At DePIN Day Singapore, moderator Adam Wozney (Akash Network) gathered four builders shaping the decentralized infrastructure economy — Harry Dewhirst (375.ai), Clemens Koczur (Impossible Cloud Network), Mark Rydon (Aethir) and Markus Levin (XYO) to explore how artificial intelligence and DePIN are converging to power the next technological cycle.
Wozney opened with a simple observation: we are at an inflection point. AI is scaling faster than any technology before it, consuming vast amounts of compute, storage, and bandwidth — while raising deep questions around access, control, and energy use. DePIN, meanwhile, is redefining how these resources are owned and distributed. The intersection of these two forces could reshape the very fabric of the internet.
Centralization and participation
“AI is unfathomably huge — the only way it can scale is through decentralization,” said Harry Dewhirst, founder of 375.ai.
As models grow, so do their demands — for GPUs, data, power, and context. The only sustainable way forward, Dewhirst argued, is to distribute participation: allow individuals and communities to contribute resources and be rewarded for doing so.
“Every piece matters,” he said. “Compute, storage, sensing, energy — the more people who contribute, the more capable the system becomes.”
Markus Levin (XYO) agreed: the AI ecosystem today is limited not by ambition, but by scarcity. Compute and data are concentrated in a few places, leaving large parts of the world underrepresented. “If you train AI only on well-mapped regions,” he explained, “you end up with biased intelligence. DePIN can fill those blind spots.”
Levin pointed to XYO’s 600,000+ active nodes in Africa, collecting local data that could be used for weather prediction, agriculture, or geospatial AI — data that centralized players often overlook because it’s not tied to immediate profit.
A year of inflection
The panelists were unanimous that 2025 marks a turning point.
DePIN projects are maturing from vision to viability. “We’re seeing real revenue, real infrastructure,” said Mark Rydon (Aethir). “For years, people talked about potential — now we’re seeing fundamentals.”
The World Economic Forum predicts the DePIN market will reach $3.5 trillion by 2028 — a number once unimaginable, now increasingly plausible.
“DePIN is no longer a concept,” noted Clemens Koczur (Impossible Cloud Network). “It’s becoming part of the enterprise stack.” He pointed to growing corporate interest in data locality — the desire to store and train models closer to where data is generated, ensuring compliance, privacy, and competitive edge. In this new paradigm, DePIN isn’t a replacement for hyperscalers but a complement — offering resilience, security, and sovereignty.
Beyond GPUs: the balance of compute, data, and energy
Much of today’s conversation around AI revolves around GPUs. But as Harry Dewhirst reminded the audience, “It’s not just about GPUs — it’s about balance.”
AI’s progress depends on a delicate equilibrium between compute, data, storage, and energy. Too much of one without the others creates bottlenecks. “There’s far more compute right now than meaningful information,” he said, “which is why so much data has to be synthesized.”
For Levin, the solution lies in distribution — not only of compute but also of data ownership. “Most of the world’s data still comes from a handful of sources,” he said. “We need a global, diverse network feeding AI from the edge.”
Koczur added that decentralized storage will be essential for the next stage of AI adoption:
“Companies will increasingly train proprietary AI models on localized data they control. They’ll want to store it in their environment, not in a public hyperscaler. That’s where decentralized cloud makes sense.”
The energy question
No conversation about AI infrastructure is complete without addressing its energy appetite.
AI’s electricity consumption could reach 8% of global demand by 2030, equivalent to the entire power usage of Japan. Can DePIN help mitigate this?
The panelists were cautiously optimistic. Dewhirst predicted a future where households act as micro power stations, feeding energy back into decentralized networks. Levin emphasized economic incentives: “If the incentives are right, communities will come together to build renewable capacity — that’s where DePIN and AI align perfectly.”
Mark Rydon highlighted new models for decentralized training and inferencing, like those being explored by Akash Network — distributing AI workloads across thousands of idle devices instead of centralized data centers. “You don’t need new power for a laptop someone already owns,” he said. “This is how AI becomes sustainable.”
Clemens Koczur, with experience in the energy and commodities sector, noted that decentralization could relieve stress on existing power grids. “In Germany, new data centers are being blocked due to energy congestion,” he said. “Decentralized networks can redistribute that load and accelerate innovation.”
When asked what success would look like five or ten years from now, the answers were surprisingly human.
For Dewhirst, success means that “DePIN” as a term will disappear. “It’ll just be life,” he said — a world where people contribute data, compute, or energy without realizing they’re part of a decentralized network.
Levin looked further ahead, predicting the convergence of AI, DePIN, and quantum computing into a seamless global system — one that powers everything from healthcare to robotics.
Rydon and Koczur both echoed the same vision: greater ownership and understanding of the data individuals generate every day, and a deeper appreciation of the value they bring to the digital ecosystem.
