Headlines about AI electricity demand tend to settle on a single number — data center consumption will double by 2030, or triple, depending on which report is cited. The more rigorous underlying analysis, however, is built around explicit scenarios with meaningfully different outcomes, and treating any single point estimate as a planning certainty is a mistake that can leave developers and investors badly exposed.
The IEA's Own Framework Acknowledges Substantial Uncertainty
The International Energy Agency's modelling of data center electricity demand is explicitly scenario-based, built around sensitivity cases including a Base Case, a more aggressive Lift-Off scenario, a High Efficiency scenario reflecting faster hardware and software efficiency gains, and a Headwinds scenario reflecting greater bottlenecks across the energy and chip supply chain. The Base Case projects global data center electricity consumption roughly doubling from approximately 485 TWh in 2025 to around 945 TWh by 2030, with AI-focused consumption specifically tripling over the same period — but the IEA is explicit that this central projection sits within a meaningful range, not a precise forecast.
Critically, the IEA's own analysis notes that bottlenecks across energy supply chains and chip manufacturing are reducing the likelihood of more aggressive near-term scenarios materialising, even as booming investment and project pipelines might otherwise suggest faster growth — a useful illustration of how physical and supply chain constraints, not just demand intentions, shape the actual realised outcome.
Why Single-Number Planning Is Dangerous
- Grid and generation investment decisions made against an overly aggressive demand assumption can result in stranded or underutilised capacity if actual demand falls short
- Conversely, planning against an overly conservative demand assumption can leave developers and grid operators badly under-prepared for genuine growth, compounding the interconnection queue problems already straining the sector
- Efficiency improvements — which the IEA describes as occurring at a rate unprecedented in energy history on a per-task basis — can meaningfully offset aggregate demand growth even as usage expands, making the net demand trajectory genuinely uncertain rather than a simple extrapolation of current growth rates
The right way to plan for AI electricity demand is not to pick the most likely single number — it is to build infrastructure and investment strategies that perform reasonably well across a plausible range of outcomes.
What Scenario-Aware Planning Looks Like in Practice
For developers, this means designing facilities and securing power capacity with enough flexibility to scale up if demand runs ahead of expectations, without being so aggressively over-built that a slower demand scenario leaves significant capacity stranded. For investors, it means stress-testing underwriting assumptions against both higher and lower demand scenarios, rather than relying on a single base-case projection that may not reflect the genuine range of plausible outcomes that credible energy forecasters themselves acknowledge.
Grid operators and policymakers face an analogous challenge at a larger scale — investing in transmission and generation capacity that can accommodate a meaningful range of data center demand growth without either over-committing capital to a scenario that does not materialise or under-investing relative to genuine need.
Building Scenario Resilience Into Investment Strategy
DATAPERT incorporates scenario-based energy demand analysis into the investment intelligence work we conduct with clients, helping ensure infrastructure and capital strategies remain resilient across a genuine range of plausible futures. Explore our data center development advisory or start a project to discuss scenario planning for your portfolio.
