Europe's AI Strategy Needs Market Shaping, Not Just Subsidies

Europe's AI dependence runs deeper than policy admits

Europe's artificial intelligence ecosystem remains structurally captured by U.S. technology giants, even as Brussels pours resources into domestic AI capacity. According to analysis published by Tech Policy Press in partnership with the AI Now Institute, the continent's current strategy—subsidizing infrastructure, channeling capital into venture funds, and loosening regulations—fails to address the fundamental problem: European AI companies operate within a market where value flows inexorably upstream to American labs and hyperscalers.

The pattern repeats across sectors. When open-source platform Hugging Face partners with Nvidia, or German AI firm DeepL integrates with AWS, these arrangements reinforce rather than challenge existing dependencies. The report argues that neither consumers, businesses, nor public institutions can make meaningful choices at any layer of the technology stack under current conditions.

Why it matters

Europe's AI industrial policy currently treats technological trajectory as inevitable rather than shapeable. If the continent continues betting exclusively on frontier model access while ignoring market structure, it risks deepening dependencies just as the AI market enters a period of fundamental uncertainty—one where open-weight models, commodification, and shifting value capture could radically alter competitive dynamics.

Four strategic shifts for European policymakers

The analysis, detailed by Tech Policy Press, proposes four interconnected approaches that diverge from conventional thinking.

First, policymakers should take market uncertainty seriously. The question of whether value ultimately concentrates in applications, models, or infrastructure remains unanswered. If open-weight models reach functional parity and commoditize, value may flow to whoever controls inference compute—likely the same hyperscalers. Alternatively, AI labs might respond by moving upstream into enterprise products and agents that create new lock-in points. Each scenario demands different policy responses, yet Europe's current strategy assumes a single trajectory.

Second, Europe must actively shape market structure rather than simply respond to it. The AI market is already being shaped—by Google, Microsoft, Nvidia, Amazon, and Meta through what functions as private industrial policy. These firms subsidize computing power and build captive ecosystems that determine what the market looks like. The question isn't whether markets get shaped, but by whom. Europe possesses tools beyond subsidies and occasional antitrust cases: procurement, standardization, trade policy, taxation, prohibitions on bundling, and coordination among domestic firms.

Third, cloud dependence and AI sovereignty constitute a single problem. Most demand for AI currently means proprietary foundation models bundled with hyperscaler compute. The proposed Cloud and AI Development Act (CADA) aims to triple data center capacity but reserves only a fraction of public sector demand for European providers—effectively subsidizing the same oligopoly it claims to challenge. Europe needs to prioritize diversification of its cloud market, not generic capacity expansion.

Fourth, Europe must decide what AI is actually for. Even a market where AI is cheap, open, and competitive doesn't automatically serve public interest. The analysis distinguishes between replacement logic—where AI displaces workers to capture economic activity—and augmentation, where AI expands what workers can accomplish. This choice involves power and politics, not technological inevitability.

A collective action problem

The framework acknowledges that established European companies lack individual incentives to defect from existing arrangements. They resell hyperscaler capacity, accept investment, and participate in partnership programs. Forgoing current revenue for a market that only exists if everyone defects together represents a classic collective action problem—precisely what states are designed to solve.

These details were first reported by Tech Policy Press in partnership with the AI Now Institute.