Grass GRASSBehind $GRASS doubling in two weeks: from selling data packages to “AI proxy internet access,” an old DePIN project relies on narrative to reset its anchor and survive
Author: Gemini, Deep Tide TechFlow On September 28, Grass officially published a lengthy declaration titled Abundant Intelligence by Wynd Labs CEO Andrej. This long article did not reveal new client lists nor announce substantial buyback plans, but it provided a coherent logic explaining the recent price fluctuations of the GRASS token. Over the past two weeks, GRASS's price surged from around $0.35 to above $0.70, nearly doubling. Capital had already priced this in on the market, and this article served more like a "post-event manual": Grass is attempting to shift its narrative from a DePIN project selling training data to an AI real-time information layer with a larger valuation model. However, whether this new narrative can sustain the current market cap long-term still depends on resolving the multi-million-dollar gap between its real cash flow and token capture mechanism. Changing the data: from "data packet outsourcer" to "AI agent network" In the past year, the market's understanding of Grass's business had a clear ceiling: using residential IPs to crawl public web pages and package snapshots to sell to AI labs. Although this business generates cash flow, under the industry consensus that "quality training data is running out," the market often views it as a transitional business. The core appeal of Grass CEO's long article yesterday, as I understand it, is to break this ceiling expectation and create a new narrative. To understand Andrej's new approach, we first need to clarify a fatal shortcoming of current large models: they are outdated upon release. No matter how many trillions of tokens are fed during training, once weights are frozen and the model is released, its world knowledge stops at the training completion moment. When users ask about real-time news or compare the latest product prices in chat, the model must connect to the internet during the inference phase to fetch the latest public web pages. However, large models themselves "cannot access the internet." Most mainstream internet sites today deploy strict anti-crawling firewalls like Cloudflare and Datadome. When AI labs try to crawl web pages in real-time using data center servers, requests are mercilessly blocked due to data center IP characteristics. Cloudflare has grown into a nearly $100 billion market cap giant by "keeping machine crawlers out." This is precisely the commercial story Grass is trying to reconstruct. Grass's foundational asset was the global network of millions of ordinary users sharing real residential IPs from home. Previously, Grass used this network to batch crawl and clean web pages in the background, packaging offline corpora to sell to AI labs; now, Andrej plans to turn this network into a "real-time proxy internet access channel" for AI models. In this roadmap, the official plan outlines a toolchain enabling AI to fully access the real world: Search API (solving "where to find"): Grass has accumulated massive public web copies through long-term crawling, and this API acts like a dedicated search engine for AI, allowing models to horizontally index the most relevant target web pages when facing complex queries. Contents API (solving "how to view"): After finding target pages, models no longer directly request from data centers and hit firewalls but access through millions of real residential nodes behind Grass. To the target site's firewall, this looks like normal browsing from an ordinary US resident's home, bypassing Cloudflare and other blocks, fully rendering and returning the page content to the model. Multimodal retrieval (solving "how to understand complex content"): The real internet contains not only text but also images, PDF reports, and embedded videos. This module enables real-time called models to directly comprehend non-text information. Following this logic, some overseas crypto analysts' views on GRASS become easy to understand. If Cloudflare's business logic is "charging websites and keeping AI crawlers out," then Grass's business model is the opposite: "charging AI companies, using real residential networks as disguise channels to let AI in." Thus, Grass's capital narrative has undergone a complete re-anchor: no longer an "off-chain data outsourcer" trading at traditional DePIN hardware multiples discount, but packaging itself as an "AI real-time retrieval layer and proxy gateway" enjoying high multiples in the primary market like Exa and Parallel. This explains why, even without any substantial new client announcements, this logically coherent roadmap alone was enough to trigger speculative capital in the secondary market on this narrative shift. Is there real revenue? This version is more inclined to be a project with real revenue and logically self-consistent. From this principle, looking at GRASS's fundamentals, it is a growth-stage data company with tens of millions of dollars in real annual revenue. According to data disclosed in the official July token holders call, its cash generation has passed the early proof-of-concept stage: Full-year 2025 revenue is $17 million. The first half (H1) of 2026 revenue already reached $17 million, nearly 7 times growth compared to the same period last year. Official expectations for 2026 full-year revenue from "training data" alone will reach $65 million to $75 million. The business shows clear seasonality, with some large orders originally planned for H1 (about $15 million) postponed to Q3 due to AI labs' training cycles. The company is currently profitable on the books. By building its own computing and storage facilities in 2025, monthly cash expenses (Opex) have been compressed to $2 million to $3 million (mainly infrastructure), far below its revenue pace. Architecturally, Grass avoids the traditional Web3 trap of "equity companies taking profits while tokens get scraps." Its client contracts and business revenue flow directly into Grass DataCo and the foundation, while the underlying developer Wynd Labs exists only as an engineering and business service provider. In theory, this ensures all protocol revenue belongs to the network itself. Who pays for GRASS? The business is running, and the structure is clean, but "how profits flow back to tokens" remains Grass's biggest challenge. Under the current mechanism, Grass shows typical "growth-stage tech startup" logic, but this sharply conflicts with the secondary market's craving for liquidity—the "crypto logic." Extremely restrained buyback efforts For secondary market traders, real cash buyback and burn is the only hardcore value capture. But Grass's official stance is very reluctant here. Except for a small $350,000 open market buyback in November 2025, large-scale buybacks have basically stalled. The official logic is: reinvest most profits into computing, storage, and network expansion, with long-term returns far exceeding direct secondary market support. This heavy asset investment strengthens the business moat but causes the token to lose the most direct buying support in the short term. Node incentive reality game In the network's second phase (Stage 2), Grass made a key tokenomics change: stopping GRASS token inflation to subsidize nodes, switching to paying rewards from protocol-earned USDC real revenue. From a token inflation perspective, this cuts off the bottomless sell pressure from circulating supply, an absolute positive. But from the network's physical supply side, contradictions emerge. Some ordinary nodes providing broadband find they only earn a few cents to tens of cents in stablecoins over months and start shutting down. Grass's new narrative of "breaking through Cloudflare's blockade" relies on these millions of real residential nodes. If low returns cause the underlying network to shrink, its business moat will be physically eroded. Stakers' insufficient yield experience Although the July governance vote passed to allocate part of USDC income to GRASS stakers, the actual amount and share ratio remain insufficiently transparent in public details. When holders cannot intuitively feel the tens of millions of dollars in protocol revenue through annualized yield (APY) statements, the token price can only oscillate between earnings expectations and early chip unlocking panic (such as late October tail investor unlocks). Therefore, Grass used a new roadmap to complete the contextual shift from a "static data packet company" to an "AI dynamic information layer." In crypto market capital games, such narrative upgrades from established projects toward hot sectors are often the most effective short-term catalysts. But after short-term sentiment vents, what ultimately determines GRASS's long-term pricing is still the real business fundamentals. The market will finally test not how cutting-edge the API functions described in the long article are, but whether the business can deliver tens of millions in financial guidance over the next six months as promised. The core contradiction is when this strong cash flow entity will be willing to convert on-paper profits into real GRASS token buy orders in the secondary market. Until narrative and real cash flow fully align, the CEO's statement only gives capital a reason to re-enter. If clear token buybacks and profit distributions are not forthcoming, I believe this expectation-fueled doubling rally will ultimately be a short-lived capital game rather than a true asset value reappraisal.