The EU AI Fund Controversy: Why GPU Subsidies Fail to Reach Real Entrepreneurs

A Public Challenge to the EU AI Fund

Recently, an AI entrepreneur publicly spoke out on social media, leveling sharp criticism at the actual effectiveness of the EU AI Fund. The entrepreneur stated that they had applied for the EU's GPU resource support program for AI startups but never received any GPUs—nor did they know anyone who had actually benefited from the program.

These remarks quickly sparked widespread discussion in the tech community and once again thrust the EU's AI industrial policy into the spotlight.

The EU AI Fund: The Gap Between Ideals and Reality

Policy Intent: Closing the AI Gap with GPU Subsidies

The EU's intent in establishing the AI Fund was clear: in the global AI race, Europe is visibly lagging behind the United States and China. To close the gap, the EU planned to lower barriers to entry by providing computing resources (primarily GPUs) to AI startups, nurturing the development of a domestic AI ecosystem.

This approach is entirely sound in theory—GPUs are the core infrastructure for AI training and inference, and the high cost of compute is one of the biggest bottlenecks facing small and medium-sized AI companies.

Why are GPUs so critical? GPUs (Graphics Processing Units) were originally designed for graphics rendering, but their highly parallel computing architecture makes them ideal hardware for deep learning training. Modern AI chips like the NVIDIA H100 cost $30,000–$40,000 per unit, and training a medium-scale large language model often requires hundreds or even thousands of GPUs working in concert for weeks. This means compute costs have become one of the highest barriers to AI entrepreneurship, directly determining which teams can compete in cutting-edge model development.

It's worth noting that the EU has historically excelled at regulation in the digital policy space but has been relatively slow on industrial support. The AI Act passed in 2023 is the world's first systematic AI regulatory framework, but critics point out that the EU's investment in legislative regulation far exceeds its substantive support for domestic AI industry. European AI funds (including the EuroHPC Joint Undertaking and AI-specific programs under the Horizon Europe framework) were established precisely to fill this gap beyond regulation—but their execution mechanisms inherently carry the DNA of the EU's bureaucratic system.

Execution-Level Doubts: Where Did the Resources Go?

However, based on this entrepreneur's firsthand experience, the policy's execution appears far removed from its original intent. They used a particularly harsh word—"cronyism"—implying that these funds and resources ultimately flowed to groups with close ties to decision-makers, rather than to frontline entrepreneurs who genuinely need support.

From an economics perspective, this phenomenon can be explained by "rent-seeking" theory: when governments control the allocation of scarce resources, rational actors will invest their energy in influencing allocation decisions rather than in genuinely productive activities. In the EU's GPU allocation program, the backgrounds of review committee members, formatting requirements for application materials, and relationship networks with intermediary institutions can all become hidden factors influencing where resources flow. This doesn't necessarily imply corruption, but structural information asymmetry and relationship network advantages objectively cause resources to tilt toward groups that "know how to apply" rather than those that "most need support."

This criticism is far from isolated. The EU's various tech funding programs have long faced similar controversies: cumbersome application processes, lengthy approval cycles, and opaque resource allocation, ultimately resulting in "trillions of euros in funding never actually reaching the target audience."

Deeper Issues: Structural Challenges in Europe's AI Ecosystem

The Contradiction Between Bureaucracy and Innovation Speed

The EU's funding allocation mechanism is essentially a massive bureaucratic system. From project application, review, and disbursement to final implementation, multiple levels of approval are typically required. For a field like AI where technology iterates on a monthly or even weekly basis, this pace is clearly fatal.

When an entrepreneur spends months waiting for GPU approval results, the market window may have long since closed.

Take the EuroHPC Joint Undertaking as an example—this is the EU's most important computing infrastructure project, currently deploying supercomputer nodes across multiple European countries, including LUMI in Finland and MareNostrum 5 in Spain. However, these resources primarily serve academic research institutions and large enterprises, with extremely high access thresholds for startups. The application process typically requires submitting detailed technical proposals, passing peer review, and demonstrating the project's "European scientific value"—an evaluation system designed for academia that is nearly an insurmountable obstacle for commercial AI startup teams that need to iterate rapidly.

Lack of Transparency in Fund Flows

The deeper issue lies in the transparency of fund flows. EU tech funding programs typically involve multiple intermediary institutions—national-level agencies, regional distribution platforms, technical evaluation committees, and more. Each layer of intermediaries consumes a portion of resources and increases rent-seeking opportunities.

Ultimately, taxpayer money is significantly diluted through layer after layer of transfers, and the proportion actually used to support innovation is questionable.

Comparison with the US AI Startup Model

By contrast, the US AI startup ecosystem relies more heavily on market-driven mechanisms:

• Direct support from cloud computing giants: AWS Activate provides startups with up to $100,000 in cloud service credits, with applications completable in days; Google for Startups Cloud Program offers up to $200,000 in GCP credits; Microsoft's Azure for Startups is deeply integrated with its venture ecosystem. The core logic of these programs is: cloud providers subsidize early users in exchange for long-term customer relationships, aligning commercial interests with entrepreneur needs, resulting in execution efficiency far exceeding government-led allocation systems.
• Accelerator resource matching: Top accelerators like Y Combinator and a16z have established direct compute resource pipelines with cloud platforms, helping early-stage projects quickly access computing power
• Decentralized allocation: Market-driven models far exceed government-led allocation systems in efficiency

This contrast starkly reveals the enormous difference in execution efficiency between the two approaches.

Beyond the GPU Resource Dilemma: Deeper Fractures in Europe's AI Ecosystem

The GPU resource allocation problem is just one cross-section of Europe's AI ecosystem challenges. The deeper structural issue is the continuous brain drain: top European AI researchers flow en masse to US tech companies, and a few success stories like DeepMind (UK) and Mistral (France) cannot mask the overall ecosystem's weakness. According to Stanford University's AI Index Report, Europe's private AI investment is roughly one-fifth that of the US. When difficulty accessing compute, inefficient funding support, and limited market size compound, European AI entrepreneurs face systemic disadvantages that no single GPU allocation program can remedy.

Implications for Europe's AI Strategy

This incident reflects the core contradiction facing Europe's AI strategy: The EU is attempting to use top-down industrial policy to drive a field that fundamentally requires bottom-up innovative vitality.

If the EU genuinely wants to support the AI startup ecosystem, it may need to consider the following improvements:

• Simplify application processes: Reduce intermediary layers and establish fast-track channels directly accessible to entrepreneurs
• Increase transparency: Publicly disclose detailed data on fund flows and GPU allocation, subject to public oversight
• Introduce market-driven mechanisms: Partner with cloud computing platforms to directly subsidize entrepreneurs through compute credits, drawing on the proven efficient distribution models of AWS, Google Cloud, and similar platforms
• Establish feedback mechanisms: Regularly collect genuine feedback from recipients and applicants, dynamically adjusting policies

Conclusion

One entrepreneur's complaint may be just an individual case, but the systemic issues it reveals deserve serious reflection. As the global AI race intensifies, whether Europe can truly convert policy resources into innovation momentum concerns not just the fate of individual entrepreneurs, but Europe's position in the future technology landscape.

Not a single cent of taxpayer money should be wasted in inefficient bureaucratic circulation—it should be transformed into computing power that genuinely drives innovation.

Key Takeaways

• An AI entrepreneur publicly questioned the EU AI Fund's GPU allocation program, stating they never received resources and don't know anyone who benefited
• GPUs as core AI training hardware cost $30,000–$40,000 per unit, making compute costs one of the highest barriers to AI entrepreneurship
• EU tech funding programs have long faced structural problems including cumbersome bureaucratic processes, opaque fund flows, and excessive intermediary layers—economic "rent-seeking" theory explains the resource misallocation
• EU computing infrastructure like EuroHPC primarily serves academic institutions, with extremely high access thresholds for commercial startups
• Compared to the US market-driven compute allocation model, the EU's top-down industrial policy has clear efficiency disadvantages
• Europe's AI strategy needs fundamental reform in simplifying processes, increasing transparency, and introducing market-driven mechanisms