Skip to main content
Kuzco, launched in 2024, was a distributed GPU inference network on Solana — “contribute your idle GPU, earn points” — backed by a16z CSX and Multicoin. In early 2025 the team abandoned the consumer crowdsourced-node model entirely, rebranded as Inference.net, and became a professional AI inference infrastructure company serving enterprise customers (SOC 2 certified; customers include Cal AI and Gravity Ads). The cause was not weak incentive design — it was a task-to-means mismatch: enterprise AI inference demands stable low latency and SLA guarantees, which GPUs scattered behind residential broadband structurally cannot deliver. Grass succeeded with millions of consumer nodes only because web scraping is naturally fault-tolerant. SeedNode Ring’s reputation computation is a data-indexing and inference task, not a compute-scarcity problem — so we do not use “run a node” as a growth mechanic. Where decentralization is genuinely needed is the verification and notarization of scores — closer to an oracle model with a small set of professional validators than to mass-market mining.