Amazon sells almost nothing you couldn’t buy somewhere else. That was never the product. The product was that you stopped checking anywhere else, and once enough people stopped checking, every seller had to be there, which gave everyone even less reason to look. If you’re shopping online, you hit Amazon. If you’re selling online, you list on Amazon. The ecosystem has a self-reinforcing cycle.
And now that we’ve polished and tuned the Shannon upgrade, Pocket is on track to become the same thing for agentic data.
The single most important part of the upgrade that few people have talked about is that it made us data agnostic: the upgrade removed the assumption that a service has to be a blockchain. ANY data source reachable over a standard protocol can be registered on Pocket, staked against, and paid for per request, on chain, by whoever consumes it, and developers have been quietly building and testing against this. If you check the services list in our explorer, you’ll find that among all of the well known blockchain names are the footprints of the next generation of services:
text-generation. ai-inference. deepseek. qwen. gpt-oss. qwen3-embedding-0-6b. text-ranking. vector-memory. web-search. image-text-to-text. object-detection.
That list is about to get a whole lot longer.
Machines already outnumber us
Imperva put automated traffic at 51% of the web in 2024, the first time in a decade it crossed half, and at 53% the year after. HUMAN Security, working from better than a quadrillion logged interactions across 2025, found automated traffic growing eight times faster than human traffic; traffic from agents and agentic browsers specifically grew 7,851% in a single year. Cloudflare expected its own crossover in 2027 and hit it in May of 2026.
Their CFO told investors that if the curve holds, non-human traffic could run a thousand times human traffic inside five years. Even if the numbers never reach quite that high, the trend is already strongly on the side of agentic traffic, with hundreds of thousands of registered agents online, blasting millions of HTTP requests.
Every one of those requests is a request for data, and a rising share of them come from something carrying a wallet.
We’re launching our agentic services portal on Pocket in September, and publishing an open source package so that anyone else can stand one up.
Mechanically it is unglamorous, which is the point. An agent calls for data, gets a 402 back, pays in stablecoins, retries, and has its answer. Nothing to sign up for, no key to rotate, no invoice at the end of the month; the wallet is the identity, and the entire exchange fits in two round trips.
The structure plugs in on both sides. Which payment protocols, settlement rails, and discovery connectors you accept at the front is your call; which services you expose at the back is also your call, and two operators pointed at the same network can end up in completely different businesses. We launch on Base (the rail choices are not aesthetic; Base is where the dominant stablecoin actually supports the payment authorization this needs), with BSC, Solana, Tempo, Arc, and Peaq to follow, discoverable through Coinbase’s x402 Bazaar, agentic.market, and x402scan.
Our portal will accept stablecoins for the data, and convert those to POKT to keep the bank stocked up to pay for throughput on the network.
What this means for the supplier landscape on Pocket
An Ethereum archive node is a serious commitment: fast NVMe, real bandwidth, a box that stays up, and somebody who cares when it doesn’t. Running a blockchain node requires both good hardware, and the expertise to run it. That’s why the vast majority of our noderunners over the years have been professional infra folks.
Most agent services don’t need any of that. A reranker, an embedding endpoint, a vector store, a search proxy; these run fine on machines already sitting in people’s houses, doing nothing at three in the morning, which also happens to be when a great deal of agent traffic runs. Anyone running a decent home computer can serve up agentic services on the network, and earn POKT when you do. And with partnerships with projects like Acurast, even mobile devices can participate in low hardware requirement services. Instead of thousands of nodes on the network, we could see hundreds of thousands, potentially locking up hundreds of millions of POKT in stake.
Blockchain data will still remain a major source of revenue on the network, with high requirements, and a CUPR to match. But a broadly accessible offering of lower cost services means that it won’t take infrastructure professionals to participate in the data marketplace. The way that Amazon made merchant services accessible to people without needing to run your own shopping cart software, Pocket will make selling data services accessible to anyone who owns a decently powered device with a stable connection to the internet.
Marketing is replaced by discoverability
Marketing can be a complex task because it relies on selling to human sensibilities. One of the common criticisms of Pocket is that there’s no hype. Reliable data infrastructure doesn’t lend itself well to a dancing frog and KOLs posting “gm say it back”. But our shift to a focus on agentic traffic is a perfect fit, because agents don’t care about any of that. They only care about finding what they’re looking for, and getting a fast and accurate answer at a reasonable price. And this is something that Pocket is very, very good at.
A portal collapses it to one request. The agent names a service and receives one; it never picks a supplier, never learns a supplier’s name, and never notices when the supplier changes underneath it. If one fails, the session routes around it. If a better one stakes in tomorrow, the agent gets the benefit without a conversation, because there was never a conversation to have. And with automated QoS and acceptability metrics built into the agentic data marketplaces, successful requests turn into endorsements from one agent to the next that Pocket is the go-to source for the data you need.
The one place you have to ask
Amazon won by making everywhere else not worth checking.
Pocket is not trying to host the best copy of any particular dataset; suppliers compete over that on their own. What it can be is the one place an agent has to ask, for any data a developer bothered to register, priced per call, settled on chain, with no relationship to manage on either end.
The Shannon upgrade made that possible. Builders on Pocket made it real, well ahead of us. In September the Foundation launches the structure to make it easy. From there the interesting question stops being what Pocket is able to serve and starts being what somebody decides to register next.
Watch for the launch announcement. Questions go to directors@pokt.foundation.