H100 GPUs at$7.57/hr
Same NVIDIA silicon, none of the ceremony. Per-second billing, no contracts, live in five minutes. Train like a frontier lab. Pay like you read the invoice.
01 — The problem
with renting silicon
Somewhere along the way, renting a graphics card started costing more than the graphics card.
The big clouds call it enterprise pricing a premium for the logo on the invoice. The chip doesn't know the difference. Your burn rate does.
So we built the GPU cloud we wanted to buy: a meter that only runs while your job does, hardware that shows up in minutes, and a bill you can read out loud without flinching.
02 — The number that matters
$0.00
Same chip. Same CUDA. Different invoice.
The rate card
All on-demand · all per-second · all yours in minutes
Indicative on-demand rates, June 2026 — final price confirmed at launch. Per-second billing on everything. There is no fine print; this is the fine print.
03 — The stopwatch
Card swipe to CUDA in five minutes.
THINGS THAT DID NOT HAPPEN: a sales call · a capacity request form · a 14-day approval · a reserved-instance contract
$0.000000this is your meter, running on an H100, right now
04 — The meter
It only runs when you do.
Per-second billing means exactly that. Stop the box and the spend stops with it — mid-epoch, mid-thought, midnight.
Your idle weekend costs $0.000000. We think that's a feature, not a discount.
05 — Private capacity
Buy it by the rack.
Reserved clusters of 10 to 200 GPUs for teams whose training runs outlive quarters. Terms from 30 days to 6 months. Prepay and watch the rate drop. Same five-minute provisioning — multiplied.
