Why the fastest chips in the world sit idle waiting for a grid connection — and how mobile gas turbines broke the logjam.
GPUs100,000
1,000600,000
150.0 MW required
Nvidia delivers chips in months · Grid connection takes a median of 5 years
Deployment Speed: Grid vs. Mobile Turbine
$1BPurchase Price
1+ GWFleet Capacity
~$1 /WCost Per Watt
600KH100s per GW
⚡ xAI's Memphis Cluster
When xAI built its first supercomputer cluster in Memphis, Tennessee, it deployed 100,000 GPUs drawing approximately 150 megawatts of power. The local grid couldn't supply enough electricity fast enough.
Elon Musk leased roughly 35 mobile gas turbines from APR Energy to power the cluster while grid connections remained pending. Environmental groups sued over air permits, and the U.S. Department of Justice intervened to keep the turbines running during the permitting process.
The experience revealed the real bottleneck in AI scaling: not chips, but power delivery speed. Months later, Musk personally acquired APR Energy for $1 billion.