Grass Revenue and the Home Internet Behind AI Training Data

Intorduction

In September 2026, an outside review of Grass revenue put the first half of the year at about $14.53 million recorded and about $11.35 million in cash received, according to ChainCatcher. Two months earlier, Grass had put the same half at about $17 million.

Grass (GRASS) pays people to share their spare home internet so AI labs can collect public web data. In AI crypto, revenue that someone outside the company has checked is worth a close read. So is the gap between that check and the company’s own number.

This article covers how a Grass node earns, what the review proves and what it does not, how the figures line up, where the network’s traffic comes from, and the risks for people who share their connection.

What Grass Is and How a Node Earns

Grass is a Solana-based network that turns idle home bandwidth into a web data pipeline for AI. Its first airdrop went out on October 28, 2024, and nearly 1.5 million addresses claimed it within days, The Block reported. That first round handed out 100 million GRASS, 10% of the total supply.

The Request Path From AI Lab to Home Connection

A customer wants a public web page. Its request travels through a Grass router to a node, which is a browser extension, desktop app, or phone app on someone’s home connection. The node fetches the page and sends back an encrypted response.

Grass’s docs describe the node as the network’s gateway to the internet. It routes traffic through the user’s IP address, kept apart from the user’s own browsing. A node earns by the traffic it relays and by a reputation score built from uptime, completeness, consistency, timeliness, and availability. Where the node sits also matters, because demand varies by region.

Grass says nodes carry only public web data, never personal files, passwords, or browsing history. It also says a node uses idle bandwidth and steps back when the owner is active, in an August 2026 explainer.

Routers, Validators and Two Kinds of Traffic

Routers are the hubs between nodes and the validator. Per the docs, they meter and encrypt traffic and report bandwidth, latency, and node connectivity.

As of September 2026, router rewards follow delegated stake. A model tied to validated bandwidth is planned for after the network decentralizes. A router that loses connection to more than a quarter of its node pool can be slashed.

The validator is where provenance comes in. Grass offers two traffic types. With Partially Encrypted Traffic, the validator completes the connection to the website, so it can check data quality and trace where the data came from.

With Fully Encrypted Traffic, data stays encrypted end to end. That gives more privacy but fewer guarantees on quality and higher fees.

For an AI customer, the first mode is the one that matters. It is how Grass can say where a record came from. That design is set out in Grass’s own docs.

Four Entities Behind One Network

Grass is not one company. A November 2025 update lists four parts. The Grass Foundation is an ownerless entity, Grass OpCo runs network operations, Grass DataCo runs the data business, and Wynd Labs works as an outside service provider.

The revenue review covers Grass DataCo. That is the entity that supplies data to AI customers.

What the Revenue Review Covers

Two periods are covered, most of 2025 and the first half of 2026.

Figures from ChainCatcher, September 22, 2026.

The $4.52 million gap has two parts, CoinGabbar reports. About $2.75 million is invoices for work still in progress. About $1.78 million is payments due under 60-day terms. The reviewer, named as Regen Financial, says it examined every work order in both periods and matched them against cash receipts.

Coverage of the review cites Grass’s post on X as the source, not a published report document.

Recorded, Received and Audited Are Three Different Things

Recorded revenue is work Grass has booked. Cash received is money that has landed. The two numbers answer different questions, and in the first half of 2026 they sit further apart. About 78% of recorded revenue had arrived as cash, against about 92% for the 2025 period.

Some of that gap is timing. The newer period has more invoices still inside their 60-day window.

The third term matters most. Coverage describes the review as a due diligence procedure, not an audit. An audit gives a formal opinion on a company’s full financial statements. This review checked one set of records against another.

Anyone citing the Grass revenue figures should say which of the three they mean.

Grass’s Own Numbers Against the Review

Grass has published revenue figures twice before. Setting them against the review shows where they agree and where they do not.

Figures from Grass’s November 2025 and July 2026 updates, and from ChainCatcher.

2025 Roughly Matches

Grass’s July figure and the review both put 2025 near $17 million. The review starts on March 14, and Grass has described the first quarter of 2025 as bringing close to no revenue.

The November projection did not hold. Grass’s own July numbers put the second half of 2025 at about $14.3 million. With Q3 at about $4.3 million, Q4 came in near $10 million against the $12.8 million projected.

The First Half of 2026 Does Not

In July, Grass said the first half of 2026 brought about $17 million. The review puts recorded revenue for the same six months at about $14.53 million. That is a gap of roughly $2.5 million.

None of the sources explains it. Neither Grass nor the coverage of the review has said why the two numbers differ.

The Full-Year Projection

Grass projected $65 million to $75 million for 2026. With $14.53 million recorded in the first half, the second half would need about $50 million to $60 million. Grass pointed to about $15 million of business moved into Q3 as part of that path. The next checked figure will show how much of it landed.

Further reading. Akash Network (AKT): GPU Compute, BME Burn, Cosmos Exit puts a company revenue claim next to independent on-chain data, the same test applied here.

Where the Traffic Comes From

The network looks huge from the outside. In November 2025, Grass reported 8.5 million monthly active participants, up from 3 million at its first airdrop, with 38% of activity from mobile.

Actual work is far more concentrated. Grass’s Stage 2 rewards page says 150,000 users received about 90% of the network’s traffic. Set against the November 2025 participant figure, that is under 2% of users carrying nearly all of the load.

Reward design explains part of it. Network Points track real bandwidth, valued at $0.0049 per point, with 5x multipliers for the desktop app and 10x for the Android app over the browser extension. A second score, Uptime Points, weighs connection quality and IP location. Stable, well-placed connections earn more, so they end up doing most of the work.

That design also blunts a familiar problem. Networks that pay for presence attract fake presence; a pattern covered in Sybil Attacks in Decentralized AI Networks: What Works. Paying for relayed traffic makes idle sign-ups worth little.

Rewards Paid in USDC Instead of New Tokens

Stage 2 rewards are paid in USDC, not GRASS. Grass links the choice to regulatory uncertainty. Its July recap says the payout adds zero net new tokens to circulating supply.

For node operators, that changes what the reward means. Pay in a stablecoin does not move with a token. It also has to come from somewhere real, meaning revenue or reserves. Grass has not said where the USDC pool comes from or how large it is.

The shift mirrors a change elsewhere in DePIN. io.net (IO): Adoption, IDE Tokenomics, and Real Risks covers io.net’s move from fixed emissions to supplier payouts tied to revenue. Grass has gone a step further by taking its token out of Stage 2 node pay altogether.

Claims for Stage 2 run from July 22, 2026 to January 22, 2027. Unclaimed rewards stay with the project after that, per CoinGabbar.

Real Use Cases, Sorted by Maturity

Details from ChainCatcher and Grass’s November 2025 and July 2026 updates.

The training data business is the only one with checked revenue behind it. Grass says its customers include leading private AI firms and Fortune 100 companies, some from the Magnificent Seven, and that nearly every AI customer has come back. It has not named any of them.

Live Context Retrieval is the one to track. It aims to give AI models live access to current web data through the node network, instead of static datasets for training. That is a different customer with different timing.

The Risks Behind the Numbers

A Limited Review of a Report Nobody Can Read in Full

The review is the best evidence Grass has published, and it is still narrow. It checked work orders against cash. It gave no opinion on the full accounts. The public sees summaries in news coverage and a post on X, not the report itself, and the reviewer’s standing can’t be checked from public sources.

A company’s revenue claim is only as strong as the check behind it. This one is real but limited.

Company Figures Have Run Ahead of Checked Results

Twice now, Grass’s own numbers have come in above what followed. Q4 2025 landed near $10 million against a $12.8 million projection. The first half of 2026 came in at about $14.53 million recorded against the roughly $17 million Grass stated in July.

Neither gap is large on its own. Together they set a pattern worth tracking. They also sit in front of a $65 million to $75 million full-year projection that needs a much stronger second half.

A Small Core Carries the Network

A headline of millions of participants suggests a network that cannot be shut down or bent by a few players. The Stage 2 data says about 150,000 users carry about 90% of traffic.

That core is still large in absolute terms. But the real capacity, and the reward flow, sit with a small fraction of the edge. If those users leave, or if one region’s connections get blocked, capacity drops faster than the headline figure suggests.

Home Connections Carry Reputation and Security Exposure

Security researchers have studied networks that rent out home internet connections for data collection, and none of the reports below names Grass. Gen Digital found in September 2026 that residential proxy networks route AI scraping through home IP addresses. People who share their connection can see their IP reputation damaged when that traffic misbehaves.

Google has gone after the worst of this market. In January 2026, it disrupted the IPIDEA network, whose exit nodes more than 550 threat groups used in a single week. In July 2026, it cut the NetNut pool by millions of devices, and researchers found that none of more than 20 bandwidth-sharing apps they checked showed a consent prompt.

Grass presents itself as the opposite of that model, with opt-in sharing and paid rewards, per its own explainer. The exposure for the person at home is still similar. Sites see requests from their IP, and Cloudflare told iTnews that signals beyond the IP address separate human and bot traffic. A home address can be flagged even when the owner did nothing.

Unnamed Customers Hide Concentration

Grass has not named a single customer. If one or two labs make up most revenue, one contract ending would move the numbers sharply. The July note that about $15 million of business slid from one quarter into the next shows how lumpy these contracts already are.

Impersonation Targets Node Operators

A network with millions of wallet-holding users draws scams. Grass’s own August 2026 explainer warns users about phishing impersonators. The Stage 2 claim window, open until January 2027, gives them an obvious hook.

What This Means for Node Operators, AI Data Customers and DePIN Builders

Node Operators

Your IP address is the product. Know that sites will see traffic from it that you did not send. Install only from Grass’s official channels, and check any claim link against the official site before connecting a wallet. Stage 2 rewards arrive in USDC, so the value of that payout does not depend on the token.

AI Data Customers

Ask which traffic type your data came through. Partially Encrypted Traffic is what lets Grass trace a record’s origin, and that trail is the useful part for any compliance review. Ask too whether the provider can show collection rules for sites that restrict automated access.

DePIN Builders

Grass shows the value and the limits of an outside revenue check. Publish the full report, not a summary. Line new figures up against your old ones before others do it for you. Recorded revenue, cash, and audited accounts are separate claims, so label each one.

What to Watch

Second-half revenue. Grass projected $65 million to $75 million for 2026. A checked figure for the second half will show whether the business moved into Q3 arrived. No date for the next update has been set.

A full published report. The next review would carry more weight as a readable document with the reviewer’s scope and methods, rather than a post on X.

Live Context Retrieval. Grass said in July 2026 that first products would launch later that summer. A live product with named customers would be the first revenue line beyond bulk training data.

The Stage 2 claim deadline. Claims close on January 22, 2027. The share left unclaimed will say something about how many of the reported participants are still active.

Router rewards. The docs plan to move router pay from stake to validated bandwidth after decentralization. That change would tie more of the network’s rewards to work done.

Conclusion

The checked record is clear on one point. AI labs are paying Grass for web data, with about $32.11 million recorded since March 2025 and about $27.59 million received in cash. That is paying demand, checked by someone outside the company.

The record is less clear on the rest. The review is limited in scope and not published in full. Grass’s July figure for the first half of 2026 sits about $2.5 million above the checked number, and its 2025 projection ran ahead too. A small core of nodes carries most traffic, and customers stay unnamed.

Grass revenue is evidence of use, not a verdict. The next test is simple. A second-half number, checked and published in full, will show whether the network’s growth matches its own forecasts.

FAQs.

Subscribe below for no-hype AI crypto analysis before the market catches up.

Editorial & Disclaimer Note: Content on CryptoAIAnalysis is independently researched and written using publicly available documentation, technical resources, and observable network data. The aim is to explain AI-powered crypto and blockchain systems clearly, highlight real-world use cases, and discuss limitations alongside potential. This content is provided for informational and educational purposes only and does not constitute financial, investment, or legal advice. Cryptocurrency and AI-related investments involve risk, and readers should always conduct their own research before making decisions.