Mario Draghi calls for Europe to build AI data centers

Mario Draghi warned that Europe has less than 5% of global AI compute capacity versus 75% for the U.S., with its AI power-capacity shortfall projected to widen from roughly 3GW currently to 14GW by 2030. He estimates rapid AI adoption could add 0.3-0.4 percentage points annually to euro-area productivity growth, while Europe’s data economy could exceed €800 billion, or more than 5% of GDP, by 2030. Draghi called for pooled corporate commitments to fund new data centers, citing ASML, Capgemini and Amadeus-backed Mistral compute purchases targeting 1GW of capacity by 2030; permitting and grid delays remain a major constraint, with German projects taking about 42 months versus 24 months in the U.S.

Analysis

The investable implication is not a near-term AI revenue step-up for ASML, Capgemini or Amadeus; it is a European power-and-permitting bottleneck trade. Incremental compute demand will be constrained by substations, high-voltage cabling, transformers, switchgear and grid-connection queues, concentrating pricing power with Schneider Electric (SU.PA), Siemens Energy (ENR.DE), ABB (ABBN.SW), Prysmian (PRY.MI) and Legrand (LR.PA). These suppliers can monetize announced capacity before the underlying facilities reach utilization, while data-center developers carry multi-year execution and power-price risk.

ASML's read-through is principally political: a sovereignty-led capex agenda could broaden European semiconductor and compute subsidies, but the cited corporate commitments do not yet alter its order-cycle economics. The more relevant 6-18 month catalyst is whether EU funding is directed toward domestic chips, power infrastructure and expedited grid approvals rather than primarily software or cloud consumption. Capgemini could gain consulting and integration work, but AI implementation revenue is labor-intensive and may initially dilute margins if utilization or pricing lags hiring.

Consensus may overvalue the headline capacity deficit as immediately addressable demand. Higher financing costs, volatile wholesale power prices and procurement fragmentation can delay projects even where strategic intent is strong; a 1GW commitment spread over several years is insufficient to move broad equipment-company earnings without follow-on contracts. The thesis is falsified if European grid-capex plans, transformer orders and connection approvals fail to accelerate over the next two reporting cycles, or if power constraints force hyperscalers to concentrate deployments outside continental Europe.

Key Decisions for Investors

• Build a 6-12 month long basket of SU.PA, PRY.MI and ABBN.SW versus a short position in DLR or EQIX only if European power-connection delays remain elevated; equipment vendors have earlier revenue recognition and less utilization risk. Target 10-15% relative return; exit if order intake and backlog commentary fail to improve by 1H results.
• Keep ASML as a watch-list beneficiary rather than add solely on this development. Upgrade only on evidence of funded EU semiconductor/compute programs or a measurable increase in European customer backlog; the risk is multiple expansion without an earnings revision.
• For CAP.PA, wait for disclosed AI-services bookings, pricing and utilization before taking directional exposure. A long is justified only if AI revenue growth exceeds the associated delivery-cost growth; otherwise the likely outcome is revenue growth with margin pressure.
• Monitor ENR.DE and LR.PA for transformer, grid-automation and electrical-distribution order commentary over the next 1-3 months. Confirmed backlog acceleration would support adding exposure; rising European power prices without corresponding order conversion would favor taking profits on the grid-equipment basket.