Executive Summary

The European Union's AI Act remains a focal point for enterprise technology strategy, even as its enforcement timeline extends to December 2027. New research from the diginomica network of CIOs and digital leaders highlights a paradox: while European enterprises are increasingly disillusioned with costly frontier AI models, their preparedness for the AI Act's requirements lags. Only a quarter of surveyed organizations believe they are in scope, and a third have not assessed their obligations. Meanwhile, a separate trend sees enterprises moving away from "token burn" dependency on frontier models, favoring smaller, task-specific AI solutions. This dual shift carries profound implications for investment, innovation, and Europe's competitive position in the global AI race.

Introduction

The EU AI Act, first proposed in 2021, aims to establish a risk-based regulatory framework for artificial intelligence. With enforcement now expected in late 2027, European companies face a protracted compliance journey. Yet the diginomica network survey, conducted mid-2026, reveals a cautious pace: only 35% of digital leaders have begun tracking or watermarking AI-generated content, a requirement with an August 2026 deadline. Meanwhile, the narrative around frontier AI models has evolved. Enterprises are questioning the value of deploying expensive large language models (LLMs) for routine tasks, citing escalating costs and diminishing returns. This recalibration is not merely a cost-saving measure; it reflects a strategic reassessment of AI's role in business operations and European technological autonomy.

Main Analysis

Compliance Gaps and AI Literacy

The AI Act's emphasis on transparency and human oversight is manifest in provisions for AI literacy and watermarking. The diginomica survey indicates that while 38% of organizations are partially compliant with AI training requirements, only 13% claim full compliance. More concerning is that 36% of leaders have not conducted an assessment of the Act's applicability to their business. This slow response may stem from uncertainty about the Act's final provisions, but it also signals a broader challenge: integrating regulatory compliance into AI innovation strategies without stifling progress.

The Frontier Model Reassessment

The term "frontier model addiction" has entered the enterprise lexicon, describing the tendency to default to the largest, most expensive AI models. Recent market signals suggest a shift. Investment banks such as Morgan Stanley now advise that "you really don't need the latest cutting-edge incredibly expensive model to summarize an analyst report." This pragmatism is driving the adoption of smaller, fine-tuned models that deliver adequate performance at a fraction of the cost. Enterprises are also exploring open-source models to reduce dependency on US hyperscalers, raising strategic questions about data sovereignty and AI infrastructure.

Sovereignty and Strategic Autonomy

A recurring theme in European AI discourse is the risk of technical dependency on non-European AI providers. The EU AI Act, while regulatory in nature, is increasingly viewed as a tool to foster a thriving European AI ecosystem. The debate, captured in a recent Tech Policy Press article on the "Europe 2031" position paper, underscores the need for localized models and sovereign data centers. Enterprises that lag in compliance may also fall behind in building the internal capabilities necessary for long-term AI competitiveness.

Business Impact

The convergence of regulatory and cost pressures is reshaping corporate strategy in several ways:

• Investment reallocation: Capital is shifting from frontier model licensing to AI governance, compliance tools, and training. Companies are establishing internal AI ethics boards and investing in explainability software to meet transparency requirements.
• Procurement changes: Enterprises are favoring modular AI solutions — smaller models for specific tasks — over all-purpose LLMs. This trend is boosting European AI startups that specialize in domain-specific models for manufacturing, finance, and healthcare.
• Talent development: AI literacy mandates are driving upskilling programs, aligning with broader digital transformation goals. Companies that invest early in AI competencies will gain a competitive edge in attracting and retaining talent.
• Supply chain resilience: By reducing reliance on a few large AI providers, enterprises can mitigate risks related to pricing volatility, geopolitical tensions, and data residency requirements.

European Perspective

For the European Union, the AI Act represents both a regulatory benchmark and an industrial policy instrument. The emphasis on people-centric AI aligns with European values but also risks burdening companies with compliance costs. However, the Act's long timeline allows for phased implementation, and the growing disillusionment with frontier models may inadvertently boost Europe's homegrown AI ecosystem. The Nordic countries and Germany are already investing in AI for industrial applications, while France and the Netherlands focus on foundational research. Central and Eastern Europe could benefit from near-shoring AI development as enterprises seek alternative providers. The European Single Market provides scale, but fragmentation persists: member states have varying interpretations of the Act's provisions, complicating cross-border compliance.

Future Outlook

Over the next three to five years, the European AI landscape will likely evolve along the following trajectories:

• Regulatory maturity: The EU AI Act will inspire similar regulations in other regions, but Europe will remain a testing ground. Compliance will become a market differentiator, especially for B2B software vendors.
• Model diversification: Enterprises will deploy a portfolio of AI models — proprietary, open-source, and small-scale — tailored to specific use cases. The tokenomics debate will shift from cost containment to value creation, with ROI measured not just in efficiency but in competitive advantage.
• Infrastructure investment: The demand for sovereign AI infrastructure will drive investments in European cloud providers, edge computing, and energy-efficient data centers. Public-private partnerships under the European Digital Decade targets will accelerate this trend.
• Innovation ecosystems: Startups focusing on AI governance, synthetic data, and domain-specific models will see increased funding. Venture capital in European AI reached record levels in 2025, and this momentum is likely to continue as regulatory clarity improves.
• Global competitiveness: Europe may not lead in frontier model development, but it can excel in trustworthy, regulated AI. This niche could attract international buyers from sectors like healthcare, finance, and manufacturing, where compliance is paramount.

Conclusion

The EU AI Act and the enterprise move away from frontier models are two sides of the same coin: a maturing understanding of AI's practical limits and opportunities. European companies that act now to close compliance gaps, diversify AI sourcing, and invest in human capabilities will be better positioned to navigate the regulatory landscape and capture value from AI. The strategic imperative is clear — align AI investments with both regulatory demands and sustainable business models. Failure to do so risks ceding ground to more agile competitors and reinforcing external dependencies that undermine European digital sovereignty.