Delphi · Oct 20, 2026 · 1h 16m · 69m saved
With Andrej Radonjic, Co-founder and CEO of Wynd Labs
A network built from a pandemic arbitrage bot now delivers exabyte volumes of web data to the frontier AI labs training the biggest models in the world. Most data vendors chase whatever volume they can scrape. Andrej Radonjic, co-founder and CEO of Wynd Labs, told Yan Liberman on The Delphi Podcast that Grass, the network his team built, only works because its users explicitly opted in.
"From the internet's perspective, it just looks like users of the internet accessing the internet as usual. But in reality, it's a massive data infrastructure platform feeding the frontier of AI. At this point, we're at exabyte scale."
Andrej started building what became Grass during an applied-math PhD, when an arbitrage bot he wrote needed to scrape prices across websites that were starting to wall off their data from data-center traffic.
The full interview is covered here so you can skip it. 1 hour 16 minutes of audio, 7 minutes of reading.
Here are the 5 takeaways that matter.
Key Takeaways
Grass has scaled to roughly 8 million nodes of people who opted in to share bandwidth for a reward, delivering exabyte volumes of data to frontier AI labs
AI "flipped" the internet's access model: sites built for humans now default to blocking any traffic that looks like a data-center IP
Grass's video index has grown past 100 billion videos, now used by robotics labs training models to learn behavior by watching the open web
Andrej won't name which labs Grass serves, comparing the secrecy to defense contracting — naming a capability would expose a lab's roadmap
Grass's intellectual property sits with an "ownerless" Grass Foundation, governed by independent directors with no stake in Wynd Labs
He defines a real moat as reinvested capital that out-produces competitors per dollar, which he says Grass's self-owned infrastructure gives it
1. From PhD Project to Exabytes
Andrej traced Grass's origins to the Covid pandemic, when he built an arbitrage bot during an applied-math PhD that required scraping prices across websites. He and his co-founders noticed e-commerce sites were already walling off their data from data-center IPs years before the ChatGPT boom, and bet the trend would spread across the whole internet as machine learning grew more powerful.
Grass pays users for bandwidth instead of installing proxy software without consent
"Grass is a network of millions of nodes that people opt into to share their bandwidth and get rewarded for it." — Andrej Radonjic
He cited the "smart TV residential proxy scandal" — quietly installing proxy software on devices without consent — as the cautionary example Grass built against.
Every node on the network has explicitly opted in
"A core tenet of Grass is consent. So the nodes on the Grass network are all consenting into the Grass network and are being rewarded for it as well." — Andrej Radonjic
That network has scaled to roughly 8 million nodes and, by Andrej's account, now delivers exabyte volumes of web data to frontier AI labs — a scale he said most people can't intuitively grasp.
