How the EU AI Act Could Shape Europe’s AI Investment Appeal

AI regulation can shape where multinational companies develop and deploy AI, but it is only one part of a location decision. Companies also weigh compliance costs and timelines against access to customers, skilled workers, computing capacity and reliable power. For Europe, the practical test is whether clear, workable rules can sit alongside the infrastructure needed to build and run AI systems.

What the AI Act asks of high-risk systems

The EU AI Act requires high-risk systems to meet safeguards before they enter the market. These include risk assessment and mitigation, high-quality datasets, activity logs, technical documentation, human oversight, and measures for robustness and cybersecurity. The requirements can support accountability, but they also mean work on testing, records and compliance processes for companies developing or supplying covered systems. The Act provides for penalties for non-compliance. 3

A trade-group estimate cited in a policy analysis puts the cost of complying with the Act at €3.3 billion a year across the EU. That figure is an estimate, not a measured total, and should be read as an interested stakeholder’s assessment rather than a definitive cost forecast. 11

The revised timetable gives firms more time—but not certainty by itself

The AI Omnibus entered into force in July 2026. Under the revised schedule, obligations for stand-alone high-risk AI systems apply from 2 December 2027, while obligations for high-risk systems embedded in regulated products apply from 2 August 2028. The delay addresses, in part, the late availability of technical standards needed to support implementation. 256

More time can help companies plan compliance and give standards work room to progress. But a later deadline does not remove the underlying obligations. Firms still need predictable guidance on how the rules apply to their products and uses if they are to budget, design and launch with confidence.

Regulation is only one part of Europe’s investment challenge

The AI Act may affect the cost and predictability of operating in Europe, but the sources here do not establish that the regulation alone determines investment location or causes companies to move elsewhere. A sounder view is that compliance sits alongside other practical considerations: access to customers and talent, the availability of computing resources, and the cost and reliability of electricity.

That distinction matters for policy. Simplifying compliance may reduce avoidable friction, but it cannot substitute for infrastructure. Public investment initiatives aimed at expanding AI capacity will matter most if they translate into usable computing resources for companies—not just announced funding or plans.

Data-centre power costs complicate the location decision

Large data centres need power, and serving new demand can require investment in generation and transmission. Who pays for those upgrades is a policy question as well as a project-cost question: US utility regulators, for example, decide how infrastructure costs are allocated between large customers and other ratepayers. 17

If data-centre operators bear more of the costs their projects create, other customers may be better protected, but project economics can become less attractive. If costs are shared more broadly, projects may face a lower direct bill while households and businesses risk carrying part of the infrastructure burden. These trade-offs mean that lowering regulatory costs alone may not make a location competitive if power supply, grid access or cost allocation remain obstacles.

Could stronger post-deployment monitoring help?

A proportionate approach could retain upfront checks for high-risk systems while making requirements simpler where risks are lower, then rely on robust monitoring after deployment. That could include logging, incident reporting, audits and corrective action. This is a policy option, not a description of a change made by the AI Omnibus.

The approach would need capable regulators and enforceable remedies to be credible. It would also need to preserve pre-market safeguards where failures could cause serious harm. A balanced system could reduce unnecessary delays without treating oversight as optional—but it would not, on its own, solve Europe’s challenges around computing capacity, electricity or grid connections.

The bottom line

The AI Act changes the compliance calculation for companies operating in Europe, and the revised deadlines give firms more time to prepare. But the available evidence does not show that regulation alone decides where multinationals invest. Europe’s appeal will also depend on whether companies can access the compute and power they need, and whether compliance is clear and proportionate in practice.