Grass

Andrej Radonjic (CEO, MSc Math/Stats York University, BSc Engineering Physics McMaster) + Chris Nguyen (CTO), with Connor White and Chris Gordon named as additional co-founders in the project's own materials. Parent: Wynd Network / Wynd Labs — a centralized web-scraping infrastructure company Radonjic founded during the pandemic, then pivoted into Grass in 2023.

Grass is a data DePIN: idle residential bandwidth from user-installed extensions and apps is rented to scrape the public web at scale, then sold as structured datasets to AI labs. Self-positioned as "the data layer of AI." Operationally, it is the most commercially validated decentralized AI project in the May 2026 cohort by a wide margin — and the most disappointingly priced relative to its revenue ramp.

What Makes It Architecturally Distinct

Most "data DePINs" are residential proxy networks with a token bolted on. Grass added a verifiable provenance layer: the Sovereign Data Rollup on Solana (announced March 2024, deployed through 2025) batches scraping-event proofs through a ZK Processor and checkpoints them to Solana mainnet.

Every scrape produces a ledger entry — URL, source IP, timestamp, session key, dataset hash. The dataset bytes themselves go off-chain to Backblaze B2; the chain stores proofs, not data. The provenance pitch is the data-poisoning defense — a buyer of a multimodal training corpus gets cryptographic proof of when, where, and how each sample was scraped.

Four roles in the network:

1. Grass Nodes — user devices providing bandwidth (extension, desktop, Android)
2. Routers — geographically distributed, stake GRASS, relay traffic
3. Validators — issue scraping instructions, verify data
4. ZK Processor — generates provenance proofs

Two encryption modes: PET (partially encrypted, validator-inspectable) and FET (fully encrypted, client-private).

The branding says "Sovereign Rollup." The reality as of May 2026 is that validators remain operator-run by Wynd Labs. The decentralized-validator-committee transition was originally targeted end-2025 and has slipped past that without a public milestone date. The page should not assert validator decentralization as a current fact.

Scale (May 2026)

• ~3M active nodes (Backblaze case study, conservative active-relay floor)
• 8.5M monthly active users (Grass self-report — registered users, not necessarily relaying)
• 190+ countries
• ~100 Gbps current network bandwidth; Sion target >1 Tbps
• ~1 PB/day scraped, 2-3 PB/day uploaded (includes redundancy/checkpointing) to Backblaze B2
• Reddit's full 2024 corpus scraped in one week post-Sion (per ASXN)
• 15M articles/day collected at ~0.1% of capacity

Revenue — the Rare Positive DePIN Disclosure

This is the singular feature of Grass that separates it from token-only DePIN: Wynd Labs discloses revenue.

• Q4 2025 projected: ~$12.8M
• October–November 2025 actual: >$10M
• 2025 contracts with "two major AI organizations" (unnamed)
• Last 12 months: "seven-figure clients, including the largest AI labs" (unnamed)
• ~20 active commercial collaborations including nonprofits and academic institutions
• SERP API in trials with 3 major SEO firms + 1 leading AI lab
• VALID dataset partnership with Ontocord and LAION — 30M multimodal samples, open-source release

Token buybacks: ~$100K completed Oct-Nov 2025, ~$250K in progress. Small in absolute terms, but a working revenue-to-token feedback loop is rare enough in DePIN to deserve its own data point.

Customer concentration is the open question. "Largest AI labs" is asserted; zero are named. Whether revenue is durable enterprise sales across the OpenAI/Anthropic/xAI/Meta/Google axis or concentrated in 1–3 anchor accounts is the load-bearing unknown for the whole investment case. DePINone's central concern, and the right one.

Tokenomics

• TGE: October 28, 2024 on Solana, day-one listings on Binance, Coinbase, Bybit, OKX
• Total supply: 1B GRASS (fixed)
• Allocation: Investors 25.2% / Foundation+Ecosystem 22.8% / Contributors+Team 22% / Future Incentives 17% / Airdrop One 10% / Router Rewards 3%
• Insider concentration: 47.2% (Investors + Team)
• Vesting: Investors and team had a 1-year cliff to October 28, 2025
• October 2025 cliff unlock: ~181M tokens, ~58% jump in circulating supply; price approximately halved during the window
• Season 2 airdrop: 170M GRASS queued for H1 2026, gated on native dashboard wallet launch
• Staking: ~32% of circulating staked to routers, 7-day unstaking lock, 32–50% APY across sources

Market Data (May 22, 2026)

• Price: ~$0.29 (CoinGecko, May 19)
• ATH: $3.90 (Nov 7–8, 2024)
• ATL: $0.1663 (Feb 23, 2026)
• Market cap: $171M–$256M (snapshot variance — see note)
• Circulating supply: 587.14M (58.71% unlocked)
• FDV: ~$435M
• Drawdown from ATH: 88–92%

CoinMarketCap shows ~244M circulating / $106M market cap as of this research pass. CoinGecko, Tokenomist, and DePINport all agree on ~587M / $171–256M. CMC appears to use a March 2025 snapshot. Authoritative current is ~587M circulating; CMC's market cap is wrong for May 2026.

Funding

• Seed (Dec 2023): $3.5M led by Polychain Capital + Tribe Capital, with Bitscale, Big Brain, Advisors Anonymous, Typhon V, Mozaik
• Series A (Sept 2024): Undisclosed amount, HackVC lead, with Polychain, Delphi Digital, Lattice, Brevan Howard Digital. Insider valuation ~$1B per ChainCatcher
• Bridge (Oct 2025): $10M led by Polychain + Tribe (same as seed)
• Disclosed total: ~$14.5M, but true total is materially higher given the private Series A round size

Recent News (Major Milestones)

• May 2025 — Sion upgrade: Multimodal scraping (text, images, 4K video); ~100 TB/day → ~1 PB/day
• October 2025 — Cliff unlock: ~181M tokens released; price approximately halved
• October 2025 — $10M bridge: Polychain/Tribe redux
• October 2025 — Points overhaul: Stricter sybil/VPN penalties
• Oct–Nov 2025 — First public buyback (~$100K) + Q4 revenue disclosure
• H1 2026 — Season 2 airdrop (170M) queued, awaiting native account-abstraction wallet

There is no "Grass 2.0" branded release. The project ships features incrementally under independent product names: LCR (Live Context Retrieval API for real-time inference), Sion (multimodal upgrade), mobile, SERP API. The product surface is wider than the brand suggests.

Contrarian / Critical

Token-Fundamentals Disconnect

GRASS is down 88–92% from ATH despite a revenue ramp from near-zero to ~$12.8M/quarter. The market is pricing continued unlock dilution (Season 2's 170M + ongoing linear vesting through 2028) ahead of any forward enterprise-revenue multiple. This is the real test of the "data DePIN" thesis — whether disclosed revenue ever translates to token capture, or whether unlocks structurally outrun cash flow. If revenue grows at current pace and token continues to bleed, the conclusion is mechanical: insider distribution exceeds buyer demand.

Validator Centralization

Sovereign Rollup branding outpaces on-chain reality (see contradiction above). The provenance ledger does what it claims; the validator set issuing scrape instructions does not.

Proxyware Category Risk

The extension is closed-source. Permission scope is broad ("modify data on all visited websites"). Grass states it KYCs all buyers; the user has no direct visibility into what their IP relays moment-to-moment. The broader residential-proxy category has documented historical association with bulk social-account fraud, click-fraud rings, and PII scraping (Trend Micro research referenced in DL News). Grass denies direct involvement in any specific incident, but the category risk — your residential IP showing up in someone else's threat-intel report — is not waved away by KYC at the buyer side.

AppEsteem certification is consumer-software compliance, not Trail-of-Bits-grade code audit. Radonjic's "Apple cybersecurity review" claim (WuBlockchain, Sep 2024) is not independently verifiable; Apple does not publicly confirm vendor security reviews.

Competitive Moat Is Thin

Bright Data has 150M+ residential IPs with SOC 2, ISO 27001, CSA STAR certifications and a decade of enterprise customer base. Grass has 3M nodes with AppEsteem only. The customer-switching case has to rest on one of three legs:

1. Provenance / ZK ledger as compliance asset — interesting if AI-data regulation tightens; not yet a forcing function
2. Token subsidy lowering effective price — works as long as token holders absorb the dilution
3. Ethics — user opt-in vs. embedded proxyware — real, but procurement teams don't buy on ethics

No publicly disclosed Bright-Data-to-Grass switcher has been documented. The "we win on provenance" thesis is plausible but unproven at procurement scale.

Revenue Framing

$12.8M/Q4 is a startup metric, not a category-leader metric. Comparables: Reddit's Google data deal ($60M annual), Reuters' OpenAI training-data deal ($22M annual). Grass at current pace is a small line item next to the data-licensing comparables. Whether this is an early ramp or a structural ceiling depends on H1–H2 2026 enterprise sales — particularly whether a named anchor customer surfaces.

Positioning vs. the Landscape

• vs. Bright Data, Decodo, Smart Proxy — these are the centralized incumbents. Grass's structural advantages are (a) user-consented bandwidth via token incentive and (b) provenance ledger. Structural disadvantages are scale (3M vs. 150M+ nodes) and enterprise compliance certifications.
• vs. other DePIN data plays (Vana, Reppo) — Vana is data unions; Reppo is RLHF-quality prediction markets. Grass is the only one with infrastructure for real-time public-web scraping at PB/day.
• vs. Wave 1 GPU marketplaces (io-net, Render Network, Akash Network) — different layer entirely. Those are compute; Grass is data. They are stack-complementary, not substitutes.
• vs. AI training-data centralized buys (Reddit-Google, Reuters-OpenAI) — different sales motion. Grass sells web-scraped derivatives, not first-party publisher content. Most useful for buyers who need long-tail breadth (small sites, multimodal, geographic diversity) that no single publisher contract covers.