CPS

Run wherepower is cheapest.

CPS is a GPU scheduling API. It watches real day-ahead power prices across your sites and regions, then tells you where each flexible AI training or batch job should run to minimize power cost.

Get an API key

The decision, priced in dollars

Advisory only: CPS recommends, a human approves, your scheduler dispatches. Every decision is sealed as a tamper-evident receipt.

Example receipt: 64 GPUs, 12 hours (illustrative)

Site
Region
Projected cost
Site A
CAISO
$410
Site B
NYISO
$455
Recommendation
Site A
Approve?

Illustrative numbers to show the shape of a receipt. Real receipts list the job, the options, the prices seen and the projected cost, hash-chained to every receipt before them.

One call per job

Send GPU count, hours and a flexibility window. Get back the cheapest site and start window, priced in dollars.

POST /v1/jobs
{"gpus": 64, "hours": 12,
 "window_hours": 24}

-> site: "Site A"
   start: "02:00 local"
   projected_cost_usd: 410
   receipt: sealed

Who it is for

Neoclouds

GPU clouds with capacity in more than one market. Route flexible customer and internal workloads to the cheaper grid.

AI labs

Training-heavy teams whose power bill is the second line item after the GPUs themselves. Shift jobs in time and place.

Multi-site operators

Any operator with compute in two or more power markets who wants one place that answers: where should this job run?

How it works

  1. Register your sites. One API call per datacenter or region: name, region, price node, measured kW per GPU, PUE, GPU capacity. Your numbers, not our estimates.
  2. CPS polls real power prices. Day-ahead wholesale prices from public grid operators, refreshed daily. CAISO and NYISO live today.
  3. Submit a job. GPU count, estimated hours, flexibility window. CPS finds the cheapest site and the cheapest start window inside your window, and prices every option it considered, in dollars.
  4. Approve the decision. The recommendation is sealed as a receipt: the job, the options, the prices seen, the projected cost, hash-chained to every receipt before it.

Pricing

Pilot

$5,000

Flat. 30 days, one cluster. We help you wire it up and you get the receipts.

Production

$10

Per GPU managed, per month. Metered automatically from your usage.

CPS never dispatches workloads, never touches power or cooling hardware, and never states savings as a percentage. Costs are shown in dollars against real market prices.

Get an API key

Self-serve signup for the pilot. Your key is shown once, keep it somewhere safe.

Your API key (shown once):

Copy it now. We cannot show it again. Your principal id is , use it as the principal when you submit jobs.

Continue to the 5-step quickstart

Ready to activate billing?