An H200 supplied to Clore.ai earns up to ~$2,750 a month per card, and the contract route is where that figure lives: one card's month, every hour of it, at the highest H200 rate Clore's bare-metal catalogue carries, for 8-GPU nodes on terms from 30 days. The work that justifies 141 GB arrives with budgets and planning horizons, which is why it comes in on contract. The marketplace stays open as well, with the same one-command agent and per-minute prices you choose.
The jobs that need 141 GB are planned months ahead, and Clore's contract route is built to meet them.
Nobody picks up a 141 GB card for an afternoon. The workloads that justify it, 70B at full precision, 405B across two cards, long-context serving where the cache is the constraint, come with budgets and planning horizons attached. Clore meets that demand through the bare-metal route, on terms from 30 days out to 1,800.
The configurator treats H200 and H200 SXM as separate line items. The SXM entry sits in Iceland and Japan and sets the top of the H200 range; the standard entry covers the USA, Japan, Slovenia and India. Name your module in the listing so buyers see which one you run.
You can list an H200 server on the marketplace whenever you like. The agent is the same one command, the pricing is yours to set, and every rented minute is paid, so a node between contracts still has a way to earn.
MFP Lock is an optional layer over what a listed server earns. Locked CLORE, measured against the server's Maximum Fair Price, begins collecting MFP rewards a day after the lock, for every minute the machine is under a marketplace order. Each tier you fill also lifts how much of a Bitcoin, USDT or USDC rental you keep.
Tier 2 starts filling once Tier 1 is full, then Tier 3. Tiers 1 and 2 earn MFP rewards on marketplace orders, and every tier raises what you keep on BTC, USDT and USDC rentals.
Sized off the server's Maximum Fair Price, with tiers of 2,000, 4,000 and 3,000 CLORE per MFP point. Lock the full ceiling and every tier is full.
On the per-minute route, whether a renter pays in Bitcoin, USDT, USDC or CLORE, your earnings land in one Clore balance and withdraw in any of the four. Contract income is paid separately, in USD stablecoin or BTC.
Every minute your H200 server spends under an order is billed to the renter and added to your balance, whichever order type they picked.
Pick the coin when you withdraw. Keeping some earnings in CLORE also gives you the stake for MFP Lock, if you choose to use it.
Withdrawal details for every currency: payouts and fees →
Both doors are open. The contract route carries the top figure, and the marketplace keeps every setting in your hands.
This is where H200 demand lands: 8-GPU nodes on terms from 30 days in five regions, arranged with the Clore team rather than switched on in the console, and the route behind the up to ~$2,750 a month per card.
Install the agent, publish a spot price, an on-demand price or both, and change them whenever you like. Every rented minute is paid.
Lock CLORE against a listed H200 server and it earns MFP rewards on rented minutes, while you keep more of every BTC, USDT and USDC rental.
Have 8-GPU H200 nodes ready? Talk to us about supplying bare-metal capacity →
The agent installs onto a machine that already boots and already has a working driver, with no custom image or PXE step involved.
The documentation makes this the precondition: a working NVIDIA driver must already be installed. x86_64, Ubuntu 22.04 or newer with 22.04 LTS recommended, at least 8 GB of RAM and 32 GB of disk. AMD GPUs are not supported.
A single command from Clore's public installer repository, with your init token appended so the machine pairs to your account. Docker is pulled in as part of the install.
After the reboot the server appears against your account. HiveOS rigs take the identical command; the installer detects the OS and only the preparation steps differ.
Publish spot and on-demand prices if you want the per-minute route. If you have an 8-card node and a term in mind, the bare-metal side is where this card's demand is.
Because the hardware gets bought the way it gets used. A 141 GB card is rarely wanted for an afternoon; the jobs that justify it are production jobs with budgets and planning horizons. Clore.ai meets them through the bare-metal route, with 8-GPU nodes on 30-day minimums in the USA, Japan, Slovenia and Iceland and a 360-day minimum in India, and every card supplied that way earns up to ~$2,750 a month.
It takes a card out of the layout. Llama 3 70B at FP16 is roughly 140 GB of weights: two H100s and a tensor-parallel split, or one H200. Llama 3.1 405B at INT4 is around 200 GB: two H200s against three H100s. Renters comparing the two are comparing different job shapes, and the earnings follow: on Clore.ai an H200 earns up to ~$2,750 a month per card against up to ~$1,750 for an H100.
NVIDIA's MIG table lists both at 141 GB with up to seven instances, so capacity and partitioning match; the difference is mechanical. SXM modules sit on a vendor baseboard inside a purpose-built eight-GPU chassis, and that is the entry Clore's catalogue places at the top of the H200 range, in Iceland and Japan. NVL is a card you install in a server you already run, and it lists per minute like any other card. Name the form factor in your listing so renters know which one they are getting.
Work that runs out of memory long before it runs out of compute. A 70B model at FP16 loads on one card, Llama 3.1 405B at INT4 fits on two, and long-context serving turns the extra room into KV cache so more sessions stay resident at once. The compute is the same Hopper die as the H100, so renters pick this card for its 141 GB and its 4,800 GB/s rather than for FLOPS.
In the USA, Japan and Slovenia the catalogue runs H200 terms from a month to 1,800 days, stepping through 90, 180 and 360. The SXM entry in Iceland and Japan offers 30, 180 and 360 days and longer, and India starts at 360. The catalogue pairs its H200 cards with 1.5 to 3 TB of RAM and 3.2 Tbps InfiniBand or RoCE, which is the class of machine these buyers expect. A longer term keeps a node committed for longer and a shorter one keeps your hands free; the Clore team will talk through which suits your fleet.
Per rented minute on marketplace orders, into your Clore balance, in Bitcoin, USDT, USDC or CLORE. Those four are the platform's currencies: renters pay orders in them and hosts withdraw in them. On those marketplace orders, locking CLORE in MFP adds MFP rewards and raises what you keep on the BTC, USDT and USDC side. Contract income is paid separately, in USD stablecoin or BTC, your choice.
What a card can earn, where the contracts run, and what renters get for their money.
The ceiling for one H200 supplied to the bare-metal product, and the figure to set against your capacity plans.
Talk to us →H200 SXM in Iceland and Japan, standard H200 in the USA, Japan, Slovenia and India. Same 141 GB and the same MIG behaviour in every one of them.
Host onboarding →Room for a 70B at FP16 on a single card, the job that sends renters looking for an H200 in the first place.
All documentation →Terms each catalogue entry offers, by region. Every entry starts at 8 GPUs and runs out to 1,800 days.
The USA and Slovenia each carry two H200 entries, so demand in those regions comes in through more than one line item. The renter view of the same catalogue is on the H200 rental page.
MFP Lock, the installer and payouts, each on its own page in Clore's documentation.
The rental page works through what the extra memory unlocks, why the compute is identical to an H100's, and how the contract is configured.
Create a host account to list servers per minute at prices you set, or bring an 8-card node and a term to the contract side, where this card's demand actually sits.