Deloitte warns US AI datacenter power demand will explode 30x by 2035
The power crunch at AI datacenters is turning into a full-blown infrastructure crisis in the US.
A new Deloitte Insights report flags US datacenters used about 33 GW in 2023, with AI facilities pulling 4 GW of that. By 2035, total datacenter energy needs could hit 176 GW, and AI alone might gulp 123 GW — more than 30 times today’s AI power draw and 70% of the total. 
Datacenters are not just multiplying, they’re swelling in size too. Today’s biggest hyperscalers draw under 500 MW each. New ones under construction are expected to run over 2 GW. Mega 50,000-acre campuses that could pull 5 GW are already being planned.
The grid and power generation can’t keep up. Connection wait times stretch to seven years, while building new gas plants lags datacenter launches by a decade. Most new power still comes from gas, despite promises of clean energy from cloud giants.
Supply chain snarls and tariffs on steel, copper, and cement are slowing energy and datacenter expansions alike.
Deloitte says the fix needs a triple punch: tech innovation, regulatory reforms, and a flood of new investments.
On tech innovation, they point to more efficient datacenter hardware like optical data links and solid-state transformers for grids.
Regulatory changes could trim red tape and prioritize projects by readiness, copying the UK’s approach to kill “zombie” grid projects.
Biggest solution? Cash.
"Developing additive infrastructure will require massive funding across all of the industries involved," Deloitte states.
Failure to boost the grid risks strangling US AI progress and losing the economic and geopolitical edge.
"Staking an infrastructural lead in powering AI may now be a matter of competitiveness and even national security," Deloitte concludes.
Report link: Can US infrastructure keep up with the AI economy?
