OpenAI and Anthropic have quietly begun the process of going public, filing confidential S-1 documents with US regulators. European policymakers should treat this as a warning signal rather than a spectator event.
These filings force hard questions that AI hype has buried. How do these companies actually generate revenue? What capital demands sustain their operations? Which dependencies shape their survival?
The timing matters for Europe. The European Commission has proposed tripling EU data centre capacity within five to seven years through its Cloud and AI Development Act. Brussels has also committed up to €10 billion in public money for seven AI gigafactories, aiming to unlock €20 billion in private investment.
Meanwhile, Silicon Valley’s dominant model concentrates processors in massive facilities consuming enormous energy, water, and capital. A handful of companies control these systems. Yet this represents a corporate strategy, not an inevitable technological requirement.
Europe needs computing capacity. The harder question is what kind of ecosystem to build. Before committing tens of billions to hyperscale infrastructure and the power plants to run it, EU leaders must ask whether copying Silicon Valley serves European interests.
Smaller specialised models, dedicated processors, and local devices are shifting where computation happens. These distributed approaches often integrate better with renewable energy grids.
Security concerns also favour distribution. Iranian drone strikes damaged three Amazon data centres in the UAE and Bahrain this year. Some customer data proved unrecoverable. Concentrating critical infrastructure concentrates physical vulnerability.
Quantum computing offers another reason for flexibility. D-Wave reported in Science that its machine solved a complex materials problem in minutes, claiming the task would take Frontier, among the world’s fastest supercomputers, a million years. The quantum machine consumed roughly 12 kilowatts. Frontier draws about 20 megawatts, over a thousand times more.
Computing power does not always demand conventional infrastructure at hyperscale.
Data centres and transmission lines last decades. Computing technologies shift in months. Europe should avoid locking generational investment into approaches that cheaper alternatives could render obsolete.
The OpenAI and Anthropic disclosures will eventually become public. They face identical structural pressures: energy costs, compute expenses, model commoditisation, uncertain infrastructure returns. Those filings may help Europe separate technological necessity from Big Tech’s corporate preferences.
Europe entered the AI race late. That lateness offers a chance to avoid costly mistakes rather than repeat them.
