More than a buzzword, DeFAI is a combination of artificial intelligence (AI) and decentralised finance (DeFi) and could herald Web3’s next frontier, merging AI’s analytical prowess with blockchain’s trust less infrastructure to tackle fragmentation, opaque transactions and volatile decision-making. Projects leverage AI to parse cross-chain data, predict market shifts and automate execution - potentially democratising DeFi and neutralising exploitative practices such as maximal value extraction (MEV). Yet the DEFAI sector has fallen by over 80%, from $7billion to $1.4billion and faces existential tensions: reliance on centralised AI models, regulatory ambiguity and ethical risks of autonomous systems. And, as technical and governance challenges mount, DeFAI’s defining struggle lies not only in innovation but in balancing decentralisation against AI’s centralised underpinnings. So, will it forge an open, intelligent financial future - onboarding the next generation of Web3 users - or will it replicate the power structures it aims to disrupt? The stakes for trust and autonomy in finance have never been higher.
Year after year, Web3 pushes relentlessly toward mainstream relevance. 2024 will be remembered as the year that real-world asset tokenisation went supernova - rewiring global finance, democratising digital asset ownership and injecting $trillions into blockchain ecosystems. But, even as tokenisation reshapes economies, 2025 is poised to unleash Web3’s next evolutionary leap: its fusion with AI. Since OpenAI’s ChatGPT ignited the global AI arms race, developers have raced to merge machine learning’s raw potential with industries such as healthcare and real estate. And now, blockchain’s trust less infrastructure is the latest frontier. The collision is here and, whilst its name remains debated - DeFAI, AiFi, or OATs (Onchain Agent Terminals) - the implications are undeniable. The race for domination in the space has kicked off, with myriads of projects launching across different blockchains, including:
· Hey Anon - deployed on Sonic, Solana, Base and Arbitrum
· Paal AI - Ethereum
· Autonolas - Ethereum, Gnosis Chain, Base and Optimism
· Orbit - compatible with protocols across 100+ chains, including all EVM-compatible networks, Solana, Aptos, Sui and Bitcoin.
Image source: TeamBlockchain/Coingecko
To date, the growing DeFAI category consists of at least 8,000 projects, including Aixbt (AIXBT), Griffain (GRIFFAIN), Hey Anon (ANON) and Orbit (GRIFT). Each are attempting to carve out a spot in the DeFAI sector which currently sits at $901 million, down from its peak market cap of $7 billion in early 2025 - a mere 2% of DeFi’s $118 billion market, according to CoinGecko. The market has been experiencing some downtime but Ryan McNutt, founder of Orbit, believes the sector is warming up. However, there does appear to be a wave of challenges with only 27 out of the 82 AI agent projects launched on Solana during its Hackathon still standing. McNutt ascribes the downtime to the launch of DeepSeek, a Chinese AI model that also tumbled the entire ecosystem, including Bitcoin. Yet DeFAI’s potential lies in addressing four critical bottlenecks plaguing Web3 - gathering fragmented data across chains, interpreting opaque on-chain activity, making decisions amid volatility and executing actions securely. Blockchains generate oceans of data but privacy protocols such as Monero or Zcash obscure transaction details, whilst cross-chain ecosystems from Ethereum to Solana scatter data into silos. DeFAI aims to streamline this, leveraging AI to dissect through-noise, detect patterns in market trends and flag anomalies. Imagine an AI agent scanning Telegram, Discord and on-chain activity to predict a meme coin pump before it trends.
Source: Teamblockchain/ TK Research
But raw data alone is meaningless without interpretation. Web3’s transparency becomes a double-edged sword - whilst anyone can view a wallet’s transaction history, few can understand what happens behind cryptic wallet addresses or tangled intelligent contract interactions. Traditional platforms such as Dune Analytics help, but DeFAI tools such as Fetch.ai’s AI agents go further, translating blockchain jargon into plain language. For instance, AI could alert users that a sudden spike in stablecoin deposits to Aave signals incoming volatility - or warn that a DeFi protocol’s “innovative” yield mechanism resembles a Ponzi scheme. Yet challenges persist. How do you train AI models on non-standardised data? And, can decentralised networks such as Bittensor compete with centralised cloud giants when processing petabytes of blockchain data? Michael Casey, co-founder of the Decentralised AI Society, notes that “AI will face increasing pressure to be regulated, and big players like OpenAI are lobbying for rules that align with their models”, making it harder for decentralised AIs to compete. Currently, a body of DeFAI agents are trained and dependent on centralised Ais and, aside from centralisation, they will also face technical and regulatory challenges.
Decision-making in this environment is equally fraught. DAOs debate governance proposals for weeks whilst traders juggle APYs, liquidity pools and impermanent loss calculations. Nonetheless, DeFAI promises to automate this chaos. Picture AI simulating 10,000 outcomes of a Uniswap proposal to adjust fee tiers and then recommending the optimal choice based on data. Or, imagine a bot rebalancing a portfolio across Ethereum, Solana and Cosmos in real time, hedging against a predicted market dip spotted in social sentiment. Although projects such as SingularityNET are already testing such tools, risks continue to linger. Flawed training data could lead to cascading errors, and opaque “black-box” AI models might make decisions that even developers cannot explain an example being when a crypto user convinced an AI agent to transfer $47,000 worth of Ether. Fortuitously, it was a test. Moreover, execution is where theory meets reality; even the best strategy fails if a trade executes late, there is a smart contract glitch or a validator censors transactions. But this is where DeFAI’s promise shines - AI-powered platforms such as Flashbots combat MEV (exploitative strategies where bots front-run trades) by optimising transaction ordering. Meanwhile, agents automate complex workflows, such as collateralising a loan on MakerDAO, staking rewards on Lido and insuring the position on Nexus Mutual - all in one atomic transaction. Yet, reliance on oracles for real-world data remains a vulnerability, and regulatory ambiguity looms. For example, the EU’s AI Act demands transparency in automated decisions - a challenge for decentralised models.
So, the road ahead is both promising and perilous. Whilst DeFAI could democratise access to DeFi, enabling newcomers to navigate liquidity mining or DAO governance with AI-guided simplicity, it raises ethical questions. Should AI be allowed to spawn thousands of meme coins, flooding the market with speculative assets? And, could the large language models (LLMs) powering these agents - often trained on closed-source data - inadvertently introduce biases, favouring Ethereum-based protocols over Solana, or vice versa? Whichever, the momentum is undeniable. As AI reshapes industries from healthcare to logistics, its marriage with Web3 could redefine ownership, governance and value creation. The question is not whether DeFAI will mature, but how. Will it evolve as an open, decentralised force or replicate the centralised power structures it sought to dismantle? For now, the fusion of blockchain’s trust less rails and AI’s soon-to-be omniscient firepower offers a glimpse of a future where finance is decentralised and intelligently autonomous - and code does not simply execute, but adapts.
This article first appeared in Digital Bytes (25th of March, 2025), a weekly newsletter by Jonny Fry of Team Blockchain.