# Into: Green Compute A Bittensor subnet that rents out GPUs for AI inference with one hard condition: miners have to prove their electricity is renewable, every epoch, or they do not get paid. // Green-energy GPUs, scored every epoch --- > New to Bittensor? Start here. Experienced users can skip to the full analysis. ### What is Green Compute? Green Compute is a decentralized GPU compute network running as subnet 110 on Bittensor. According to its website, it rents out GPU capacity and serves AI inference, and miners only qualify if they can prove the power running their hardware comes from renewable sources like solar, hydro, wind, geothermal, or biogas. **The simple version:** It is like renting cloud GPUs the way you would from RunPod or Lambda, except every machine on the network has to run on verified clean power. **Centralized equivalent:** AWS, CoreWeave, or Lambda for GPU inference, but decentralized and restricted to green-energy hardware. **How it works:** - **Miners** run GPUs (the site names RTX 4090 and 5090 cards), serve real inference requests, and have to prove their power is clean through carbon registries and hardware-location attestation that is re-checked each epoch. They earn TAO plus rental fees, paid per epoch, on a "no proof, no payout" basis. - **Validators** score the miners each epoch. The site states that validators score miner work but does not publicly spell out exactly what they verify. --- ### Why This Matters - **The problem it solves:** AI inference is expensive to run and increasingly energy-hungry, and data-center power draw is a growing concern. Green Compute's pitch is to tie payouts to electricity that is provably renewable rather than treating "green" as a marketing badge. - **The opportunity:** It gives renewable-energy operators a way to monetize spare GPU capacity, and gives buyers an OpenAI-compatible API served on green hardware (both per the project's website). - **The Bittensor advantage:** The incentive layer pays miners in TAO for serving inference, and the green-energy proof is enforced as a condition of payment each epoch rather than self-reported once at signup. - **Traction signals:** Be honest here. The subnet carries an on-chain emission share around 2.6%, and its alpha price has roughly doubled over the past 30 days. But the project is early by its own roadmap, which dates a testnet to February 2026 and mainnet on subnet 110 to April 2026. No public code repository is registered on-chain, and there is no independent social or usage signal about the subnet yet. --- ## Full Analysis **Category:** Inference and Compute | **Centralized Competitor:** AWS, CoreWeave, Lambda, RunPod Decentralized GPU markets compete with centralized cloud providers on price and availability, and several Bittensor subnets already target compute and inference. Green Compute's stated differentiator is narrower: it does not just want cheap GPUs, it wants GPUs whose power is provably renewable, and it builds that requirement into the reward mechanism. **Mechanism:** Per the project's website, miners register GPU hardware (two SKUs, RTX 4090 and 5090) to subnet 110 and serve inference behind an OpenAI-compatible API. Before they can earn, they have to verify their energy source through carbon registries and hardware-location proofs, and that attestation is re-run each epoch via remote attestation. Validators score miners on their work, and payouts in TAO plus rental fees are settled each epoch, with the site framing it as "no proof, no payout." The website advertises indicative pricing from $0.40 per GPU-hour for a 4090 and $0.70 for a 5090. A few honest caveats. The implementation is not independently verifiable from here: there is no public code repository registered to the slot on-chain, the site lists no named team, and the product, pricing, and performance claims all come from the project's own marketing site. The validator side of the mechanism, specifically what validators check beyond "scoring," is not publicly documented. Treat the product details above as the team's description of what they are building, not as independently confirmed fact. On-chain, the slot is over a year old and active, with an emission share near 2.6% and pool depth of roughly 5,558 TAO backing the alpha token. The strong recent price move sits against thin public evidence of usage, so the open question is whether real inference demand shows up on the network. --- ### Risk Factors - **Execution:** Green Compute is early by its own published roadmap (testnet February 2026, mainnet on subnet 110 April 2026). Building a green-verified GPU network, the attestation pipeline, and real inference demand at once is unproven. - **Single-source documentation:** Product, pricing, and performance claims rest on the project's own website. No public code repository is registered on-chain and no team is named, so the implementation cannot be independently verified here. - **Traction unproven:** There is no independent social or usage signal about the subnet yet, and on-chain telemetry currently shows no active miners. Whether demand materializes is the central unknown. - **Liquidity:** Pool depth around 5,558 TAO is modest, so larger stakes will face meaningful slippage on entry and exit. - **Competition:** Multiple Bittensor subnets target compute and inference, and centralized clouds compete on price and scale. The green-energy requirement is the differentiator that has to pay off. --- Into the next one.