Data center energy consumption — the silent engine of the AI revolution — will quadruple by 2035, according to new industry data reported by TechCrunch. The number is striking, but what really matters isn't the volume: it's who will foot the bill and how global electrical infrastructure will absorb this increase.
The projected growth isn't linear. Data centers already consume about 1-2% of global electricity, but the curve is accelerating due to three simultaneous factors: AI inference demand (which consumes far more energy than training because it's continuous), conventional cloud computing expansion, and the proliferation of autonomous AI agents maintaining 24/7 active workloads. Training a large model consumes energy for months — but inference for billions of users consumes energy forever.
The data point that goes unnoticed in most analyses is that this 4x increase isn't evenly distributed. It concentrates in specific regions: Northern Virginia (US), Ireland, Singapore, and parts of Scandinavia already face grid pressure from data center concentration. US utility companies are already seeking permission to build new generation plants exclusively for data centers, some turning to natural gas as a short-term solution while renewable infrastructure catches up.
The most interesting development, however, is the market response. Companies like Bluecore Energy are raising capital for portable nuclear reactors. Sila is scaling battery factories with $300M. Microsoft has signed nuclear energy contracts for its data centers. The emerging trend is that future data center location will be determined not by network latency but by energy availability.
The lingering question: who bears the environmental cost? Big tech is financing renewables and nuclear, but the indirect cost — grid pressure, emissions during transition, competition for cooling water — is socialized. The 4x electricity bill isn't just big tech's. It belongs to everyone living near a data center or depending on a grid that also serves one.
Source: TechCrunch
