Grass made $17 million of revenue in all of 2025. It made $17 million again in the first six months of 2026.
GRASS has doubled in two weeks and that is not the reason being given.
The reason being given is a blog post.
The Accounts
Six months matched twelve, and it arrived without a press cycle.
Monthly operating costs run at $2 million to $3 million, so $24 million to $36 million across a year. Company projections put 2026 revenue at $65 million to $75 million.
Take the midpoints and that is a network earning around $70 million against around $30 million of costs. For a token at rank 112, an income statement that works is rarer than a narrative that sounds good.
The projection is the company’s own and the year has three months left in it.
What Actually Shipped
The two-week move traces to a repositioning document, not a release.
Grass has spent its life scraping public web pages through residential IP addresses and selling the results to AI labs as packaged snapshots. The new framing casts it as an AI real-time retrieval layer, built from a Search API for finding relevant pages, a Contents API that reaches through residential nodes to get past defences like Cloudflare, and multimodal retrieval covering images, PDFs and video.
That is a coherent shift in what the network is for. Wynd Labs chief executive Andrej laid the direction out without naming a single new customer, publishing a pricing model, or committing to a buyback. Nothing went live alongside it.
It Rose Through Its Own Unlock
An unlock of 17.13 million GRASS, around $7.24 million and roughly 2.53% of circulating supply, hit on 27 and 28 September, with daily releases continuing behind it.
The price doubled across that same window. Absorbing a supply increase of that size while rising is a strength signal, and the usual unlock warning does not apply the way it normally would.
The daily component is still running.
What Would Confirm It
A retrieval layer for AI agents is a far larger market than selling data snapshots, and the accounts say the smaller business already works.
Three things would settle it. A named client on the new APIs. A published pricing model. Or a second-half revenue figure that holds the pace.
The doubling is priced on the description. The evidence sits in the two half-years behind it.
