FAQ

Is any of this deployed?

No. There are no contracts on any network, no data centres, no sites, no partners, no audit and no launch date. These docs describe a design. Treat every mechanism described as intended behaviour, not as observed behaviour.

What am I actually buying if I stake $WATT?

A pro-rata claim on however much curtailed energy the network absorbs while your stake is in place, denominated in kWh. Not a yield, not a share of revenue, and not a fixed rate. If the network absorbs nothing during an epoch, the epoch pays nothing. That is not a bug, points are a claim on a measured physical quantity, and you cannot distribute a quantity that does not exist.

What stops the operator from inventing the energy figures?

Nothing in the mathematics. The budget for each epoch is submitted by an attestation set, and if that set colludes or is compromised, the numbers are wrong. Watt does not have a cryptographic proof that a solar site absorbed a given number of kWh, and does not claim one.

What the design does is make the assumption explicit and reduce it: independent attesters, a quorum threshold, bonds, and meter data that a counterparty on the generation side can also see. That is an economic assumption, not a proof. If you cannot accept a bonded quorum as an oracle for physical measurement, you should not use this protocol. See Legal.

Can I game the points with a flash loan?

No. Weight is stake-seconds over the whole epoch, not a balance at a snapshot. A position held for one block earns one block's worth of an epoch, a rounding error. To earn a meaningful share you have to hold a real position for a meaningful stretch of time, and carry the price risk that comes with it. There is no snapshot to time.

Why can't I see the epoch budget before I stake?

Because if you could, you would stake only into the good epochs and dilute everyone who stayed through the bad ones. The budget is published only after the epoch closes, so at the moment you decide to stake, the number does not exist for anyone. The full argument is in Energy Points.

Why do vouchers expire, and why is there no refund?

A voucher is a reservation of capacity in a window of time. If the window passes and the machine sat idle, that idle time is not recoverable inventory, it is the same reason a departed flight's empty seat has no salvage value.

Refunds would also break capacity planning. If every unused voucher rolled forward, the network would carry an unbounded liability that never settles and could never size hardware against it. Sell what you will not use, before the bucket closes.

Isn't intermittent power useless for AI?

For low-latency serving, yes. For batch work, training with checkpointing, batch inference, rendering, scientific compute, interruption is a scheduling problem, not a blocker. Watt targets the interruptible half of the workload space and does not pretend to serve the other half. See Infrastructure.

Why not just install batteries?

Batteries and behind-the-meter compute are not competitors so much as different trades. A battery time-shifts energy at a cost per kWh of storage plus round-trip losses, and then still has to sell into the same congested node that caused the curtailment. Compute converts the energy into a product that leaves over fibre rather than over wires, which is precisely the constraint that was binding. Storage is often the better answer; where transmission is the bottleneck rather than timing, it is not.

What happens if a job never completes?

Every job has an unconditional exit. If no attestation quorum resolves the job by the deadline plus a grace period, the holder can call reclaim and get the escrowed vouchers back. It needs no operator, no attester and no admin. It is gated only on a timestamp. See Redemption.

The caveat: reclaimed vouchers come back with their original expiry. Redeem with headroom.

Why an order book instead of an AMM?

Vouchers decay deterministically toward expiry. An AMM liquidity provider holding voucher inventory is short theta with no premium, they take certain time decay and are paid only volume-dependent swap fees. Plus every (SKU, expiry) pair is a separate asset, so pooled liquidity would fragment across a hundred markets. A book prices decay continuously and costs nothing to leave empty. See Marketplace.

Does the protocol earn from $WATT trading?

No. The launch venue owns the liquidity pool, so swap fees go there, not to Watt. Protocol revenue comes only from the voucher mint fee and the marketplace fee, both charged on actual use. If you are modelling this, exclude trading volume entirely. See Token.

What is the token supply?

Undetermined. Supply, allocation and vesting are TBD and will be published in full before launch. No number is given here, because publishing a placeholder as though it were a decision would be worse than publishing nothing.

Where does the 0.84 kWh figure come from?

An NVIDIA H100 SXM5 draws roughly 700 W at its rated TDP. At a power usage effectiveness of 1.2, accounting for cooling and distribution overhead, one GPU-hour sits behind about 0.84 kWh delivered. That is arithmetic, not a measurement of any Watt facility, and real PUE varies by site and season.

Is $WATT an investment?

No. It is a utility token for accessing compute capacity. Nothing in these docs is an offer of securities, a prospectus, or investment advice, and nothing here promises profit or return of capital. Access may be restricted by law where you live. Read Legal.

What is the biggest reason this might not work?

Physical execution. The token mechanism is the easy half. Contracting sites, shipping modules, energising them, keeping utilisation high enough that vouchers are worth minting, and doing it under whatever rules the local market applies to behind-the-meter load, that is where projects of this kind fail. Nothing in these docs de-risks it.