Tokenomics

Architecture & stack

How tokenized capacity maps from accelerators through orchestration to inference — and how the platform wraps it.

Architecture

One stack — from bare metal to inference

Tokenized capacity sits on real accelerators. The same mental model as a hyperscale AI cloud, with ownership and liquidity built in.

Hardware

AMD MI300X / MI325X / MI355X

MI250X / MI210 / MI300A

InfiniBand fabric

NVMe + object tiers

Token layer

LUB-* compute tokens

Marketplace & spreads

Vault staking

On-chain style ledger

Orchestration

Job queue & priorities

Kubernetes pools

ROCm container images

VPC & networking

Inference & APIs

vLLM endpoints

Embeddings API

Batch & training jobs

Live WebSocket status

Platform