Green Compute

SN110 Green Compute sells AI model answers and GPU hours the way a green electricity tariff sells power: the machines sit on UK farms running on manure gas, and only suppliers who prove their power is clean get paid.

The commodity, explained from zero

Two things are for sale. Inference is asking a trained model a question and getting an answer, metered by the token, roughly per word. GPU hours are rentals of the graphics processors that do the work, with an RTX 4090 "from $0.40/GPU/hr" and an RTX 5090 "from $0.70/GPU/hr", marked indicative. The API copies the OpenAI format, so a developer swaps a web address and a key.

The green part is the pitch. The about page says the hardware runs on "working UK farms that already produce electricity from on-farm renewables", chiefly biogas from methane that "would otherwise be vented into the atmosphere". The front page puts it less delicately: "The best clean inference on the market runs on cow shit." The customers named are enterprise teams, developers and researchers, and the enterprise page offers "dedicated capacity, token-metered billing" under contract. The front page claims "$82,000 in profitable revenue in the first month" (June 2026) and an $8,200 buyback of the subnet's token from 10 percent of contract revenue, with a wallet address to check. No customer is named except in testimonials.

Why it is on Bittensor at all

The about page gives three reasons: miners are paid for work actually done, anyone with the right hardware can add supply, and demand comes through the founders' existing business, Easy Compute. It adds: "If a solar miner in another country outperforms us, they win the traffic. That keeps us honest." The comparison breaks here: a green tariff buys from many generators, but on the day of checking the chain showed one active miner and thirteen validators, so the tariff has, for now, one supplier.

How the work gets done

Miners (the suppliers) connect RTX 4090 or 5090 nodes after submitting a carbon-credit registry attestation and proof of where the hardware sits; "no proof, no payout". Validators (the judges) route demand and score miners on "performance, reliability, and the integrity of green-power attestation", paying each epoch in emission, the subnet's share of newly minted TAO. A calculator on the front page applies a 1.5x multiplier for verified green nodes. How the score is computed is not published anywhere; there is no repository.

How you would know it works

The endpoint api.green-compute.com/v1/chat/completions answers requests, refusing without a key. The June 2026 buyback names a wallet, 5D2UXtzXhP2YCgreodgrGMs2niVZWv2Xt7cxJwYWioU6hdTH, that a reader can look up. The capacity page and the TEE counters render blank to a fetcher and cannot be checked.

What is missing

There is no code: no repository on chain or on the site, no incentive document, no scoring formula, and no miner guide beyond an application form. The chain identity fills 3 of 7 fields; sales@green-compute.com is on the site but not on chain. There is no API reference beyond one curl example, no status page and no SLA, though the sales form promises a reply within 24 hours. The registration cost is described as a burn without a number (0.0005 TAO on chain today).

Go deeper

• Front door with prices
• Plain account of the subnet
• API example and enterprise terms

Metaphor: a green electricity tariff for AI computing. Every claim is drawn from the evidence set or the subnet's own materials; "(inferred)" marks a conclusion rather than a quote. Corrections.