities retain their roles in areas such as law enforcement, border management, judicial authorities and financial institutions.

A new prohibition protects citizens. AI ‘nudification’ applications and artificially generated child sexual-abuse material are explicitly prohibited by Article 5 of the AI Act — a welcome measure, independent of any competitiveness agenda.

The context: unprecedented industrial pressure

This agreement did not emerge out of nowhere. It responds to acute concern across the digital sector over the practical difficulty of applying the AI Act and its negative impact on Europe’s ability to develop AI.

In July 2025, 45 major European companies, including Airbus and AXA, called for a regulatory pause. In September, 56 AI players, led by Mistral AI, demanded urgent simplification. The Draghi Report put the annual cost of regulatory compliance in Europe at €500 billion.

The strongest signal came days before the agreement: the CEOs of seven major European companies — ASML, Airbus, Ericsson, Mistral AI, Nokia, SAP and Siemens — co-signed an op-ed published simultaneously in the leading economic newspapers of eight European countries, following a meeting with Commission President Ursula von der Leyen.

Their message was specific: ‘More than three years after the “ChatGPT moment”, Europe is still debating regulation, while others have long since moved to large-scale deployment in physical systems and robotics.’

Industry accepts that the AI Act addresses a real need to manage AI risks. It nevertheless argues that Europe’s regulatory tempo is misaligned with the speed of global technological competition.

What the agreement does not resolve: the deeper divide

The Digital Omnibus responds to industry’s operational demands. But the seven CEOs’ op-ed also pointed to something more fundamental — and here the agreement remains silent.

The capital divide. In 2025, European companies invested roughly $8 billion in AI. US companies invested $109 billion. This 1-to-13 ratio is not a matter of regulatory timing; it concerns the structure of capital markets, appetite for risk and the ability to mobilise funding at the speed at which technology evolves.

No easing of the AI Act will close that gap. Europe would need a transformation in how it finances innovation — a challenge of an entirely different nature.

The industrial divide. Europe has global sector champions: Airbus in aerospace, ASML in semiconductors and SAP in enterprise software. But it has no general-purpose AI platforms comparable to Google, Microsoft or Alibaba. One structural reason is that European competition rules were designed to protect consumers from dominant positions. At the same time, they have sometimes made it difficult to achieve the consolidation needed to create actors capable of competing with giants benefiting from economies of scale and resources of an entirely different order.

The CEOs’ op-ed explicitly called for merger-and-acquisition rules that allow European companies to reach the capital scale required to compete in AI. The Digital Omnibus does not address this issue.

The paradigm divide. It is the deepest one, and the least easily solved through legislation.

The AI Act was designed through a risk-management lens: identify dangerous uses, frame or prohibit them, and protect citizens. That is a legitimate and important logic. But it is a defensive logic, focused on preventing harm.

In a sector where the rate at which models learn, the scale of training datasets and available computing power are decisive competitive advantages, a regulatory approach primarily adverse to risk can create a structural disadvantage for those who comply — compared with competitors operating under radically different frameworks.

The real question Europe must ask is not only ‘how can AI harms be avoided?’ but also ‘how can AI become the engine of the next wave of European growth and sovereignty?’

This intellectual shift — from protection to production — may be the most underestimated issue in the current debate.

Three decisive workstreams for what comes next

The 7 May agreement is a necessary condition for developing a European industrial base in AI, but not yet a sufficient one. These are the next steps if this political gesture is to become a genuine strategic turning point.

This is the next concrete milestone, and it is decisive. The package should address State aid for computing infrastructure, support for the European semiconductor value chain and potentially dedicated financing mechanisms for sovereign AI. If it delivers substantial financial commitments and a coherent industrial-support framework, the signal to the ecosystem will be strong. If it becomes lost in negotiations between Member States, the credibility of Europe’s ambition will be seriously weakened.

This is not about abandoning antitrust; it is about adapting it. The Draghi Report made the point clearly: European M&A rules are not calibrated for today’s geopolitical context, in which Europe’s principal technological competitors benefit from massive state support and much more integrated domestic markets. Allowing European consolidation that creates actors able to carry weight globally, while maintaining real vigilance against abuse, is a difficult but necessary balance.

This is probably the most powerful and least spectacular lever. An European AI startup seeking to operate across the continent must today navigate 27 distinct data-protection regimes, 27 public-procurement markets and 27 tax administrations. This fragmentation has no simple regulatory cure; it requires sustained political will over time. Yet its competitive cost is considerable: it makes it structurally difficult to achieve the economies of scale needed to compete.

Conclusion: can Europe change its posture?

The 7 May agreement sends a positive signal: Europe has shown it can react quickly under industrial and geopolitical pressure. Rewriting a major law within six months, through an accelerated procedure, is unusual — and reflects collective clarity about what is at stake.

But adapting is not yet choosing. Choosing would mean collectively accepting that technological competitiveness is a sovereignty issue, on a par with defence or energy — and devoting comparable resources, governance and vision to it.

Europe has talent. It has world-class industrial companies. It benefits from a single market of 450 million people. It counts champions such as Mistral AI and ASML, as well as hundreds of remarkable startups innovating under difficult funding conditions.

What it may still lack is the shared conviction that producing AI at scale — not merely regulating it — is an existential priority for the decades ahead.