Infrastructure

The token mechanism is only interesting if something physical sits under it. This page describes the intended physical layer. None of it is built. No site, operator, partner or capacity figure is announced, because none is settled.

Behind the meter

A generation site has a point of interconnection, the meter where its output enters the grid. Everything on the generator's side is behind the meter.

Watt's design places modular data centres there. The consequence is specific: a behind-the-meter load can consume output that the grid has instructed the site to curtail. The energy never crosses the interconnection point, so it is not constrained by the transmission capacity that caused the curtailment in the first place. It is also not a grid connection of its own, so it does not sit in an interconnection queue.

This is the whole reason the design works. A data centre in front of the meter, however close to the solar farm, is just another grid load buying grid power at grid prices, and it inherits every constraint that made the curtailment happen.

Modular

The units are containerised: compute, power distribution and cooling shipped as a sealed module and connected on a pad. This is a deliberate trade against conventional data-centre construction.

The last row matters more than it looks. Curtailment profiles are not permanent. A transmission upgrade can eliminate the surplus that justified a site. A fixed building would be stranded; a container is moved to the next site.

Curtailment profiles

Curtailment is not random. It follows structure, and the structure is what makes the load schedulable.

Daily. Solar curtailment concentrates around midday, when generation peaks and demand has not. The shape is the well-documented mid-day trough in net load.

Seasonal. Spring and autumn are the usual peaks in temperate markets: strong irradiance, mild temperatures, so little heating or cooling demand to absorb it.

Local. Curtailment is a property of a node, not a country. A site behind a congested transmission corridor can be curtailed heavily while a site fifty kilometres away is never curtailed at all. Site selection is largely the work of finding nodes with a persistent, structural surplus rather than an occasional one.

The output of this is a profile: for a given site, the hours in which surplus is expected to exist and its rough magnitude. Profiles drive two things, where modules go, and how far forward the protocol is willing to open expiry buckets for minting.

Watt publishes no profile for any site here. Real profiles come from real metering at real locations, and there are none.

Matching the load to the source

Not every workload suits intermittent power. The design targets those that do:

• Interruptible. Batch training with checkpointing, batch inference, rendering, scientific compute. Work that can pause when the sun goes behind a cloud and resume without losing progress.
• Latency-tolerant. A site chosen for its curtailment is not chosen for its proximity to users. Workloads that measure in hours, not milliseconds.
• Bandwidth-modest. Moving a training dataset once is fine. Streaming terabytes continuously to a rural node is not.

Low-latency serving is the obvious poor fit and the design does not pretend otherwise.

Why compute moves to the energy

The two problems in Introduction, energy that cannot reach a load, a load that cannot reach energy, have two possible fixes. Move the energy, or move the compute.

Moving the energy means transmission. A new line is a decade of permitting, right-of-way acquisition, litigation and construction, with losses over distance and a route that must satisfy every landowner along it. It is the correct long-term answer for the grid as a whole. It is not an answer available on the timescale on which curtailment is happening now.

Moving the compute means putting a container on a truck. The hardware is dense, shippable and already manufactured. Siting is a pad, a fence and a connection to equipment that already exists on the generator's side of the meter. Timelines are set by procurement and logistics, not by public inquiry.

The asymmetry is stark. Electrons are expensive to move and cheap to consume where they are made. Compute is expensive to build and cheap to relocate. Given a mismatch between the two, the cheaper thing should move.

There is a second effect. A behind-the-meter load gives the generator revenue on output it would otherwise be paid to withhold, turning a curtailment event from a loss into a sale. That improves project economics for the generator, which is what makes the arrangement worth signing on both sides. Whether it does so at any particular site depends on curtailment frequency, contract structure and local market rules, and Watt makes no claim about any specific site's economics.

Status

The infrastructure described here is a plan. No modules have been deployed, no sites contracted, and no capacity exists. Statements on this page describe how the system is designed to work, not what it has done.