Inference work. Verified on-chain. Node operators earn.
Ambient runs a 600-billion-parameter model across a decentralized PoW network. Every inference call is a provably verified economic event. No cloud provider, no GPU bottleneck.
Familiar tooling. Purpose-built for compute.
Ambient is SVM-compatible — developers deploy existing Solana toolchains directly onto a chain engineered from the ground up for AI compute workloads, not retrofitted for them.
PoW consensus ties each inference call to block production. Completed model calls generate cryptographic proof, making every inference a verifiable, economically settled transaction on-chain.
Choose your workload. Choose your reward.
Inference Node
Fine-Tuning Node
Training Node
Runs adapter training passes on network fine-tunes. Reward schedule weighted by dataset contribution and verified gradient updates. Mid-tier hardware eligible.
Contributes to base model training epochs. Highest reward tier, proportional to verified compute contribution. No enterprise GPU required — distributed workload sharing applies.
Executes live model calls against the 600B-parameter network. Rewarded per verified inference event. Minimum: consumer-grade GPU with 16 GB VRAM.
Evaluate the spec. Then run a node.
The whitepaper covers consensus mechanics, inference pipeline design, and reward schedules in full. Start there, then follow the node setup guide to go live.