Ignore rosy datacentre expansion projections – there isn’t enough power
Summary
Uptime Institute’s latest report warns that electrical supply — both grid capacity and generation — is not being added quickly enough to match the explosive expansion of datacentre capacity driven by AI. The industry may face a meaningful slowdown in new builds because large, power-hungry facilities require sources of energy that take years to deliver, while short-term options such as on-site gas turbines and repurposed crypto-mining sites are constrained.
The report estimates AI-related datacentre load could hit about 10 GW by the end of 2026 and projects 75–125 GW additional server-farm power demand through 2030. Carbon capture and storage (CCS) is highlighted as a potential mitigation for emissions if gas-fired generation is used, but most CCS technologies are immature or face storage and cost barriers.
Key Points
- Uptime Institute predicts power will become the defining constraint on datacentre growth in 2026 and beyond.
- Global AI-driven datacentre load may reach roughly 10 GW by end of 2026; 75–125 GW extra may be required by 2030.
- New generation and grid upgrades (solar ~5 years, wind/gas ~6 years, nuclear ≥10 years) lag the much faster timescale of datacentre projects.
- Short-term supply tactics (on-site gas turbines, former crypto-mining sites) are limited by equipment shortages and dwindling sites.
- Using natural gas to meet demand risks undermining net-zero commitments; CCS can help but commercial options are limited and storage sites constrain deployment.
- Industry is fragmenting: a high-density class for extreme AI workloads is emerging, while much capacity remains for traditional, lower-power workloads.
Context and relevance
This matters to operators, cloud customers, infrastructure investors and regulators. The findings link directly to the AI infrastructure boom since 2022 and explain why many ambitious expansion forecasts may be unrealistic without big, coordinated investment in generation and transmission. It also connects energy policy, corporate net-zero pledges, procurement (PPAs) and siting decisions — including the practical limits of retrofitting CCS and where captured carbon can be stored.
For enterprises planning capacity or negotiating cloud contracts, the report signals potential delays, higher costs for guaranteed low-carbon power, and possible geographic constraints on where new datacentres can be located.
Why should I read this?
Quick and blunt: if you’re assuming datacentres can keep growing forever to feed AI, you’re probably wrong. This piece explains why power — not racks or chips — is the real bottleneck, and why that will change costs, timelines and where operators build next. Read it if you want to avoid being surprised by project delays, higher energy bills or strained sustainability claims.
Author style
Punchy take: Uptime Institute doesn’t mince words — power limits will force tougher choices on builders, cloud buyers and policymakers. If you work in infrastructure or energy, this is essential context; if you’re just curious about AI’s real-world costs, it’s a useful reality check.
Source
Source: https://www.theregister.com/2026/01/14/datacenter_expansion_power_limit/