AI data centre power demand is growing faster than the grid can respond. JLL’s 2026 global data centre outlook, released this week, delivers a stark conclusion: power availability has displaced land cost and capital access as the primary constraint on data centre development. Multiyear grid connection waits are now a structural feature of major markets, not an anomaly. For developers, occupiers, and energy investors, the implications are immediate and far-reaching.
Why AI data centre power is now the site constraint
For most of the past decade, data centre site selection turned on three variables: land availability, fibre connectivity, and access to capital. Power was a consideration, but rarely the deciding factor. That calculus has shifted decisively.
JLL’s outlook identifies AI data centre power as the single most consequential variable in site selection today. The reason is structural. Grid interconnection queues in the United States, United Kingdom, and across continental Europe now stretch three to five years in many jurisdictions. A developer who secures land and financing today may still wait half a decade before drawing a single megawatt from the network.
Scale of demand
The numbers behind this shift are striking. A single large-language-model training cluster can consume 50 to 100 megawatts continuously. Hyperscale campuses are now routinely planned at 500 megawatts or above. That is equivalent to the baseload demand of a mid-sized city, concentrated on a single site and required with near-perfect reliability.
The timeline mismatch
Grid operators were not designed to absorb demand spikes of this magnitude at this speed. Transmission infrastructure investment cycles run ten to fifteen years. AI adoption cycles run eighteen to thirty-six months. The mismatch is not a policy failure — it is a physics problem compounded by planning timelines.
Markets already constrained
Northern Virginia, the world’s largest data centre market, has seen utilities impose effective moratoriums on new large-load connections in certain zones. Ireland’s grid operator has maintained connection restrictions for hyperscale facilities since 2022. Parts of the Netherlands and Singapore have imposed outright development pauses. The pattern is consistent: the markets most attractive for AI infrastructure are precisely the markets where grid headroom is most constrained.
How developers are responding
Faced with multiyear waits, the most sophisticated developers are not simply queuing. They are restructuring their energy strategies from the ground up.
On-site generation
On-site generation is the most immediate lever. Gas turbine installations, fuel cell arrays, and small modular reactor agreements are all being pursued as ways to bypass the interconnection queue entirely. Several hyperscalers have signed letters of intent with SMR developers, accepting a five-to-seven-year delivery timeline in exchange for guaranteed, grid-independent capacity.
Restructured power purchase agreements
Long-term power purchase agreements are also being restructured. Rather than sourcing renewable energy from existing projects near established markets, buyers are now co-developing generation assets in markets with genuine grid headroom — accepting higher transmission costs in exchange for connection certainty.
Geographic diversification
A third response is geographic diversification. Markets including Texas, parts of the Midwest, the Nordics, and select Southeast Asian jurisdictions are attracting investment precisely because their grid infrastructure can still absorb large new loads within commercially viable timelines.
What this means for occupiers
The power constraint is not only a developer problem. Corporate occupiers — enterprises and cloud-native businesses deploying AI at scale — face real risk of delayed infrastructure access if energy strategy is treated as a downstream procurement decision.
Historically, a corporate technology team would select a colocation provider based on latency, compliance, and cost, then leave power sourcing to the operator. That model is no longer adequate. Occupiers who do not ask hard questions about a facility’s grid connection status, its backup generation capacity, and its position in the local interconnection queue are accepting delivery risk they may not have priced.
JLL’s outlook recommends that corporate real estate and technology teams integrate energy due diligence into site selection from the earliest stage. The question is no longer whether a building can be built — it is whether it can be powered, and when.
The policy and investment implication
The long-term resolution to the AI data centre power gap requires policy action, not just private ingenuity. Transmission permitting reform, accelerated interconnection processing, and coordinated investment in high-voltage infrastructure are all necessary conditions for the market to function efficiently.
The United States Federal Energy Regulatory Commission has introduced reforms to streamline the interconnection queue, but implementation is uneven across regional transmission organisations. In the United Kingdom, the National Energy System Operator is piloting a connections reform programme that prioritises shovel-ready projects. Progress is real but incremental relative to the pace of demand growth.
For investors and developers, the practical implication is clear. Markets that move fastest on grid reform will capture a disproportionate share of AI infrastructure investment over the next decade. Power policy is now industrial policy, and the jurisdictions that understand this earliest will define the geography of the AI economy.
