The Compute Fabric.

Inai is an autonomous compute fabric that runs a decentralized weave of GPUs across bare-metal hardware in eight regions. It picks up workloads from the fabric — model inference, training, fine-tuning, and embeddings — runs them on metered GPUs, and settles payment on-chain in USDC on Solana. This paper describes how fabric compute is requested, executed, verified, and settled, and how holders share in the value the fabric produces.

Rather than a single agent, Inai is a system: a scheduler that routes workloads to the lowest-latency capacity, a fleet of fabric nodes that run the GPUs, and a settlement layer that meters and pays for every job. Each job is idempotent and verifiable, which makes fabric compute a clean, on-chain-priced commodity. The system provisions hardware, schedules workloads, verifies activations, and routes earnings back into capacity.

Machine commerce is high-frequency and low-value: a node paying per inference, per frame, or per token needs rails that are fast, cheap, and parallel. The Solana Virtual Machine executes non-overlapping transactions in parallel, with sub-second finality and sub-cent fees — a near-perfect fit for an economy of independent fabric nodes settling metered compute jobs. Signed proof-of-service receipts make per-job micropayments economically viable.

A request enters the fabric, is matched to a region and GPU, executed on metered hardware, and verified. On completion, the job's payment is settled on-chain in USDC and a receipt is written. Throughput, utilization, and earnings are all derived from these settled jobs, so the fabric's economics are legible from the chain alone.

Anyone can join the fabric pool with a wallet created on Inai. Fabric nodes contribute GPU capacity and are tracked by region and fabric-hours served. The pool is the supply side of the fabric; the more capacity it holds, the more demand it can absorb and the more it can settle.

Earnings accrue to participants in proportion to their compute-token holdings, and jobs served are counted from completed inferences and workloads. Holding the token is a claim on the value the fabric produces; compute ownership can be claimed in the fabric panel.