lium

LIUM

SN51

Lium (lium.io) is an agent-first GPU cloud on Bittensor Subnet 51. It enables users to rent NVIDIA GPUs through the web interface, command-line interface, SDK, or API. GPU providers earn 95% of rental revenue plus bonuses. The project provides documentation at docs.lium.io and a provider portal at provider.lium.io. Its CLI and SDK are available at GitHub · Datura-ai/lium. Official community and support channels are discord.gg · Lium, X · @lium_io, and support@lium.io.

EVERY CLAIM IS SOURCE-LINKED · “—” MEANS NOT VERIFIED, NOT ZERO · METHODOLOGY · REVISIONS

[OVERVIEW]

Mission

Lium operates a GPU-rental marketplace on Bittensor Subnet 51. Independent providers supply machines; renters deploy pods through the web, CLI, Python SDK or API. Provider income combines customer-paid rentals with subnet incentivesSourcesGitHub · Datura-ai/lium-iodocs.lium.io · Architecture

Team

The terms effective September 27, 2026 identify Datura AI Corp, incorporated in Saint Kitts and Nevis, as the service operator. The public repository is maintained under Datura-ai. Individual founder identities and funding amounts were not established from these sourcesSourceslium.io · TermsGitHub · Datura-ai/lium-io

Key milestones

  • 2026-09-27Published service termsSourcelium.io · TermsThe effective terms identify the operator, prepaid service model, refund limits and governing law. This is a legal-document milestone, not the project launch dateSourcelium.io · Terms

[TECHNOLOGY]

CONSENSUS_MECHANISM
Bittensor subnet scoring / Yuma Consensus
BLOCKCHAIN_TYPE
Bittensor subnet 51; GPU marketplace
PROGRAMMING_LANGUAGES

Scalability & architecture

  • A CPU coordinator manages one or more GPU nodes. Providers can self-host the coordinator or use a Lium-operated central provider server; GPU nodes remain provider-operatedSourcedocs.lium.io · Architecture
  • Programmatic pod management, file transfer, volumes and backups are exposed through a Python SDK. This is application-level compute scaling, not a separate blockchain throughput claimSourcedocs.lium.io · Lium

[TOKENOMICS]

MAX_SUPPLY
21.00M
EMISSION_SCHEDULE

Bittensor documents a 21 million cap for each subnet alpha token, including subnet 51. Issuance is not liquid float: pool reserves and staked positions are included in issued supply. Lium describes provider rewards paid in alpha, with a 9% rented pool and a shared 91% ceiling split dynamically between eligible unrented nodes and designated burner UIDs. These percentages refer to SN51 provider emission, not total network supply or an annual yield. Current issuance, circulating supply and net burn were not measuredSourcesbittensor.com · Moneydocs.lium.io · Emission

TOKEN_UTILITY

SN51 represents subnet-specific alpha exposure within Bittensor, not shares in Datura AI Corp or ownership of GPU hardware. Providers receive rental earnings and subnet rewards in alpha. Staking and unstaking exchange TAO and alpha through the subnet pool, exposing holders to fees and price impactSourcesbittensor.com · Moneydocs.lium.io · Rental Feeslium.io · Terms

[CHART]

TradingView • BINANCE:SN51USDT

[MARKET_DATA]

PRICE_USD
$23.04
Market Cap
$116.99M
Fully Diluted Valuation
$116.99M
24h Volume
$1.58M
24h Change
+5.44%
All-Time High
$37.11(Jul 17, 2025)
All-Time Low
$0.164238(Mar 26, 2026)

Last updated: Nov 21, 2026, 9:00 PM UTC

[ECONOMICS]

Trading pairs

PairExchangeSource
SN51/TAOBittensor subnet 51 poolSourcesbittensor.com · MoneyGitHub · Datura-ai/lium-ioSource ↗
SN51 is the dossier ticker for subnet-51 alpha. This is a protocol alpha/TAO staking pool, not a centralized-exchange order book. Staking and unstaking incur pool fees and price impact; no live pool reserves, spread, executable depth, CEX listing or holder concentration were measuredSourcesbittensor.com · MoneyGitHub · Datura-ai/lium-io

[ECOSYSTEM]

Integrations

Bittensor Subnet 51Incentive network

Provider hotkeys and validator scoring connect the rental marketplace to subnet emissionSourcesdocs.lium.io · Architecturedocs.lium.io · Emission

Intel TDX and NVIDIA Confidential ComputingConfidential-compute stack

Eligible CVM hardware provides isolation and remote-attestation primitives. This is a documented technology integration, not a claim of a commercial partnershipSourcedocs.lium.io · Cvm

StripePayment processing

Service terms identify Stripe for card top-ups; the sanctions policy describes payment screeningSourceslium.io · Termsdocs.lium.io · Export Compliance

dApps

Lium GPU marketplaceGPU compute

Renters deploy metered pods on independently operated GPUs, through web and programmatic interfacesSourceslium.io · TermsGitHub · Datura-ai/lium-io

Provider PortalInfrastructure management

Providers register nodes, set constrained pricing, monitor rentals and view payment informationSourcesdocs.lium.io · Architecturedocs.lium.io · Rental Fees

[UNIQUE_VALUE_PROPOSITION]

Core value proposition

Combines an hourly GPU marketplace with Bittensor-funded provider incentives and developer-friendly pod orchestrationSourcesGitHub · Datura-ai/lium-iodocs.lium.io · Emission

Key differentiators

1

Provider earnings combine a documented 95% share of rental payments with separately scored subnet emission; the platform retains the remainder of rental paymentsSourcesdocs.lium.io · Rental Feesdocs.lium.io · Emission

2

Provider documentation distinguishes self-hosted and managed coordinators, while maintaining independent GPU hostsSourcedocs.lium.io · Architecture

3

CVM-capable hosts add hardware isolation unavailable on ordinary rented containersSourcesdocs.lium.io · Securitydocs.lium.io · Cvm

Competitive edge

A CLI and Python SDK reduce manual deployment steps; emission can reward eligible idle capacity before rental utilization grows. Subsidized supply is not proof of sustainable demand or lower total workload costSourcesdocs.lium.io · Liumdocs.lium.io · Emission

Target audience

AI developers, researchers and agents renting NVIDIA GPU capacity, plus hardware providers seeking rental and subnet-reward incomeSourcesGitHub · Datura-ai/lium-iodocs.lium.io · Architecture

[COMPARATIVE]

Competitive positioning

Relative to a conventional GPU rental service, Lium adds Bittensor alpha incentives and validator-scored participation. Relative to a fully permissionless compute network, its single team validator, hosted billing and service terms introduce substantial operator dependence. This comparison is structural, not a measured market-share or price rankingSourcesGitHub · Datura-ai/lium-iodocs.lium.io · Validatorslium.io · Terms

Strengths

Weaknesses

  • A single team-operated validator creates a scoring-control concentration explicitly documented by the projectSourcedocs.lium.io · Validators
  • Non-CVM hosts can access sensitive workload material; container isolation alone does not protect against the hardware operatorSourcedocs.lium.io · Security
  • Idle rewards depend on changing model caps, eligibility and competition; token income should not be treated as a fixed hardware returnSourcedocs.lium.io · Emission

[ROADMAP]

Roadmap

GPU marketplace and programmatic toolingSourcesGitHub · Datura-ai/lium-iodocs.lium.io · LiumCOMPLETED

Confidential-compute provider pathSourcedocs.lium.io · CvmCOMPLETED

  • CVM deployment instructions and hardware requirements are published. Deployment coverage and independent attestation testing were not measured; no dated future roadmap was establishedSourcedocs.lium.io · Cvm

Risk factors

Outlook

Bull case: repeat paid workloads sustain utilization and rental income while the SDK and confidential nodes improve usability. Bear case: demand stays subsidy-dependent, alpha rewards fall or operator/security constraints deter customers. Neither published reward formulas nor GPU availability establish a guaranteed yield or token priceSourcesdocs.lium.io · Rental Feesdocs.lium.io · Emissiondocs.lium.io · Securitydocs.lium.io · Cvmlium.io · Terms

[GOVERNANCE]

Governance model

Operational administration is team-led. Lium documents a single team-operated SN51 validator, while the service operator controls admission, account suspension and platform terms. This should not be described as a verified token-holder DAOSourcesdocs.lium.io · Validatorslium.io · Terms

[SECURITY]

Security practices

  • Non-CVM hosts retain privileged access to pod files, environment variables and process memory. Documentation warns against placing private keys, API secrets or proprietary model weights on such machinesSourcedocs.lium.io · Security
  • CVM nodes use Intel TDX hardware isolation and attestation, with SGX and NVIDIA Confidential Computing requirements. Supported GPUs are Hopper and Blackwell; ordinary consumer GPU pods must not be assumed confidentialSourcedocs.lium.io · Cvm
  • Validators check node hardware, performance and the required Sysbox runtime. These operational checks are not an independent security audit or a proof of arbitrary workload correctnessSourcesdocs.lium.io · ArchitectureGitHub · Datura-ai/lium-io
  • Support guidance warns about impersonation and seed-phrase theft and directs users to an official Discord ticket. It is not a disclosure of a protocol exploit or quantified lossSourcedocs.lium.io · Official Support And Scams

[REGULATORY]

Compliance status

The service prohibits restricted destinations and sanctioned parties, including specified ownership-controlled entities. Enforcement described includes payment screening, account-origin review and node-location review. These are company policy statements, not certification of complete complianceSourcedocs.lium.io · Export Compliance

Jurisdictions

[COMMUNITY]

[GITHUB_REPOSITORIES]

Datura-ai/lium-ioPRIMARY

Public provider node, miner and validator implementation with links to CLI, documentation and supportSourceGitHub · Datura-ai/lium-io

[SOCIAL_MEDIA]

[COMMUNITY_PLATFORMS]

[TECHNICAL_DETAILS]

Block Explorers

Node requirements

Minimum Hardware
Separate CPU coordinator and supported NVIDIA GPU node, with NVIDIA drivers, Container Toolkit and required sysbox-runc runtime. The live model list is authoritative; GPUs not on the validator allow-list are rejectedSourcedocs.lium.io · Architecture
Recommended Hardware
For confidential workloads, compatible Intel TDX/SGX hardware and Hopper or Blackwell GPUs with NVIDIA Confidential Computing are required; standard consumer GPUs are not CVM-capableSourcedocs.lium.io · Cvm

API Endpoints

GPU model catalogue
https://lium.io/api/machines
Documentation →

Technical specifications

Chain Design
A provider coordinator signs with a registered SN51 hotkey. Validators probe GPU nodes, and renters run container workloads on those nodesSourcedocs.lium.io · Architecture
Subnet Identity
netuid 51 is a Bittensor subnet identifier, not an EVM chain ID. CryptoIA retains the celium slug while the official product and repository use LiumSourcesGitHub · Datura-ai/lium-iodocs.lium.io · Architecture
Consensus Detail
Yuma distributes subnet incentives based on validator scoring; this should not be confused with consensus over the contents of renters’ workloadsSourcesdocs.lium.io · Emissiondocs.lium.io · Architecture
Rental Economics
The service is prepaid and usage is metered per second. Published provider revenue share is 95%; rental payments and subnet emission are distinct income streams. Current protocol revenue and realized provider income were not measuredSourceslium.io · Termsdocs.lium.io · Rental Feesdocs.lium.io · Emission
Data Retention Risk
When prepaid balance reaches zero, the terms state pods are stopped and removed and local data is deleted. Backups and external storage remain importantSourcelium.io · Terms