Clore.ai's B300 contract is built for the long run: a B300 node earns up to ~$3,400 a month per card on terms from 720 days, the longest on the platform, and the rate stays the same all the way out to 1,800. That ceiling is what one card makes on the contract route when it runs every hour of the month at the best B300 rate in the catalogue, supplied in 8-GPU nodes across the USA, the EU and Japan through a direct agreement with Clore. Blackwell Ultra brings 288 GB of HBM3e on every card, the largest memory Clore contracts.
The B300 is sold on the longest terms on the platform, at a rate that holds from 720 days to 1,800, for the biggest memory Clore offers.
Nobody contracts a B300 for a weekend. With a 720-day minimum and terms out to 1,800, the customers behind this card are inference services with committed users, funded research programmes and companies moving a cloud bill they already track. That is the kind of steady demand a contracted B300 node serves.
Clore's configurator prices the B300 the same at 720, 1,080, 1,440 and 1,800 days, so a longer contract keeps the card at the top of its range for the whole term. On most other contracted cards a longer term trades rate for length; on this one it simply adds time.
288 GB of HBM3e per card at 8,000 GB/s. Llama 3.1 405B at INT4 fits on one card and DeepSeek-V3's 671B at FP4 on two, which is the capacity argument that sends renters to Blackwell Ultra.
Any B300 server you list on the marketplace can take part in MFP Lock. CLORE locked against its Maximum Fair Price starts earning a day later, a share of the network's emission for each minute the server is rented through the marketplace, and every tier you fill leaves more of each Bitcoin, USDT or USDC rental with you. It is entirely optional and adds to per-minute rental income rather than replacing any of it.
Tiers 1 and 2 share in the CLORE emission while the server is rented on the marketplace, and the CLORE you lock stays yours to unlock later.
Locked against the server's Maximum Fair Price, with Tier 1 at 2,000, Tier 2 at 4,000 and Tier 3 at 3,000 CLORE for each point. A full lock across all three tiers gives the most.
The B300 earns on two tracks: the contract route that sets its ceiling, and per-minute listings you price yourself.
Nodes of eight GPUs supplied for 720 days or longer, agreed directly with the Clore team before any hardware is placed, with income paid in USD stablecoin or BTC.
List a B300 server like any other card, set your rates, and each rented minute pays out in BTC, USDT, USDC or CLORE.
Marketplace withdrawals in all four currencies are covered under payouts and fees →
The ceiling per card, the length of the commitment, and the regions where Clore sells it.
Earned in 8-GPU nodes on the contract route, the top of the B300 range on Clore.ai.
Four lengths in the catalogue, 720, 1,080, 1,440 and 1,800 days, all priced alike.
Clore sells B300 contracts in all three, each on the same set of terms.
B300 nodes ready to commit? Talk to us about supplying bare-metal capacity →
Contract supply starts with a conversation, and the software side is a single installer command on a machine that already boots.
There is no switch for this in the console: share the node count, region and term you have in mind through the support page, and the team takes it from there.
Before anything else, nvidia-smi has to answer: the installation docs make a working NVIDIA driver the first requirement. x86_64 on Ubuntu 22.04 or newer, or a HiveOS rig.
One command from Clore's public installer repository with your init token appended. After a reboot the server appears in your account, online.
Set spot and on-demand rates yourself on the marketplace side. Earnings credit per rented minute and withdraw in BTC, USDT, USDC or CLORE.
It means a node is placed for two years or more on a single agreement. 720 days is the shortest B300 term, and the configurator offers 1,080, 1,440 and 1,800 above it, all at the same rate, so a longer term stretches the income period without lowering the monthly figure. For an operator that is long, planned demand for the capacity, with the terms of each agreement set out with the Clore team before hardware is committed.
The largest models there are, at four bits. Llama 3.1 405B at INT4 is roughly 200 GB of weights and runs on a single card; DeepSeek-V3, a 671B mixture of experts, needs about 336 GB at FP4, or two cards; mixture-of-experts models at the trillion-parameter scale are what this much memory per card is for. Long programmes, inference services with committed users and multi-year research, are what fill a 720-day term.
In dense FP4 and in memory. NVIDIA's HGX figures give the eight-GPU B300 system 108 PFLOPS of dense FP4 against 72 for HGX B200, and 2.1 TB of total GPU memory against 1.4 TB, while FP8, FP16 and BF16 match. Networking doubles as well, 1.6 TB/s per system against 0.8 TB/s, which matters to tenants who scale past one node. That makes the B300 the card for four-bit inference on the largest models, which is exactly the work long contracts are signed for.
Start with a message through the support page: your node count, region and the term you can commit to. Income on this route is paid in USD stablecoin or BTC, your choice. Contract supply is agreed with the Clore team before any hardware is placed, and the catalogue lists B300 nodes with 1.5 TB of system memory, NVMe storage across eight 3.84 TB drives and a 3.2 Tbps Ethernet or RoCE fabric, so have your own node's details to hand. The same machines can also run the hosting agent and be listed per minute at your own prices.
On the contract side, 720 days is the shortest B300 term. The per-minute marketplace is always open: install the agent, list a B300 server at your own price and it earns for every rented minute. If you run Blackwell or Hopper nodes you would rather commit for shorter periods, the B200, H200 and H100 contracts start at 30 days.
Because on this part the term does not buy a discount. Clore's configurator returns the same figure at 720, 1,080, 1,440 and 1,800 days, so the up to ~$3,400 a month per card holds whichever term you supply on. A longer B300 term extends the same top rate instead of trading it down, which makes the longest terms as attractive per month as the shortest.
From Clore's bare-metal catalogue and NVIDIA's own specifications.
Every one of them sits inside a single agreement, the longest commitment any card on the platform carries.
All documentation →The most Clore contracts on any card, enough to run a 405B model at INT4 on one GPU.
Bare metal →Unique among the parts Clore contracts: the top rate holds for the full length of even the longest term.
Host onboarding →Minimum and longest terms from Clore's bare-metal catalogue. The B300 carries the most memory per card and the longest commitments.
Every contracted card starts at 8 GPUs. The renter view of the B300 contract, with NVIDIA's HGX numbers, is on the B300 rental page.
Installation, MFP Lock and payouts are documented here; the contract itself is agreed with the Clore team.
The rental page works through the same term from the customer's side, with the 288 GB argument and NVIDIA's own HGX numbers.
Open a host account, tell us about your nodes, and supply B300 capacity on terms from 720 to 1,800 days, each card earning up to ~$3,400 a month.