$6.5m · Growth/Late · AI · London, UK
Perceptron, a decentralised network for sourcing and verifying AI training data, has closed a $6.5m strategic round. The financing brought together Sigma Capital, Selini Capital, QCP Capital, P2 Ventures, CoinDCX Ventures, Momentum6, DeFi Capital, Walrus Foundation, Aethir, Colosseum, GuruDev Capital, Tempo Finance, NewTribe Capital, Digital Consensus Fund and CodeCraft Capital. Perceptron said the capital will support the launch of its data-questing platform, expand contributor tools and rewards infrastructure, and help scale the network towards five million nodes.
Turning contributor reach into a data supply chain
Perceptron is trying to connect two sides of the AI data market. Companies can commission specific datasets through “data quests”, while individuals contribute bandwidth, local knowledge or specialist expertise and receive rewards. That shifts the network from passive collection towards a demand-led workflow: a customer defines the data it needs, contributors gather it, and the platform packages the result for delivery. The company says this can reduce reliance on centralised scraping and closed data partnerships, but the sources reviewed do not disclose customer pricing or name users of the commissioning product.
The supply side is already sizeable on the company’s account. Tech.eu reported more than 300,000 daily active users across over 807,000 nodes, after an earlier phase reached 200,000 users through communities including Telegram and Discord. That reach can make unusual languages, professions and local contexts easier to recruit, but raw participation is not the same as dependable training data. Perceptron will need to show that commissioned quests can produce consistent, traceable outputs without letting contributor incentives reward volume over usefulness.
Quality control is the next execution test
Perceptron’s published roadmap put the first data-questing release in the third quarter of 2026, followed by voice, image and market-connect quests in the fourth quarter. Its validator network, enterprise connections and a model-quality layer were listed for the second quarter of 2027. Those are plans rather than delivered capabilities. Commercial progress will therefore depend on whether the company can turn its broad contributor base into datasets that enterprise customers can verify, integrate and reorder, while preserving the ownership and reward model promised to contributors.
Explore this theme
Building AI products: 5 lessons from our founders' workshop — Lutz Finger, Cherry and Jasper Masemann; partner-authored material hosted on EUVC · 31 July 2024. Its discussion of minimum viable quality and training-data bias sharpens the central challenge for a distributed data network: useful coverage only matters when the resulting datasets meet a customer’s quality threshold.
