Data centre firms say the AI bottleneck is now power and labour

Executives from NTT Data, Brookfield and Infosys say electricity and skilled labor—not chips—are now limiting the pace of AI data-center construction. In congested European markets, grid connections can take up to 10 years, highlighting a significant infrastructure bottleneck for AI expansion.

Analysis

The bottleneck shift changes where AI infrastructure capex can convert into revenue: scarce grid access and qualified labor can leave purchased compute idle, delay data-centre commissioning, and push utilization and customer monetization out even if chip deliveries remain available. That creates a second-order risk for semiconductor demand if deployment schedules slip, while improving the strategic value of power access, grid equipment, and projects already permitted and connected. But scarcity does not automatically mean supplier profits: utility approvals, labor availability, and project execution can limit how much demand equipment vendors capture.

Europe is the clearest near-term exposure; lengthy connection queues make regional capacity less fungible and could favor sites with secured power over otherwise attractive locations. Over 1–3 months, watch hyperscaler capex phasing, commissioning dates, and evidence of deferred deployments. Over 6–18 months, labor costs and grid investment timelines may shape returns on data-centre capital more than chip availability. The contrarian risk is treating this as a durable chip-demand collapse: delayed installations may defer orders rather than cancel them, and constraints could ease unevenly as projects secure power.

For INFY, the article supplies no company-specific evidence of incremental data-centre staffing revenue, contract wins, or financial exposure. Its mention is not a basis for changing estimates or taking a position. The thesis weakens if commissioning schedules hold, power-constrained projects secure connections faster than expected, or hyperscalers maintain chip orders despite delays.

Key Decisions for Investors

• No immediate INFY trade: verify whether Infosys has material, disclosed data-centre infrastructure or deployment work before translating the broad labor constraint into company revenue or margins.
• Set an alert on hyperscaler commissioning delays and capex phasing over the next 1–3 months. Repeated deferrals would raise downside risk to near-term AI hardware demand; stable deployment schedules would argue against that read-through.
• For an infrastructure relative-value watchlist, compare grid-equipment suppliers and power-secured data-centre operators with broad AI semiconductor exposure. Consider a long-versus-short expression only if valuations do not already reflect the commissioning bottleneck; avoid assuming every grid vendor captures the scarcity rent.
• Track European connection approvals, project-level power availability, and skilled-labor costs over 6–18 months. Faster-than-expected connections or easing labor constraints would falsify the durable infrastructure-scarcity thesis.