Europe's ambitious public compute infrastructure investment is solving the wrong problem. While the European Union has committed billions to AI factories and gigafactories, the startups actually using this infrastructure have fundamentally different needs than the frontier AI model developers the system was designed to serve. A new policy analysis reveals that applied AI startups in regulated sectors like healthcare and defense have genuine compute demands, but they're being overlooked in favor of a strategy optimized for large language model (LLM) developers.

What Are AI Factories, and Why Are They Failing to Meet Real Demand?

The European Union has selected nineteen AI factories as part of its tech sovereignty strategy, with several AI gigafactories announced to follow. These facilities were conceived as dynamic ecosystems built around AI-optimized supercomputers, designed to provide computing resources and support services to European industry and scientific users developing large AI models. However, the underlying assumption that commercial demand would follow supply has proven flawed.

The debate over public compute investment in Europe sits between two opposing views. One camp argues that public compute capacity is essential for European AI competitiveness and technology sovereignty. The other questions whether sufficient private-sector demand exists to justify the scale of investment committed. Critics often generalize from the trajectory of a small number of frontier-LLM developers to the entire universe of AI firms, concluding that since few such firms are emerging at scale in Europe, public compute investment is misdirected.

But this framing misses a crucial distinction. Firms building frontier AI, such as France's Mistral.AI, which recently secured $830 million in debt for its own data center buildout, represent only one segment of the AI economy. The real opportunity lies elsewhere.

Which AI Startups Actually Need Public Compute Infrastructure?

A detailed examination of Poland, the fourth-largest national recipient of EU AI factory funding and an emerging innovator according to the European Innovation Scoreboard, reveals where genuine demand exists. Researchers conducted interviews with one AI factory host, four ecosystem enablers including industry associations and research bodies, and five AI startups operating in markets considered strategic for the EU.

The findings challenge the prevailing narrative about European AI compute needs. Applied AI startups in regulated and strategic sectors have real, sector-specific demand for public infrastructure. These sectors include:

• Healthcare and Life Sciences: Startups developing AI applications for medical diagnosis, drug discovery, and patient care require substantial computing power and strict data protection guarantees.
• Space and Defense: Companies building AI systems for defense and aerospace applications need secure, sovereign compute infrastructure with robust intellectual property protections.
• Regulated Industries: Any sector with stringent regulatory requirements benefits from public compute that can guarantee data sovereignty and compliance.

The critical insight is that generalizing from frontier-LLM developers to the entire universe of AI firms misrepresents actual demand. These applied AI startups are not building the next ChatGPT; they are building specialized tools for high-value, regulated markets where European companies have genuine competitive advantages.

What Do Applied AI Startups Actually Want From Public Infrastructure?

The research identified concrete requirements that applied AI startups have from public compute infrastructure. These are not abstract wishes but practical operational needs that, if met, could unlock significant value from public investment.

• No-Cost Access with Flexibility: High-performance computing access at no cost is welcomed, provided it is unbureaucratic, time-sensitive, protects data and intellectual property, and enables easy switching between AI factories and commercial cloud providers.
• Human Support and Expertise: Localized human support and access to relevant datasets are equally valued as raw computing power, helping startups navigate technical challenges and accelerate development.
• Data Sovereignty Guarantees: Clear, codified sovereignty guarantees are essential for startups operating in regulated sectors where data residency and protection are non-negotiable compliance requirements.
• Portability and Interoperability: The ability to move workloads between different AI factories and commercial providers without friction is critical for reducing vendor lock-in and managing risk.

These requirements reveal a fundamental mismatch between what policymakers built and what the market actually needs. The current infrastructure was optimized for frontier model training, not for the operational realities of applied AI development in regulated industries.

How to Redesign Public Compute Infrastructure for Real Impact

Policymakers and AI factory operators can address the identified obstacles to unlock genuine value from public investment. The key is shifting from a one-size-fits-all approach to infrastructure tailored to different types of AI innovation.

• Secure National Investment Portions: Establish dedicated compute capacity reserved for applied AI startups in strategic sectors, ensuring that public resources are allocated to firms with genuine additionality needs rather than those that would secure private funding regardless.
• Bridge Academic and Commercial Cultures: Create operational frameworks that allow AI factories to function as both research institutions and commercial enablers, with clear processes for intellectual property protection and commercialization support.
• Codify Sovereignty Guarantees: Develop explicit, legally binding frameworks that guarantee data residency, processing location, and compliance with EU regulations, removing uncertainty for startups in regulated sectors.
• Enable Portability and Interoperability: Design technical standards and contractual terms that allow startups to move workloads between AI factories and commercial cloud providers without friction, reducing switching costs and vendor lock-in.

The central implication for policymakers is one of design: public compute infrastructure tailored to the requirements of applied AI startups, rather than optimized for frontier model training alone, is well positioned to translate public investment into durable value for European AI innovation.

What Role Do Gigafactories Play in This Strategy?

The EU has announced several AI gigafactories, which represent a significant capacity upgrade from current AI factories. These facilities are expected to concentrate far larger compute resources and are still anticipated to be selected through a European tender, with selection now expected no earlier than the fourth quarter of 2026.

However, current applied AI startup priorities do not center on gigafactory access. The operational and technical findings from the Poland case study extend to gigafactories under broadly the same conditions. The more foundational challenges, including energy constraints and the need for human support and data access, sit at the system level and cannot be solved by compute capacity alone.

This suggests that gigafactories, while important for Europe's long-term AI competitiveness, should not distract from the immediate need to make existing AI factories more responsive to applied AI startups. The two initiatives serve different purposes and should be designed accordingly.

Why This Matters for European AI Competitiveness

Europe's AI compute strategy represents one of the largest public technology investments in recent years. Getting the design right is crucial not only for return on investment but also for Europe's ability to compete globally in AI-driven innovation. The current approach risks funding infrastructure that sits underutilized while the startups that could generate genuine commercial value struggle to access the support they need.

The Poland case study is particularly instructive because it tests whether public AI compute investment can generate genuine additionality in an environment where such support may be particularly needed. Poland is an emerging innovator with recent dynamism in AI startups but without the established frontier AI ecosystem of countries like France or Germany. If public compute infrastructure can boost dynamic ecosystems for AI in emerging innovation hubs, it validates the EU's broader ambitions. If it fails to do so, the investment risks becoming a subsidy for research rather than a catalyst for commercial innovation.

The path forward requires policymakers to shift their focus from building the largest possible compute infrastructure to building the right infrastructure for the right firms at the right stage of development. For applied AI startups in regulated sectors, that shift could unlock billions in commercial value and establish Europe as a leader in practical, deployable AI innovation.