AI Data Centers Set to Consume 4× More Electricity by 2035
New research finds data center power demand through 2033 could rival India's entire national consumption, laying bare the energy bill of the AI boom.
W
We Read AI
We Read AI
Data centers built to serve AI workloads through 2033 could consume electricity equivalent to India's current national demand, according to a new industry projection covered by TechCrunch. The forecast puts hard numbers behind a growing concern: the AI boom's carbon and grid cost is arriving faster than utilities can build capacity.
The math
The base case has data-center electricity consumption roughly quadrupling by 2035, driven almost entirely by inference workloads for large models. Training draws attention, but production inference is where the sustained load lives.
What's giving way
Grid operators in Virginia, Ireland, and Arizona have already paused new hyperscaler connections. Nuclear PPAs — including Amazon's Talen Energy and Microsoft's Three Mile Island restart — are moving from novelty to standard practice.
The efficiency argument
Optimists point to model compression, smaller specialized models, and better hardware utilization as offsets. The report is skeptical: efficiency gains have been more than absorbed by demand growth for three years running.
Source: TechCrunch
#infrastructure#energy#data-centers