AI Data Center Power Infrastructure & Grid Engineering
The short version: An engineering breakdown of AI data centers' grid demands confirms what we've been saying: these are power plants' worth of load, not office buildings, and communities deserve a say.
Our Take
Credit to Keentel Engineering for pulling back the curtain on what actually happens when a hyperscale AI campus asks to plug into the grid. Their five-layer model is a useful reminder that "the cloud" starts with steel, copper, and megawatts — transmission-voltage interconnections, EMT modeling, voltage ride-through studies, and battery storage systems built specifically to handle the violent load swings of GPU training runs. This isn't a server farm humming along like a big-box store. It's industrial-scale infrastructure that utilities are having to re-engineer their planning models around.
Here's the part the engineering talk skips over: every one of those interconnection studies, substation upgrades, and reactive-power fixes gets paid for somehow, and it's rarely the data center absorbing the full cost. Grid upgrades get socialized into rate bases, utilities request new peaker capacity or gas plants to firm up these loads, and ratepayers in the surrounding county often find out only after the substation is already under construction. The technical sophistication described in this video is real — but so is the political question of who approved these interconnection queue jumps and what ratepayers were told before they signed off.
If a project like this is headed for your grid, you have a right to know before the transformers show up. Check our facility map to see what's already been sited near you, and use our take-action resources to find out who on your utility commission or county board approved the deal.
This is GridWatch the USA’s original commentary. The video above is the work of Keentel Engineering, published on YouTube — full credit to the creator.