Decentralized artificial intelligence (AI) as a movement is not only about redistributing power; it also emphasizes the importance of enabling previously impossible use cases for both technologies. By applying the crypto-native ethos of decentralization and self-sovereignty to AI, this industry sector seeks to mitigate concerns over privacy, equity, and accessibility. The convergence of crypto’s efficiency, borderlessness, and programmability with the rapid pace of innovation in AI has the potential to transform the future of nearly all economic activity. After explaining some observations about decentralized AI in practice, this article breaks down the decentralized AI stack as it is built today in the following categories: 

• Agents
• Decentralized Compute
• Confidential Compute
• Data Collection
• Training
• Developer Platforms
• Gaming
• Monetization
• Predictions
• Trading Algorithms
• Agents
• Compute

A New Era for AI in Crypto

Converging AI with crypto is a nascent development, but a lot of infrastructure is being built to support it. And the beginnings of the agentic web are visible. 

Agents will be executing yield farming strategies in the near future, and LLMs will provide medical advice in a verifiable and privacy-preserving way. In theory, crypto could also see agents issuing their own digital currencies to streamline agent-agent interactions. 

On the “user” side, decentralized AI enables participation in the governance and value creation of these systems. It also promises a future with more transparency around training and inference and credible fact-checking (e.g., models referencing block explorers or onchain social data).

Much of the potential of decentralized AI hinges on the belief of a natural symbiosis between crypto and AI. Crypto can enhance the capabilities of AI technology via permissionless and composable infrastructure. With these enhancements come new use cases that include open-source model development, verifiable inference, immutable ledgers, and democratized access to AI resources like compute, storage, bandwidth, and data. 

But this innovative relationship is not unidirectional. AI also stands to improve crypto infrastructure for users and builders. Developer experience can be leveled up by integrating AI models into smart contract development and agents can help abstract blockchain complexities away from users.

Deep Dive: The Decentralized AI Stack 

Growing interest and development within the new decentralized AI paradigm is creating a complex, rapidly accelerating landscape that changes constantly, and trying to keep up is no easy task for the average reader. Oasis has produced snapshot of the decentralized AI landscape that is shown below. Each of the sections that follow the landscape visual provide pieces of all the specific use cases and emergent areas of development in this sector.

Agents

Agents are having a moment in crypto. Driven partly by the popularity of Truth Terminal and related memecoins, this sustained trend also includes a wide array of emerging agentic products. Users and builders have high expectations for increased crypto adoption due to agents based on opinions that these products are seen as either automation tools that improve UX or as intelligent assistants that will be paid in crypto. Some notable projects in this category are listed below:

• Decentralized Finance:‍
  • OMO - Reliable agentic yield farming‍
  • Intentify - Operating system for autonomous onchain AI agents‍
• General:‍
  • Virtuals Protocol - Agent creation engine‍
  • PAAL - Swarm intelligence for a variety of use cases‍
  • Hyperspace - Generative browser powered by peer-to-peer AI network‍
  • NetMind - Decentralized compute platform‍
• Infrastructure: ‍
  • Theoriq - Base layer for multi-agent systems‍
  • Naphtha AI - Modular AI platform for autonomous agents‍
  • Giza - The full Web3 agent stack‍
  • Olas - Protocol & dev tools for agent creation & management‍
  • Fetch.ai - An open marketplace for AI agents‍
• Large Language Models:‍
  • Venice AI - ChatGPT with privacy and censorship resistance ‍
  • Supersight - LLMs specified for crypto use cases‍
  • MagnetAI - programmable action agent platform, tokenization of AI models

Decentralized Compute

As AI models become increasingly complex, demand for compute grows, and there’s a scarcity of top-tier GPUs, resulting in long wait times and rising costs. Computing resource centralization is an increasing issue, and small research groups have had difficulty accessing H100s. Decentralized compute is therefore one of the largest segments and it solves a real problem in the AI space. Some leading projects in this category are: 

• Hyperbolic - Open AI cloud service with GPU marketplace
• Aethir - Distributed cloud computing infrastructure
• Akash Network - Decentralized compute marketplace 
• Bittensor - A platform focused on the decentralized production of AI
• Gensyn AI - Decentralized compute platform for machine learning
• SpheronFDN - P2PGPU marketplace 
• Cluster Protocol - Decentralized compute platform 
• Destra Network - A dedicated GPU network 

Confidential Compute

Confidential computing via cryptographic proofs or hardware attestations is crucial in terms of sharing compute resources, as in decentralized AI. It’s currently one of the missing building blocks for broader DeAI adoption. Here are several good examples:

• Oasis - The network for AI verifiability
• Marlin Protocol - GPUs for onchain execution
• VannaLabs - Decentralized execution layer for AI/ML
• EZKL - Prove authenticity of AI models and verify execution 
• Nillion - Confidential compute network for AI training & inference 
• Arcium - Parallelized confidential computing network
• Aziel Network - Modular network dedicated to verifiable AI
• Automata - A modular attestation layer 

Data Collection

Large models require an almost infinite supply of data to scale. But there is only so much public, human-generated data, which raises the question of whether the lack of training data could stall model performance improvements. Deep web and proprietary data access may help keep the LLM growth trajectory intact, and numerous DeAI projects are working in this direction. Some examples of this work are:

• Masa AI - An AI data network for monetizing personal data.
• Grass - A decentralized web scraping protocol 
• Sarvis - Train custom AI models in a privacy-preserving way
• Sahara AI - A platform for B2C knowledge sharing
• Rivalz - Decentralized solution for data processing and storage

Training 

There are a growing number of projects that enable incentivized and verifiable training or fine-tuning mechanisms onchain. Distributed training for large foundational models is often viewed as unrealistic due to technical complexities, but recent improvements in frameworks like Distributed Low-Communication (DiLoCo) seek to shift that narrative. Some interesting examples in this space include:

• Prime Intellect - train state-of-the-art models through distributed training across clusters 
• Synesis One - A decentralized data crowdsourcing platform for training 
• NEAR Tasks - Users complete micro-tasks that train AI for rewards 
• Sapien - A platform for labeled data essential for training AI models
• Flock.io - A decentralized, collaborative training platform

Developer Platforms

Both generalized and AI-specific blockchain platforms want to attract developers to use their tools to improve their AI stack. Strong communities are emerging, and for now there is no right or wrong approach. Consider these examples:

• 0G Labs - A modular AI chain, acting as DA layer for AI dApps
• NEAR - A platform that combines blockchain abstraction with AI capabilities
• Oasis Sapphire - Dev platform optimized for building confidential/secure dApps
• Galadriel - A Layer 1 blockchain specifically built for AI projects
• SingularityNET - A library of AI algorithms
• Pond - A model development base designed for all types of dApps
• Nimble Network - An agent infra provider that supports dApps

Gaming 

Crossovers between crypto and gaming were an easily understood and widely circulated narrative. Adoption of this narrative, on the other hand, has been something of a slow burn. But with AI now in the mix, there is a resurgence of gaming applications utilizing the same computing resources for rendering and generative AI (a.k.a., GenAI or GAI) capabilities within their games. Here are some examples:

• Alethea.ai - AI agents with unique emotional use cases, expressive gaming interactivity
• AI Arena - A gamified AI education platform
• Momo.ai - Develops AI bots to build community within the web3 gaming space
• WORLD3 - Bitcoin-based AI gaming project
• Play AI - Offers an intelligence layer for integrating AI capabilities into gaming
• iAgent - AI agents specifically for gaming applications
• Chibi Clash - An AI-driven game built on Base 

Monetization

Tightly connected to the data collection projects are those builders who think about ways to monetize either participants in the data economy, model builders, or even the model inference. This is an exciting space with lots of ongoing development from new teams, such as:

• deltaDAO - Open data economy that utilizes onchain tech
• Ritual - Facilitates monetization of AI resources via collaborative infrastructure
• Ocean Protocol - Privacy-preserving platform to monetize AI models & data 
• Inferium.io - A marketplace specifically for inference models
• Desights.ai - Train models on custom data & monetize 
• Allora - A marketplace for machine intelligence
• Sentient - Decentralized artificial general intelligence 

Predictions

Prediction markets are popular. Especially recently, when it even broke into traditional media due to the US election. Utilizing AI models to predict outcomes like these is not far-fetched. Many dApps are trying to stand out in this particular category, including:

• yPredict.ai - Ai-powered crypto market predictions & research
• Kryll - An automated crypto trading platform 
• predicting AI - AI-driven market-making
• Predictoor - Offers AI-based predictions tailored for traders
• PredX - A prediction-focused platform with AI insights & forecasts 
• yAI - Prediction market platform 

Trading Algorithms 

Prediction markets but for crypto or stock prices. Historically, it has been difficult to do, but with a massive upside. There’s no shortage of players attempting to beat the market. Here are a few examples:

• Dither - AI time series models for trading
• Numerai - Uses ML to predict stock market movements
• Hummingbot - Swarm intelligence for trading, enabling automated strategies 
• Cryptoindex - An AI-powered crypto indexer
• Taurus AI - Offers a marketplace for market trading bots