Gartner warns datacenters hit 1,200 TWh by 2030, grid access becomes the bottleneck

AI is already changing how many datacenters are built. Now it is changing something more fundamental: whether the power grid can keep up. Gartner warns that global datacenter electricity consumption is set to jump, and by 2030 it expects demand to pass 1,200 TWh per year, even as it suggests grid supply may be insufficient to support new datacenter capacity.

Start with the nearer-term math. Gartner estimates global datacenter electricity consumption will reach 565 terawatt-hours (TWh) in 2026. It also forecasts power demand rising from 104 GW in 2025 to 132 GW this year. Put simply: the ramp is fast, and the question is not whether energy use keeps growing, but whether grids and planners can deliver enough electricity in time to match the buildout.

Why is this happening? Gartner points directly at AI workload design and procurement choices. It says the ballooning requirement for compute power to run AI workloads is driving the increase in datacenter power consumption. The research firm also calls out the incentive problems underneath the buildout. In Gartner's framing, fear of missing out (FOMO) pushes otherwise sensible companies to spend heavily on AI projects, despite often seeing little return.

The key link is that “AI-optimized servers” are what keep pushing consumption higher. Gartner notes that hyperscalers and other buyers have been funneling much of their server budgets into heavily configured systems meant to meet AI processing requirements. This year, Gartner expects AI-optimized servers to account for 31 percent of all datacenter power consumption. Then it sharpens the forecast for the next step: by next year, their combined power consumption will surpass that of all conventional servers in operation. Gartner also previously forecast that AI would overtake other server workloads, like databases and analytics, and become the top workload by server deployment by 2027, and the continued expansion it describes is meant to be consistent with that pattern.

This is where the “power wall” idea gets teeth. If total datacenter electricity consumption is estimated by Gartner to pass 1,200 TWh by 2030, grid supply may struggle to support additional datacenter capacity. The Register notes there have been earlier warnings that “bit barn” energy demands can outpace the grid's ability to deliver. Even the earlier numbers from Wall Street do not look comforting. Goldman Sachs estimated datacenter energy use would more than double by the end of the decade, but if Gartner's figures are correct, demand is already higher than where that Goldman report estimated it would be for 2027.

And it is not just a one-firm scenario. Energy infrastructure business Schneider Electric published four scenarios for future electricity consumption by AI datacenters at the start of last year. Gartner's latest estimate for total datacenter electricity demand in 2030 surpasses even Schneider's most aggressive forecast. That matters because scenarios are supposed to bracket plausible outcomes. When a fresh estimate overtakes the “most aggressive” prior scenario, it suggests planners might be underestimating how fast the demand curve can move.

So what should operators and investors do with a problem like this? Gartner’s director analyst Linglan Wang frames it as a new scaling constraint. She comments: “Surging demand for compute-intensive AI workloads is driving unprecedented datacenter power growth, while AI capacity is now constrained by power availability, making datacenter power security the new battle ground for scaling and protecting margins in the global AI race.” In other words, power availability is becoming a gating factor for growth and for profitability, not just an operational detail.

Wang's suggested mitigation is also practical, not mystical: infrastructure and operations (I&O) leaders must prioritize efficiency upgrades and secure grid access, and they need to invest in high-efficiency cooling systems and edge computing to mitigate power constraints. For executives, the second-order implication is that “building faster” may stop working if the limiting factor shifts from capital budgets to utility queues, interconnection timelines, transformer capacity, and cooling design constraints. In the US, the Register notes power grid operators and datacenter developers are already in a bind, which is exactly the kind of real-world friction that makes theoretical projections hurt on a spreadsheet.

The strategic stakes for peers are straightforward. If electricity access becomes the new battle ground, boards and CFOs will have to treat power planning like a core investment thesis. That means underwriting projects with grid realities, not just capex and timelines for servers. It also means competitive advantage could shift toward teams that can secure access, improve efficiency, and architect deployments around power availability, rather than teams that only optimize for compute density. In 2026 to 2030, the winners may not be the ones who just buy the fastest hardware. They may be the ones who can actually plug in.