The Multi-Billion Dollar Business of Idle Bandwidth: How Grass Turns "Unused Capacity" Into AI Infrastructure

TechFlow Summary: DePIN has been hyped for years, yet truly successful projects remain few and far between. Grass monetized your idle network bandwidth to earn $17 million last year, and raked in the exact same amount again in the first half of this year. Now it is pivoting to the AI inference search market. Multicoin suggests this could be a business several orders of magnitude larger than initially anticipated for pre-training data. This article utilizes the 19th-century historical anecdote of "shipping ice to India" to clearly and thoroughly explain the logic behind idle capacity arbitrage.

The Big Business of Idle Capacity

"It appears, then, that the sweating inhabitants of Charleston and New Orleans, Madras and Bombay and Calcutta, drink of this well." (Henry David Thoreau)

In the 1830s, India possessed abundant goods that New Englanders desired: luxuries such as indigo, silk, spices, saltpeter, and cotton textiles.

But this affection was one-sided. India had no demand for New England's main exports, which included dried fish, beaver pelts, and whale oil.

As a result, many ships loaded with Indian cargo bound for Boston were typically completely empty on their return journey to Calcutta. Captains had to fill their holds with ballast stones to prevent the ships from capsizing in the rough seas on the voyage back.

However, New England did have a highly sought-after commodity that India entirely lacked, and the supply was virtually infinite: ice.

Ice cut from frozen ponds around Boston, including the pond made famous by Thoreau, became one of New England's most important exports in the early 19th century. The breakthrough came from Frederic Tudor, who developed a packaging method that allowed large blocks of ice to arrive with minimal melting during transit to tropical regions like Charleston, New Orleans, and the Caribbean.

The journey to Calcutta was, of course, much longer. But Tudor packed the ice tightly and layered it with sawdust for insulation, hoping that the majority of it would survive the voyage.

It was worth a try. Even if half of the product melted away, Tudor estimated he could still make a profit because his transportation costs were almost negligible: captains were willing to ship the ice to India for nearly free, since doing so saved them the trouble of loading ballast stones into their holds.

Their logic was that Tudor's ice could serve as ballast.

This strategy worked! The ships didn't sink, less than half the ice melted en route, and the rest was sold at a massive markup to British officials who desperately needed ice to keep their Gin & Tonics cool.

Even if the ice sold for six times the price in New England, the business only profited because transportation was essentially free.

This might be the very first business built around existing idle capacity.

Examples after that have been few and far between.

Airbnb and Uber are the most famous, but outside of those, they are so rare that there isn't even a fixed term to describe this business model of utilizing idle resources in the economy.

It could perhaps be called overcapacity utilization, capacity arbitrage, or the sharing economy.

Or we could simply call it DePIN.

Happening in Real Time (As Chatbots Love to Say)

In my view, using magical internet currency to kickstart real-world infrastructure has always been the most original and interesting idea in the crypto space.

But it has never truly caught on. Projects like Helium and Hivemapper represented grand, hopeful ideas, yet have failed to gain significant traction thus far.

A handful of DePIN projects offering idle GPU compute power have also failed to make an impact, despite the severe shortage of GPUs.

Geodnet is one of the rare success stories in DePIN, but it serves a niche market.

Overall, the results have been too meager, causing the term "DePIN" to start feeling as dated as "NFT," "Play-to-Earn," or "Web3."

However, recently it has gained momentum again, with more people believing that the Grass protocol might finally be the DePIN champion we've been waiting for.

Grass, the protocol that converts your idle network bandwidth into AI infrastructure, may already be DePIN's best proof of concept. The Grass network generated $17 million in revenue in 2025, growing rapidly: it earned another $17 million in just the first half of 2026.

Not bad!

And people believe it is about to become significantly better, much better.

To date, Grass's way of making money has been routing web scraping requests from AI labs through residential IP addresses. It's a good business because: 1) websites frequently block these labs from scraping data directly; and 2) Grass can (almost) use our idle network bandwidth for free.

But now, Grass is launching a new service—real-time context retrieval—leveraging the data it has collected from millions of residential network connections (including mine).

Beyond scraping data for training tasks, Grass will allow chatbots like Claude and Grok to search the internet for content not contained in their training datasets when users ask questions requiring up-to-date information.

This could get expensive. For example, if you ask Claude which restaurants are currently open, it can't just go directly to Google like we do, because Google is meant only for human use.

Instead, large language models must pay via APIs to access search engines specifically built for machines, which can be thought of as a pay-per-use version of Google.

For Grass, this should represent a much larger opportunity. In a recent report announcing an investment in the GRASS token, Multicoin Capital wrote: "We believe the final total addressable market for reasoning-time search and retrieval could be several orders of magnitude larger than the market for pre-training web data."

To grasp how large "several orders of magnitude" is, look at the cost of accessing the web via APIs: Anthropic and OpenAI pay $5 per 1,000 web searches.

You can see how cash-burning this could be for them. ChatGPT alone receives billions of requests daily, and it is estimated that between 14% and 35% of those require web searches to obtain real-time information.

Multicoin estimates that for Grass, capturing even a "fraction of that request volume could support hundreds of millions of dollars in annual revenue."

Billions of dollars would make Grass the most successful real-world story in the crypto space to date.

If so, this would be a DePIN story about leveraging our idle capacity to connect to the internet.

Just like shipping ice to India back in the day.

Join the official Coincamps community:

X: https://x.com/coincamps

Telegram: https://t.me/coin_camps