Own Your Mind - SN51 subsidy analysis
until 7 October 2026, when the endpoint began requiring a validator signature
A real, busy, open-source GPU marketplace with strong datacenter-GPU prices and the best agent/scripting tooling on our list. The trade-offs are structural: offshore terms with no refunds and a $100 liability floor, no SLA, a pseudonymous team, hosts that can read pods on standard nodes, and a supply side whose prices rest on Bittensor emissions still exceeding rental revenue.
site reachable · checked 9 h ago · daily check
by the bestcloudmining editorial desk · curated review · prose updated 2026-11-04 · facts as of 2026-11-04 · how we verify
Lium is a GPU rental marketplace where independent providers list machines at their own price and renters launch containerised pods billed per second [2][7]. It runs as Bittensor subnet 51 [8] and was branded Celium before the rename [15]. The operator is Datura AI Corp, a company incorporated in Saint Kitts and Nevis [1]; the public lead is pseudonymous [10]. On the check date the public feed listed 1,746 GPUs, of which 1,396 were rented [3].
Renters start pods on provider machines and pay per second from a prepaid balance [2]. On the check date 22 of 97 listed GPU models were rentable, from consumer RTX cards up to H200, B200 and B300 [2]. Access runs through the web app, a CLI, an SDK and an API [8]. On standard nodes the provider operates the host and can read a running pod's files, environment and memory [5]. Intel TDX confidential-VM nodes are the exception: the provider cannot read their memory or files [5]. Spot nodes are cheaper, but an interrupted spot rental earns no credit [6].
The terms are governed by the laws of Saint Kitts and Nevis, with disputes going to its courts [1]. The service is provided with no warranty of uptime [1]. Top-ups are non-refundable, and unused balance is not refunded when an account is closed [1]. Liability is capped at what you paid in the past twelve months or 100 US dollars, whichever is more [1]. To rent, the balance must cover 15 minutes of the pod plus 15 minutes of each running pod; when the balance reaches zero, pods keep running for 2 minutes and are then deleted [2][4]. Payment options are card, US bank debit, crypto, TAO or alpha [4], and crypto, TAO and alpha top-ups earn a 2.5% bonus [2]. No KYC procedure is published [4].
Lium exposes live prices, fleet counts and launch success rates without an account [2][3][13]. A single validator operated by the Lium team checks every connected node every 15 minutes [12]. The public validator source shows that all emission slots the docs label burn route to UID 47, the Datura payout key [9]. An independent analysis notes that the billing endpoint began requiring a validator signature on 7 October 2026, so outside revenue checks are harder now [10]. Proof is therefore partial: hardware checks and optional attestation exist, but machine scores and payouts rest on the operator's own validator [9][12].
Single H100 80GB HBM3 launches succeeded 93.41% of the time over 14 days on the operator's own statistics API [13]. A hands-on test sponsored by Lium found that each charge came within $0.0014 of the listed price multiplied by the seconds, and that failed starts cost nothing [11]. The same test found that the hardware behind one GPU name varies widely by host in PCIe generation, disk and network speed [11]. Lium offers no SLA [1]. The supply side is subsidised by subnet emissions that still exceed rental revenue, by a ratio the independent analysis puts at 2.3x [10]. We found no reports of hacks or lost balances, but Reddit and Trustpilot could not be reached on the check date.
Providers set their own prices, so asks move by the minute [2]. Lowest live asks per GPU-hour on the check date: RTX 3090 $0.16, RTX 4090 $0.40, RTX 5090 $0.50, A100 SXM4 80GB $1.68, H100 PCIe $1.79, H100 80GB HBM3 $2.08, H200 $3.60, B200 $5.88 and B300 $6.80 [2]. On consumer cards Vast.ai was cheaper in the sponsored test: an RTX 5090 from $0.40 against Lium's $0.71 [14]. Persistent volumes cost $0.00005 per GB-hour, about $0.036 per GB-month, plus egress [16].
Lium is a legitimate, busy GPU marketplace with some of the lowest datacenter-GPU asks on our list and strong tooling for scripted and agent-driven rentals [2][8]. The trade-offs are real: offshore terms with no refunds, no SLA, a pseudonymous team, and standard nodes where the host can read your data [1][5][10]. Use it for short, non-sensitive jobs, top up only what you will spend, and pick TDX nodes for anything confidential [5]. Teams that need refunds or contractual uptime are better served by a named-jurisdiction cloud [1].
| Documented price | $0.160 per GPU/hr · $1.79 per GPU/hr (H100 PCIe min) |
|---|---|
| Payouts | daily (providers) |
| Notes | Host-set marketplace, observed 2026-11-04 from lium.io/pricing and pricing.json, lowest live asks per GPU-hour: RTX 3090 $0.16, RTX 4090 $0.40, RTX 5090 $0.50, A100 SXM4 80GB $1.68, H100 PCIe $1.79, H100 80GB HBM3 $2.08, H200 $3.60, B200 $5.88, B300 $6.80. Fleet 2026-11-04: 1,746 GPUs listed, 1,396 rented, 344 available, 22 of 97 GPU models rentable. Balance rule: 15 minutes of the pod plus 15 minutes of each running pod; at zero balance pods run 2 minutes, then are deleted. Crypto/TAO/alpha top-ups earn a 2.5% bonus. Providers keep 95%. Launch success 93.41% for H100 80GB HBM3 over 14 days (operator API). |
Run the long math before paying: cloud mining calculator →
Ready to try Lium? Start small, verify one payout, and only scale up once the hashrate or compute actually matches what you paid for.
Based on our checks as of 2026-11-04: Lium scores 6.6/10 in our weighted review. We logged 2 red flags (listed above). User reports and operator background are documented with dated sources in the meta-review below.
Singapore-run GPU aggregator reselling capacity from 5+ neoclouds; H100 from $2.64/hr on-demand, per-minute billing.
Host-set GPU marketplace with per-second billing; on-demand, interruptible and reserved tiers, plus serverless and clusters.
Developer GPU cloud with published per-second rates across Community and Secure Cloud, plus serverless endpoints and clusters.
No comments yet - be the first. Posting runs a short proof-of-work in your browser (anti-spam), no account needed.