In January 2025 Nvidia CEO Jensen Huang outlined the next step in the evolution of artificial intelligence. In his view, it should be physical AI (Physical AI, PAI) — a global “brain” for robots and autonomous devices, enabling machines to navigate the real world.
The blockchain industry picked up the idea, proposing a decentralised version. In February the analytics platform Messari published a report on a new form of interaction among people, machines and Web3 — DePAI. The notion of an infrastructure based on DePIN and a stack for developing a future “robot economy” inspired crypto influencer Miles Deutscher. In April he called DePAI “the biggest trend in crypto for the next two years”.
This article explores how DePAI could help build a balanced future for people and machines.
At CES 2025 Huang announced Cosmos — a platform for building World Foundation Models (WFM) aimed at revolutionising systems of physical AI (PAI): robots and autonomous vehicles.
Cosmos uses WFM to model real-world conditions, allowing AI systems to be tested. A controlled testbed reduces costs and speeds development.
Huang noted the open nature of Cosmos, whose source code is available on GitHub.
“We sincerely hope that the openness of Cosmos will do for the world of robotics and industrial AI what Llama 3 did for industrial AI,” he added.
According to Nvidia, PAI enables autonomous systems — such as robots, driverless cars and smart spaces — to perceive, understand and carry out complex actions in the real world.
Large language models (LLMs) such as GPT and Llama can generate human language and abstract concepts, but are limited in their grasp of the physical world’s laws. PAI can complement them by training spatial understanding.
3D data for AI are generated via high-fidelity simulations that serve both as data sources and training environments. The process starts by creating a digital twin of a space, for instance a factory, then adding sensors and robots. As sensors capture rigid-body dynamics (movement and collisions) or the interaction of light with the environment, real scenarios are modelled in the virtual copy.
PAI opens new possibilities for many industries. In robotics, the concept can improve a range of unit capabilities across products:
A PAI model is rewarded for successfully completing required actions and improves over time. Machines gradually develop complex skills, including precise motor actions needed for real-world use: careful box packing, assisting in vehicle assembly or navigating spaces.
The crypto community embraced PAI, proposing blockchains as an antidote to AI centralisation. In a February report, Messari analyst Dylan Bane added the word “decentralised” to the acronym.
The acronym reads: DePAI — decentralised networks of physical artificial intelligence.
DePAI sits at the intersection of AI, robotics and Web3. It could power a future economy in which automation is balanced with human needs.
What began as generative AI for content is becoming autonomous agents capable of making decisions on their own. The remaining step is to give them effective “bodies”.
Messari’s DePAI map lists six main groups by activity.
DePIN projects such as the geodetic network GEODNET or the Google Maps alternative Hivemapper naturally slot into a new DePAI stack. The products of such startups can become robots’ ears and eyes in the real world and in real time, serving as their data-collection layer.
The startup WeatherXM incentivises users to deploy personal weather stations and upload climate data in exchange for tokens. Such data can be used by devices within DePAI. For example, integrated smart-home systems could automatically adjust ventilation or temperature based on current weather.
In robotics, the startup Frodobots uses DePIN to deploy low-cost “sidewalk” delivery robots worldwide. The data used helps address the complexity of human decision-making in real-world navigation. Gamers can practise controlling machines in an accessible format.
Also in the mix is a company creating digital twins of physical robots on the Polkadot blockchain — Robonomics Network. It explores connecting robot operating systems and IoT devices with a blockchain to publish tasks and deliver services via smart contracts.
Machine coordination networks are being built by blockchain platforms focused on DePIN — Peaq and IoTeX. Their high-throughput architectures provide the many parallel transactions DePAI requires.
The Posemesh data-transmission protocol from Auki focuses on creating a global network of spatial intelligence. It aims to ensure privacy and help build a virtual map for robots. Its functions are not limited to logistics and are also used for AI self-training in simulation.
In investment DAOs — Xmaquina. The project focuses on creating a structure for collective ownership, management and development of costly AI robots. It allows communities and investors to co-finance and develop technologies while distributing benefits fairly.
In a global machine economy, DePAI could, in theory, return power to people that currently belongs to corporations. Users already need to know who owns the technology, which manufacturer sets interaction standards and who ultimately earns the profits.
On April 14, 2025, the well-known crypto analyst Miles Deutscher called DePAI one of the biggest trends in the industry.
The image attached to the post illustrates the complexity of relationships within the DePAI stack. The dynamics, shown as vectors, indicate the purpose of actions among infrastructure participants.
The Peaq blockchain team, which is actively involved in DePAI’s development, highlights seven components every autonomous system needs:
While most of these are familiar to the average Web3 user, the future machine economy needs explanation. By uniting leading crypto technologies, the model could deliver the following benefits:
Like any new technology, DePAI faces serious challenges that must be solved before mass adoption. The main problems are:
The vision of a machine economy that ends to people’s advantage is enticing, but whether robotics giants such as Tesla and other corporations will take part remains an open question.
Web3, with blockchains as its foundation, is likely the best solution today for implementing decentralised AI. DePAI, for its part, clearly sets out the benefits for all sides of future progress — but who participates will be decided over time.
