AI Act and Datacenters: What European Regulation Changes for AI Infrastructure

The AI Act does not target datacenters — yet it will transform them. By requiring model providers to document training in detail, the European regulation turns infrastructure into a link in the compliance chain. Operators able to document what runs in their facilities hold a new selling point.

A calendar reminder. The regulation entered into force in August 2024. Obligations for general-purpose AI models have applied since August 2025, and those for high-risk systems arrive between August 2026 and August 2027. Beyond 10^25 training FLOPs, a model is presumed to pose systemic risk — with reinforced obligations.

These obligations flow down the chain. The model provider must document the energy consumed by training, the type of compute used and the origin of the data. Yet all of this information physically lives in a datacenter: tomorrow, customers will ask their host for reports the host has never had to produce.

A Clause That Changes Contracts

The first effects are contractual. Hyperscalers and AI labs now embed reporting clauses in their colocation agreements: consumption per rack, hourly energy mix, logging of training clusters. A site unable to provide this data closes itself off from an entire customer base.

The reverse is also true: a datacenter that industrializes this reporting — automated records, hourly data, per-cluster traceability — can charge for compliance as a service. Documentation becomes a product, not a chore.

What This Changes for Investors

AI Act compliance creates a segmentation between documented and opaque assets. A datacenter whose instrumentation natively produces the required records captures demand from model providers subject to the regulation — soon, everyone training in Europe or for the European market.

The risk is symmetric: an asset bought without instrumentation will need retrofitting to remain rentable to AI customers, a cost that must sit in the business plan from acquisition. The question 'does the site produce its records?' joins the classic trio of power, PUE and connectivity.

How We Apply This at FINXIA

In the TITAN DC AI strategy, we now score the documentary compliance of assets: per-rack instrumentation, consumption historization, ability to isolate records per customer. Brown transformations embed this reporting from the design stage, alongside liquid cooling.

This requirement echoes our reading of digital sovereignty: customers training models subject to the AI Act prefer documented European infrastructure over opaque offshore capacity. Regulation becomes a commercial advantage for those who prepare for it.

Key Takeaways

The AI Act pushes compliance down to infrastructure: training energy, compute and data will have to be documented, and datacenters will supply the evidence. Instrumented assets win; the others will pay for the catch-up.

Worth watching: the first GPAI codes of practice published by the Commission, and the arrival of AI Act reporting clauses in colocation tenders.