Datura AI Corp
Datura AI Corp operates Lium, a decentralized GPU-compute marketplace that connects independent GPU providers with renters needing on-demand AI/ML compute. Lium lets developers and AI agents launch containerized GPU pods from the web, CLI, Python SDK, REST API, or agent tooling, with per-second billing.
At a Glance
- Machine-learning engineers and AI researchers
- AI startups and indie developers
- Developers building inference, training and agent workloads
- Bittensor subnet participants and TAO holders
- +2 more
AI Tools by Datura AI Corp
(1)
Lium
GPU Compute Marketplace for AI
Discussions
No discussions yet
Be the first to start a discussion about Datura AI Corp
Latest News
Case study: how Ahurasense runs B300 workloads on Lium
Lium repository adds an MCP server for agent hosts to rent and use GPUs
Lium CLI and SDK 0.20.0 release
Lium publishes guides for LLM inference, LoRA fine-tuning, ComfyUI, vLLM, PyTorch training and giving Claude Code/Codex access to GPUs
Products & Services
Permissionless marketplace for renting provider-owned NVIDIA GPU machines as Docker-containerized pods. Live listings expose GPU, CPU, memory, VRAM, storage, network, location, reliability and price; rentals are metered per second with no minimum term.
Command-line interface for listing and quoting nodes, launching and removing pods, SSH, file transfer, spending, volumes, backups, provider operations and agent workflows.
Python package for programmatic pod orchestration, including a machine decorator for offloading functions to rented GPUs and a client API for renting, waiting, executing on and releasing pods.
OpenAPI 3.1-described API for account, pod, provider, storage, billing and marketplace operations.
Market Position
Lium positions itself as an agent-first, open and decentralized GPU marketplace with independent global providers, per-second billing, no minimum commitment and pricing often below centralized clouds. It competes directly with peer-to-peer or decentralized services such as Vast.ai and io.net, and with managed GPU clouds such as RunPod; its differentiators are Bittensor staking/slashing, open-source tooling, agent automation and broad provider-supplied hardware. Trade-offs versus managed competitors include variable availability and the absence of a formal uptime SLA or publicly stated enterprise compliance certifications.
Leadership
Founders
Pierre (known as Fish/Datura)
Bittensor community participant and large-scale miner who brought experience in Bittensor subnet mechanics and GPU operations; public project profiles identify him as the founder behind Datura/Lium.
Executive Team
Pierre (Fish/Datura)
Founder
Bittensor operator and former large-scale miner associated with the creation and operation of Datura and Lium/Subnet 51.
Founding Story
Public project histories describe Lium as beginning on Bittensor Subnet 51 in late 2024 under the name Celium. The initial vision was to create an open, decentralized alternative to centralized GPU clouds by aggregating independent providers, validating their hardware and performance, and matching spare capacity with AI/ML workloads; the project later rebranded to Lium.
Business Model
Revenue Model
Prepaid pay-as-you-go GPU rental marketplace. Renters top up by card through Stripe or supported cryptocurrencies and are charged per second while pods exist; providers set listing prices, Lium collects the rental payment and pays providers 95% of the rental fee. The network also uses Bittensor emissions and staking/slashing incentives for providers.
Pricing Tiers
Rates vary by GPU model and provider. The official marketplace has listed consumer GPUs from roughly $0.03/hour and H100 listings around $1.70-$2.50 per GPU-hour at the time of capture; there is no minimum rental period.
Target Markets
- Machine-learning engineers and AI researchers
- AI startups and indie developers
- Developers building inference, training and agent workloads
- Bittensor subnet participants and TAO holders
- GPU owners and infrastructure providers
- Organizations seeking lower-cost alternatives to AWS, Google Cloud and Azure GPU capacity
- AI model training and fine-tuning
- LLM inference and serving OpenAI-compatible endpoints
- Image generation and Stable Diffusion/ComfyUI workflows
- Speech and media workloads such as Whisper
- Short-lived experiments and burst GPU capacity
- AI coding-agent execution through Claude Code, OpenAI Codex and other agents
- Ahurasense