Data centers are the physical foundation of the AI revolution. Behind every training run, every inference call, and every agentic task lie highly optimized compute installations — scaling to sizes never seen before and pushing the limits of power, cooling, and networking. In an AI-driven economy, the ability to build, supply, and operate them is becoming a decisive competitive and national advantage. For corporate leaders, infrastructure investors, and policymakers alike, understanding the dynamics of this build-out race — and what could disrupt it — has never been more urgent. This study is the first of two reports on the AI data center ecosystem.
The scale of today's AI infrastructure investment is without recent precedent. Hyperscalers are committing capital at a pace that is fundamentally reshaping global construction, energy, and supply chain markets. Training and deploying large language models demands compute infrastructure optimized to the highest level — physically dense, power-intensive, and built on highly specialized facility design ranging from advanced liquid cooling systems to high-voltage grid infrastructure.
Modern AI campuses now rival the most energy-intensive industrial assets — steel mills included — with the largest planned facilities consuming the equivalent output of a major offshore wind farm at peak. The engineering, logistical, and financial demands of building at this scale are redefining competitive dynamics across the entire datacenter value chain. Our study maps the trajectory of global capacity growth through 2035, tracing how exponential compute demand is translating into physical infrastructure requirements — and who is best positioned to meet them.
The long-term case is compelling — yet financing pressures, supply constraints, and public opposition could trigger a temporary slowdown. Enterprise AI adoption remains in early deployment phases, and new workload categories — from physical AI and autonomous systems to large-scale agentic applications — are just starting to scale. The structural argument for continued datacenter expansion holds.
Despite the robust long-term outlook, a near-term correction cannot be ruled out. Hyperscaler capital expenditure is broadly approaching operating cash flow — and for some players already exceeding it — making sustainable growth contingent on substantial AI-linked revenues materialising soon. Supply chain dynamics amplify the risk: even a modest slowdown in demand can trigger strong upstream contractions. Hard constraints — from grid access to memory supply — must be continuously overcome, and in some regions public opposition is already influencing permitting. How to position for resilience, not just growth, is a central question our study explores.
Europe's position in the global AI infrastructure race is under significant pressure. While the US maintains a commanding lead and China continues to scale aggressively, Europe's share of global datacenter capacity is projected to decline. The constraints are well-documented among industry participants: power grid access, high energy costs, regulatory complexity, and limited site availability consistently rank as the primary barriers in our survey of industry experts.
What makes this more than a technology challenge became starkly visible in June 2026, when a US export-control directive forced Anthropic to suspend foreign access to its two most advanced models overnight — a vivid reminder that reliance on foreign compute and foreign models is a dependency that can be revoked without warning. Domestic AI infrastructure is increasingly a prerequisite for digital sovereignty and for a competitive AI ecosystem — giving nations and firms sovereign control over how sensitive data is processed, securing dependable access to compute capacity, and establishing the seed and scale for a strong, independent AI industry. Without sufficient compute at home, Europe risks losing twice over: the productivity gains that AI brings to the wider economy, and a stake in the AI economy driving this revolution.
Closing the gap requires more than financial incentives. Our study examines which structural reforms — spanning grid regulation, permitting frameworks, and capital mobilization — could realistically shift Europe's trajectory, and identifies where the window for effective action is beginning to narrow.
This publication by Roland Berger offers decision-makers a rigorous, data-driven view of global datacenter dynamics through 2035 — combining market modeling, financial scenario analysis, and an extensive industry survey. For investors, operators, policymakers, and corporate leaders building AI strategy, the full study provides the analytical foundation needed to navigate a market where the decisions made today will define competitive positions for years to come.
This study is the first of two publications detailing our perspective on the AI datacenter ecosystem. Register to receive part one now, and be the first to receive part two when its released.
Sign up now to access the full study. You will also receive regular news and updates, delivered straight to your inbox.
_person_144.png?v=1687234)
_person_144.png?v=1687234)
