DeAI
Decentralised AI. An umbrella term for blockchain-based projects that build AI infrastructure (compute, data, inference, models, agents) without a single central provider controlling the system.
Also known as: decentralised AI, decentralized AI, crypto AI
DeAI is the subject of Own Your Mind. The term covers any crypto-native attempt to build AI infrastructure that doesn’t rely on a single central company to run it. That includes decentralised compute marketplaces (Akash, Render, io.net, Aethir), decentralised inference networks (Venice, Morpheus, Bittensor subnets, Sentient), decentralised training efforts (Templar, Prime Intellect, Nous Research Psyche), decentralised data marketplaces (Ocean, Vana, Grass), decentralised agent platforms (Virtuals, Autonolas, ElizaOS), and the privacy-preserving infrastructure underneath all of them (Phala, Oasis, Nillion).
The honest problem with DeAI as a category is that “decentralised” gets used very loosely. Some DeAI projects are genuinely permissionless networks with open governance, fair launches, and no single operator with kill-switch authority. Others are regular startups that bolted a token onto a centralised product so they could raise on crypto terms and call themselves decentralised. Both get labelled DeAI in marketing. The OYM Freedom Score exists precisely to separate the two.
Three tiers of DeAI are useful to keep in mind. Core DeAI means AI is the founding thesis and the protocol is designed around AI workloads from day one (Bittensor, Gensyn, Venice, Morpheus, Sentient). DeAI-adjacent means the protocol is general-purpose but AI is a primary use case (Phala, Nillion, Akash, Oasis). AI-washed means an existing crypto project bolted AI onto the marketing after the narrative shifted (many failed DeFi projects that rebranded in 2024-2025). Own Your Mind only reviews the first two tiers; AI-washed projects don’t make the cut.
The larger question DeAI is trying to answer is whether AI infrastructure can be built in a way that no single provider controls it. The answer today is “partially, in specific domains, with real tradeoffs.” Decentralised training is still slower and more expensive than centralised. Decentralised inference can match centralised on latency but usually trails on model quality. Decentralised compute marketplaces work but rely on subsidies to stay competitive. None of this means DeAI is fake. It means the current state is closer to early infrastructure than mature product, which is exactly where a research site covering the space matters most.