That headline is the bare-metal contract route: Tesla T4s supplied to Clore.ai as a block of 8 or more, on terms of 30 days and up, can earn up to ~$270/month per card. A single T4 can also be listed per minute at your own price, and the demand is real: 11 of the 13 T4 servers listed on 21 September 2026 were rented, running batch transcription, embeddings, detection and video transcode. Own T4s in quantity? The contract is where they earn the most. Own one? List it and put it in front of those queues.
Batch queues keep the per-minute T4 pool busy, and Clore.ai also sells T4 capacity on bare-metal terms. Here is where each of your cards earns the most.
Demand for Turing has held up. At the 21 September 2026 snapshot, 11 of the 13 T4 servers listed on Clore.ai were under an order, booked for transcription backlogs, embedding runs, detection and video transcode. A new listing joins a card class renters already rely on.
Clore.ai sells T4 bare metal to renters who want capacity that stays put, in blocks of 8 GPUs or more on terms of 30 days and longer. Operators who can supply eight cards fill those contracts, and the arrangement is made directly with the Clore.ai team.
Single slot, low-profile and passively cooled, a T4 slots into servers with no room for a full-height card. That makes the eight cards a contract needs easy to fit into a few machines.
MFP is an internal quality score on each server. Locking CLORE against it is optional: it earns a share of block-reward emissions while also raising what you keep when renters pay in BTC, USDT or USDC. Listing works the same with or without it.
The lock scales with the server's score rather than being a fixed amount. The maximum total is the score times 9,000 CLORE.
Tier 3 earns no emissions, yet it counts toward the full lock, and a full lock means more of every rental settled in BTC, USDT or USDC stays with you.
Rental income accrues minute by minute against the daily price you enter, and collects in your Clore.ai balance until you withdraw it.
Spot and on-demand orders both bill by the minute, so a ten-minute CI run and a week-long transcription queue each pay for exactly the time they use.
Bitcoin, CLORE, USDT or USDC, whichever suits you. MFP rewards, if you choose to lock CLORE, come in CLORE.
The docs cover withdrawals in full. Payouts and fees →
Behind the ~$270: one T4 held on a bare-metal contract all month, 24 hours a day, at the highest contract rate Clore.ai offers for the card, in blocks from 8 GPUs and 30 days. A per-minute listing earns at whatever daily price you set, for every minute a renter keeps the card.
The route for T4 fleets. Tell the Clore.ai team how many cards you can commit and for how long, and the contract is arranged with you directly.
The route for a single server. Set a daily price in USD and renters with batch queues can book the card by the minute.
Renters had most of the listed T4 pool under orders at the last snapshot, so a new listing arrives with an audience already looking.
Holding eight or more T4s? Talk to us about supplying bare-metal capacity →
If the card is already in a machine that is already on, listing it takes about twenty minutes and one command. Here is the whole process.
A single-slot board sits in almost any chassis you own, so the checklist is about software. Ubuntu 22.04 LTS on x86_64 is the recommended base with 24.04 supported; 18.04 and 20.04 are not. Install the NVIDIA driver before the agent. You need 8 GB of RAM (16 recommended) and 32 GB of disk. HiveOS rigs have their own flow.
Choose Add Server in the dashboard and it writes an installer line carrying your init token. Run it as root. Docker is installed as part of setup.
The machine reports in after a reboot and shows online. A box installed without a token can be paired afterwards with /opt/clore-hosting/clore.sh --init-token.
Enter an on-demand price per day in USD, add a lower spot price if you want interruptible work, and revise either whenever you like. Eight cards or more? Ask about a contract before pricing them one at a time.
Up to ~$270/month per card when T4s are supplied on a bare-metal contract, which needs a block of at least 8 GPUs committed for 30 days or more. A single T4 can also go on the per-minute marketplace at a price you set. Plan around the contract figure if you own T4s in quantity; the per-minute route suits a single card in a machine you already run.
Because the two routes sell different things. A per-minute listing sells hours that a renter can start and stop at will, while a bare-metal block sells a committed minimum of 8 GPUs for a minimum of 30 days, capacity that is guaranteed to stay the renter's. Renters pay more for that guarantee, which is why a contracted T4 earns more. The card is identical; the product is committed capacity.
Rental income accrues minute by minute and collects in your account balance. From there it can be withdrawn as Bitcoin, CLORE, USDT or USDC, and an MFP lock adds rewards paid in CLORE on those per-minute rentals. Income from a T4 bare-metal contract comes in USD stablecoin or BTC.
Eleven of the thirteen listed servers were rented at the last snapshot, so yes. The work that lands on these cards is the work that never needed anything newer: batch transcription, bulk detection and classification, embedding a corpus, hardware video transcode through NVENC, and continuous-integration runners that need a real GPU to execute CUDA tests. All of it runs in FP16 or INT8, which is exactly what Turing has.
Eight cards is exactly where bare-metal terms start, so you already qualify on count. Terms begin at 30 days, and the deal is put together with the Clore.ai team instead of through a switch in the console: message Support with your T4 count and the length of commitment you have in mind. Under contract, every one of those cards can earn up to ~$270/month.
It does. Every listing carries a reliability figure, the server's uptime as a percentage, and renters can filter and sort on it. The marketplace's Inference preset, for example, asks for at least 16 GB of VRAM and 95% reliability. A T4's 16 GB clears the memory bar, so a machine that stays online keeps the card in front of those renters.
Three workloads that never needed a newer architecture. Worth naming in your listing description, because they are what the people searching for this card are searching for.
Nobody is waiting on any single item, so renters judge the card on cost per finished unit, where a T4 does well.
Host docs →Encoding barely touches the tensor cores, so an eight-year-old card competes with a current one on the part of the chip that matters here.
Payouts and fees →A machine that exists for ten minutes and then goes away. Per-minute billing makes the pattern work, and every one of those minutes is paid.
MFP mechanics →The most each card can earn per month on Clore.ai, next to the architecture and bandwidth renters check first.
The install guide walks through the onboarding above, the MFP page explains the lock, and the payouts page covers how T4 earnings reach you.
The rent page makes the case from the renter's side: which queues a T4 runs well, which formats it supports, and how renters pick a listing.
List a single T4 per minute at your own price, or bring a block of eight or more under contract and earn up to ~$270/month per card.