Andrej Radonjic says the decentralized data network lacks enough compute as AI labs ask for more web data
Grass has a problem most startups would love to have. According to founder Andrej Radonjic, the company does not have enough compute to keep up with how much data AI customers now want.
Grass is a decentralized data network. The Grass Foundation operates it, and Wynd Labs, where Radonjic serves as CEO, handles most of the engineering.
The concept is simple once you strip away the jargon. Regular people opt in to share their home internet connection. Grass then routes web data collection through those residential IP addresses instead of through corporate data centers.
Many websites block or throttle traffic that obviously comes from data centers. Traffic from ordinary home connections tends to get through, which lets Grass gather public web data at a scale traditional scrapers struggle to match.
As of mid-2026, the network counts more than 8 million active nodes. It has delivered more than 1 billion GB of public web data in total.
At its peak, Grass reportedly scraped as much as 4 PB in a single day.
The buyers are reportedly some of the biggest names in AI. Research on the project lists OpenAI and Anthropic among the major labs receiving Grass data.
Saturated, and needing far more
In a recent podcast and an accompanying essay, Radonjic described a network that has hit its limits.
By his estimate, Grass needs roughly 30 to 50 times its current infrastructure to meet demand. The network has been described as “saturated,” meaning existing nodes and compute can’t absorb the requests coming in.
Two types of demand are driving the pressure. The first is high-quality training datasets, the raw material AI labs use to build and improve their models. The second is real-time data for inference, the stage where a finished AI model answers a user’s question and needs current information to do it well.
The money side
Grass reportedly brought in approximately $17 million in revenue in 2025. It then reported another $17 million in the first half of 2026 alone.
The project projects full-year 2026 revenue of between $65 million and $75 million, from training data alone.
Operating expenses have reportedly fallen to $2 to $3 million per month following a round of heavy infrastructure work. Grass shifted to its own data centers and added GPU processing capacity, which gives it more control over how data gets processed and filtered before delivery.
Grass aims to build tools for live context retrieval, letting AI systems pull real-time information from the web at the moment they need it.
The project also leans on ethics as a selling point. Grass emphasizes user consent and provenance tracking, which means recording where data came from.
Node count growth, progress on owned compute, and whether second-half revenue tracks toward the $65 million to $75 million projection will show whether Grass can turn its supply crunch into scale.
