Data, not compute, is the real AI bottleneck. Peter opens by arguing that while the market has spent the last few years obsessing over GPUs and compute networks like Aethir and Akash, the harder problem sits upstream — the high-quality training data AI models actually need is locked behind paywalls. OpenAI reportedly pays Reddit around $70 million a year, with similar eye-watering cheques going to X, and that pay-to-play economy effectively freezes out smaller AI startups. Research groups like Epoch AI project the stock of public text data will be fully exhausted somewhere between 2026 and 2032, and even Sam Altman now concedes data — not compute — is the binding constraint. Perceptron’s pitch is that a decentralised network can fix this by turning users’ idle bandwidth into a globally distributed vantage point on the live web, at roughly a 90% cost advantage to traditional centralised data providers.
A thousand eyes, one vision. Perceptron’s architecture combines Perceptron Nodes — a software client that sits quietly in the background of a user’s browser or Android device and lends out unused bandwidth — with Perceptron Agents embedded in Discord, Telegram and WeChat communities, plus a human-in-the-loop Questing app where contributors annotate datasets. The point isn’t to harvest anyone’s personal data; it’s to aggregate geographically diverse viewpoints of the public web. Peter walks through the use cases this unlocks: an e-commerce operator seeing how their products rank simultaneously in New Zealand, the UK and the US; a quant desk arbitraging cross-border discrepancies in gold, oil or crypto prices in real time; a crypto trader spotting a sentiment shift across thousands of Telegram groups before it shows up on price. Perceptron is already supplying data to Everlyn AI, a text-to-image and text-to-video platform that would have been priced out through traditional suppliers.
Freshness, sovereignty and a universal basic data income. Peter makes the case that data freshness is becoming the decisive edge for frontier models, because a ChatGPT or Claude answering questions about a fast-moving crypto market on four-month-old data is flying blind. He also makes a pointed argument about annotation bias — that when a narrow set of labellers with their own agendas decide what a dataset “means,” the models downstream inherit those opinions — and contends that decentralised annotation is the counter. In the hot-take round Peter calls himself a multi-chain opportunist who still holds Bitcoin as the anchor, argues we’re in a 2020-style bull market (not a 2022 bear), and reckons the real 10-year story of AI is that it will displace a lot of jobs but open up far more opportunity for anyone willing to pick up the tools now — pointing to Claude Code as a live example of a non-developer being able to ship working software in minutes. His sci-fi pick: Avatar — fittingly, recorded the day before a trip to Zhangjiajie, the real-world mountain range that inspired Pandora.
Brave New Coin
If you enjoyed the show please subscribe to the Crypto Conversation and give us a 5-star rating and a positive review in whatever podcast app you are using.
