In early 2025, a DAO-powered humanoid robot – designed to sort waste in a Tokyo recycling plant – malfunctioned and hurled glass bottles into a crowd. The culprit? A smart contract that gave an AI agent unchecked control over physical actuators, with no human-in-the-loop fallback. As blockchain evangelists, we often tout ‘code is law’. But when that code breaks a window – or a bone – we realize the protocol is not just an algorithm; it is a social contract with physical consequences. This incident crystallises the next frontier: blockchain must now govern the physical world, not just the financial one.
— Root: The 2022 Bear Market — we learnt that survival means respecting real-world constraints.
The context is clear. Since DeFi Summer, we have layered abstractions: base layer settlement, rollup execution, data availability sampling. Yet the vast majority of value still moves between digital tokens. Meanwhile, capital is pivoting. According to a recent market brief by Serenity (though focused on pure AI, the analogy is perfect for crypto), funding for “Physical AI” – what we would call Decentralised Physical Infrastructure Networks (DePIN) and on-chain autonomous agents – reached over USD 13 billion in 2024, second only to AI infrastructure itself. In our world, DePIN projects (Helium, IoTeX, Peaq, and fresh robotics-up streams) have raised over USD 3 billion, and the number of AI-agent tokens has exploded. But just as Serenity notes the shift from language models to world models, our industry is shifting from scaling blocks to scaling trust in the physical world. The era of pure digital speculation is yielding to the era of ‘intelligent settlement’ of real actions.
Yet we must be careful. “Code is law, but people are the protocol.” My experience auditing Uniswap’s early governance in 2020 – we published a 50-page whitepaper on democratising liquidity – taught me that even the most elegant code requires human coordination. Now, as we embed AI into smart contracts that control drones, locks, and energy grids, the complexity spike rivals that of Uniswap V4’s hooks. Remember my earlier warning: Uniswap V4’s hooks turn the DEX into programmable Lego, but the complexity spike will scare off 90% of developers. The same applies here. Pushing AI inference on-chain, especially for physical actions, introduces failure modes that pure finance never had: oracle latency, real-time safety constraints, and the need for revocation mechanisms that cannot rely on a 7-day timelock.
— Root: DeFi Summer — we learned that decentralised governance is fragile when delegation concentrates power. Users are too lazy to research and simply delegate to KOLs. Now, imagine delegating your physical safety to a KOL’s AI agent.
Let me unpack the seven dimensions of this paradigm shift, mirroring the analytical depth of the original report, but rooted in our blockchain reality.
1. Technical Route Analysis The move from EVM-compatible rollups to “physical AI” on-chain is a change in computational paradigm. Rollups optimise for throughput and finality; physical AI requires real-time, verifiable inference. Zero-knowledge proofs (ZK) are the bridge. Projects like o1 and Nexus are building ZK provers for neural networks, but latency remains a killer. In my own mentorship sessions during the 2022 Bear Market (the “Resilience Hub”, where we supported 200 junior developers), I saw how quickly teams abandoned novel tech when the market turned. The same will happen to pure ZK-AI stacks if they cannot deliver sub-300-millisecond proofs for a robot’s braking decision. The hidden risk: most teams are building on top of centralised AI APIs (OpenAI, Claude) and wrapping them in smart contracts – that is not decentralised, it is a facade.
2. Commercialisation Analysis On-chain AI applications (generative art, automated market makers with AI predictions) are the “AIGC” equivalent of our world – most mature, but no clear winner. Midjourney-style NFT generators are commoditised. DePIN, on the other hand, shows real revenue: Helium’s mobile network has 100,000+ subscribers; IoTeX’s machinefi credits are used in supply chains. Yet the holy grail – autonomous agents that earn and spend crypto for physical services – is pre-revenue. Serenity’s comment that “there are no pure-play stocks” mirrors our landscape: no major DePIN company has IPOed, and most trade on sentiment. The investor’s question: will the market reward a robotics DAO with a P/S ratio, or is it pure narrative?
3. Industry Impact Analysis The shift will reshape blockchain’s industry structure. Employment: DAO governance roles will expand; robot auditors and simulation engineers will emerge. Supply chain: DePIN needs custom hardware (sensors, edge compute) and local manufacturing. This pulls in traditional OEMs like Foxconn, which already produces Helium hotspots. The carbon footprint – physical AI consumes 100x more energy than a simple token transfer – will force blockchains to either subsidise green energy or face community backlash. We saw this with Ethereum’s proof-of-stake transition; now it will happen at the edge.
4. Competitive Landscape Analysis Layer-2s are oligopolising – Arbitrum, Optimism, Base capture 80% of TVL. DePIN is still fragmented: IoTeX leads among purpose-built L1s, but Peaq and DePIN X are closing. The true battlefield is AI-agent orchestration. Just as Serenity notes the absence of a clear leader in world models, we have no dominant “agent execution layer”. Startups like Autonolas, Fetch.ai, and Ritual are competing, but none have proven they can handle a city-wide robot swarm. The opportunity is wide – but so is the risk of a Terra-style collapse when an agent runs amok.
5. Ethics & Safety Analysis Serenity’s analysis omitted this dimension; I refuse to. The Tokyo incident was a warning. On-chain physical AI demands new governance primitives: “emergency stop” circuits that bypass the DAO in milliseconds, liability escrows that auto-compensate victims, and oracle networks that verify physical events. We borrowed these ideas from our DeFi audit days – remember the 2020 cover protocol? – but now they must be mandatory, not optional. The “alignment problem” for robots is harder than for LLMs; you cannot just feed it more data; you need hardware-level invariants.
6. Investment & Valuation Analysis The investment thesis is clear: early-stage round sizes for DePIN/agent projects have doubled in 2025, while plain L2 rounds are flat. Multiples are lofty: a robotics DAO with a testnet and a 10-minute YouTube demo can command a USD 50 million valuation. This smells like the 2021 NFT mania. My contrarian view – and here I channel the Bear Market lessons – is that the highest-valuation projects today will be the hardest-hit when the next correction comes. I tell my mentees: focus on projects that generate real fees from physical services (e.g., a solar energy marketplace) rather than pure AI agent ambitions.
7. Infrastructure & Compute Analysis Just as world models need simulation engines, our physical blockchain needs verifiable compute. TEEs (trusted execution environments) are one solution, but they are centralised. ZK-proofs are another, but they are slow. The dark horse is fully homomorphic encryption (FHE) for private, on-chain AI inference. In 2026, I convened a working group on autonomous agent accountability (the “Autonomous Agent Accountability Charter”). We concluded that infrastructure must evolve beyond ‘rollup-centric’ thinking to ‘computation-centric’ blockchains – think Intel’s SGX meets Ethereum. Until then, the latency of proving will limit real-world use to low-stakes tasks like vending machines, not cars.
— Root: DeFi Summer — I recall how Uniswap’s early governance needed town hall meetings; now physical AI governance will need ‘physical town halls’ with hardware audits.
Now, the contrarian angle. Everyone is bullish on physical AI, but I see a blind spot: delegation centralisation. In current DAO governance, 80% of voting power is delegated to 5 large entities. If those entities control the revocation keys or training models of physical agents, we end up with a plutocracy that can literally turn off your robot. The very technology we tout as liberating could become a tool for control. “Governance isn’t a technical problem; it’s a trust problem.” We need to design delegation systems that expire, require multi-sig from geographically diverse humans, and include a built-in social recovery for when an agent is hijacked.
My takeaway is not a prediction of glory, but a call to action. The industry is about to spend billions on hardware and code. Yet the most critical component – community oversight – remains underfunded. Let us build not just the most efficient physical AI execution layer, but the one that remains accountable to the people it serves. Because at the end of the day, code is law, but people are the protocol.