about

we price the electricity
behind compute.

Algorithm Energy Research reads what a utility publishes, applies it to the load a proposed site would carry, and reports the result in the units the operator runs on.

transmission structuregrid
High-voltage grid insulators against a white sky
why

why we exist

Every model that gets trained or served is, at bottom, a quantity of electricity turned into computation. That conversion is priced poorly today, and much of the avoidable cost comes from the rate schedule a site ends up on. Rates are set by documents written long before anyone planned a GPU hall, and the people building the capacity rarely see what the same load would cost somewhere else.

We think the binding constraint on intelligence over the next decade is power: how much of it can be delivered, and how well it gets priced.

what it means

what that means if we do it well

Lower electricity costs allow more training runs, more inference and more experiments to proceed.

If a dollar of electricity buys meaningfully more intelligence than it did before, that compounds everywhere computation matters: in medicine, in materials, in the sciences, and in work nobody has thought of yet. We will not cure anything ourselves. We are trying to make the input cheap enough that the people who might can afford to try.

the work

what we actually do

We read the utility tariff serving a proposed facility, price it against the planned load, and report the result in the units the operator runs on: dollars per kilowatt-hour, per GPU-hour, per million tokens. When a company is choosing where to build, we price the candidates on the same basis so the power consequences of the decision are visible before anything is signed.

The work is rate schedules, riders, demand charges and effective dates. A large share of the cost of intelligence currently hides in them.

measurement

how we will know it is working

What we check before commitment:

  1. Whether the proposed load has a documented price against the applicable filed schedule.
  2. Whether candidate territories have been compared using the same assumptions.
  3. Whether the commitment was signed with that price already in hand.

Site and power commitments shape the operating cost long before a facility opens. We make those assumptions visible while the decision can still change.