Perceptron Network is taking an unusual route to one of the AI industry's biggest costs: training data. In an interview published on July 3, 2026, co-founder and CEO Peter Anthony told crypto.news how the network gathers publicly available web data through everyday consumer devices instead of relying only on large centralized data suppliers.
Historical news event: 2026-07-03. Retrospective report prepared November 2026.
The idea is less about mining coins than about using something many people already have: an internet connection that isn't running at full capacity. Contributors install a Chrome browser extension or an Android app. Perceptron then routes requests through those devices to fetch selected public web pages and assembles the results into datasets for AI customers, according to the company.
"If you build the best model in the world, it's pretty useless if it doesn't have access to good quality data," Anthony said in the interview.
What an ordinary device contributes
One reason to spread collection across many homes is that the web looks different depending on where you are. A website or social feed can show different public results to someone in Africa than to someone in Europe. A network of consenting users in many countries can capture those differences in a way that would be costly to reproduce from one data center.
Anthony made a related point about scale. "One single server trying to monitor all these different users can't really gather meaningful intelligence at that scale," he said.
Perceptron said the network spanned more than 150 countries and about 800,000 nodes at the time. Those numbers came from the company and aren't an independent audit of daily active devices. For comparison, BlockMesh alone had reported more than 700,000 devices when it merged into Perceptron in June 2025.
Image: Perceptron Network
The company says nodes don't read private files or collect sensitive personal data from a contributor's device. Instead, the device is used to reach specified public pages. That's an important line, but any app with network access still deserves clear permissions, a readable privacy policy and outside security review. Bandwidth-sharing isn't new in crypto; we covered an early example when Mysterium's decentralized VPN reached Google Play in 2018.
From raw pages to usable data
A big pile of web pages isn't enough to train a good model. Useful data has to be cleaned, labeled, deduplicated and checked for relevance and rights. Perceptron said collected data is sent to a central server, where automated systems check it against benchmarks before contributors are credited. Anthony said the company had recently acquired a firm that makes transaction and payment verification software to help automate that validation.
He also described a planned "Data Questing" platform, where contributors would complete active tasks to create or enrich datasets, going beyond passive bandwidth sharing.
On the customer side, Anthony named Everlyn AI, which he said receives image datasets for text-to-video training, along with clients using sentiment analysis of public discussion on Twitter, YouTube and digital asset markets. He pointed to the price of data access as the opening: by his estimate, OpenAI pays about $60 million to $100 million a year to Reddit and Twitter for API access. That figure is Anthony's estimate, not a disclosed contract value.
Points now, tokens later
Node operators earn points for their connectivity, and those points are meant to convert into Perceptron's native token later. Anthony said tokens would be fed back into the ecosystem from company revenue, with some set aside for buybacks. Points aren't a guaranteed token payout, and leaving a device online isn't a fixed income.
There's also a harder commercial question about the law around collecting web data. Terms of service, copyright, personal information and regional data rules still apply when building datasets for paying customers, even when the source pages are public. Our earlier look at crypto projects involving AI shows how long the industry has promised to fix AI's data supply.
A fund for smaller AI teams
Perceptron also described a $10 million AI Data Fund for independent developers. Selected teams would get five weeks of dedicated help with data infrastructure and up to 5 TB of real-world data at no cost, the company said. Those are commitments as described by Perceptron. Which teams received support, and what was actually delivered, would need separate checking before being reported as outcomes.
In the longer term, Anthony said Perceptron wants to offer enterprise clients business intelligence and analytics on top of its data.
Why the interview matters
The July interview marks how the project now explains itself a year after the BlockMesh merger: as a data supplier and analytics business, with crypto used to track and reward contributions. Weeks later, the company announced $6.5 million in strategic funding, and it isn't alone in pairing AI with crypto incentives, as Animoca Brands' fund for apps built with Minds AI agents shows.
The model's credibility will rest on paying customers, measurable dataset quality and real protections for contributors, not on the size of a node count.
Frequently asked questions
How does Perceptron use idle bandwidth?
Contributors install Perceptron's Chrome extension or Android app, and the network routes requests through their devices to fetch public web pages. The results are checked and packaged into datasets for AI customers.
How many nodes does Perceptron have?
Perceptron said in July 2026 that it had about 800,000 nodes across more than 150 countries. Those figures were supplied by the company.
Do Perceptron contributors earn tokens?
Contributors earn points for sharing bandwidth, which the company plans to convert into its native token later. No fixed token amount or payout schedule was given in the interview.
What is Perceptron's AI Data Fund?
It's a $10 million program announced by Perceptron for independent AI developers. Selected teams get five weeks of data infrastructure support and up to 5 TB of data at no cost, according to the company.
Sources
- Crypto.news, July 3, 2026: Perceptron is turning idle bandwidth into AI training data
- Perceptron Network official site
- Perceptron Network extension, Chrome Web Store
- Perceptron Network app, Google Play
- BlockMesh newsletter, June 19, 2025 merger post (archived)
Featured image: Perceptron Network
Frequently asked questions
How does Perceptron use idle bandwidth?
Contributors install Perceptron's Chrome extension or Android app, and the network routes requests through their devices to fetch public web pages. The results are checked and packaged into datasets for AI customers.
How many nodes does Perceptron have?
Perceptron said in July 2026 that it had about 800,000 nodes across more than 150 countries. Those figures were supplied by the company.
Do Perceptron contributors earn tokens?
Contributors earn points for sharing bandwidth, which the company plans to convert into its native token later. No fixed token amount or payout schedule was given in the interview.
What is Perceptron's AI Data Fund?
It's a $10 million program announced by Perceptron for independent AI developers. Selected teams get five weeks of data infrastructure support and up to 5 TB of data at no cost, according to the company.
Last updated NOV 12, 2026
