Fourteen Trillion Dollars Moved Last Year and Nobody Thought About the Fee

Fee sponsorship looked like a checkout convenience until the buyer stopped being human. What changes when there is nobody in the loop to top up a gas balance.

Visa's network carried $14.2 trillion of payments in its 2025 financial year, across 257.5 billion transactions, and not one of those payers spent a second thinking about a transaction fee. On a blockchain, not thinking about the fee still counts as an engineering achievement.

Every consumer payment system on earth sends the infrastructure bill to the merchant: the servers, the processor, the fraud checks, the settlement. Blockchains reversed that, and the industry has spent years treating the reversal as a fact of nature rather than a design choice. Meta-transactions, account abstraction, relayers and paymasters all attack it, and they mostly work, but they live in contract code that must be written, audited and trusted. Concordium put the answer in the protocol in March 2026. The interesting part is not that it did, but what became possible once the buyer stopped being human.

The Prerequisite, Not the Price

A CCD transfer on Concordium costs about one euro cent, a figure fixed by governance rather than set by an auction. At that price the amount is not what stops anyone. The prerequisite is. A customer arrives holding a euro stablecoin and enough of it to pay, then cannot complete the purchase because the fee is owed in an asset they do not have. Nothing about their intent, balance or eligibility is in question.

That matters most when the payment is meant to be boring. Ten stablecoins across five currencies now run on Concordium as Protocol-Level Tokens, and a separate fee token undoes the point of holding one, because spending a stable asset then means holding a volatile one. For a business paying salaries or suppliers, that is an unwanted currency position plus a standing task to monitor and top up.

The barrier has always been survivable, though, because a person can be sent off to fix it.

Delegation Removes the Person Who Fixes It

An AI agent holds a balance in the currency it was funded with, an authorisation from whoever deployed it, and a task. It does not hold a reserve of some second asset for fees, and there is no reason to give it one. Doing so means running treasury operations for software, with a failure mode that surfaces mid-task: well funded in the currency that matters, stopped by an empty fee balance.

Picture an agent funded with 200 euros in a stablecoin and told to assemble a training set from paid sources. It buys forty datasets over x402 without incident, then stalls on the forty-first with 160 euros still in its account, because the sliver of gas token someone topped up at deployment has run out.

The premise of delegation is that you describe an outcome and stop touching the machinery. If the exception is logging in to top up your agent's gas, what you handed over was not the task, it was a maintenance job with extra steps.

The volume is arriving, and its shape matters more than its size. Transactions of a dollar or more now represent 95 per cent of the value moved through x402, up from 49 per cent in early 2025, so the average payment is growing rather than the count merely inflating. In a June 2026 report, Chainalysis counted more than 100 million x402 transactions on Base, up from near zero in Q3 2025, though it credits much of that raw growth to meme coin farming rather than agentic commerce. A human can tolerate a prerequisite. A delegated system cannot, because there is nobody inside the loop to satisfy it.

The Proof Is Already Running, Quietly

Concordium's Agent Registry has been live since May 2026, and registering an agent is free because the registry runs on Sponsored Transactions. Every registration is still a transaction that carries a fee, but Concordium pays it, so the developer never acquires CCD and in most cases never learns a fee existed. Agents registered elsewhere can carry a Verified by Concordium badge without moving chains. 8004scan counted more than 560,000 agents across the ERC-8004 registries as of 12 October 2026, so this is not a niche event.

The objection is that sponsorship shifts a cost onto a business that would rather not carry it. European interchange is capped by Regulation (EU) 2015/751 at 0.2 per cent for consumer debit and 0.3 per cent for consumer credit, and interchange is only the floor of what a merchant pays. On a fifty euro basket that floor is ten to fifteen cents, against a cent for the sponsored network fee. Somebody still holds the CCD, of course. Sponsorship does not remove it from the transaction, it moves that balance from thousands of buyers who mostly abandon the purchase to a smaller number of operators topping up an account sized to their throughput.

How Sponsored Transactions Work on Concordium

Concordium's governance committee approved decision GP-020 in March 2026, and Protocol 10 shipped it, separating the account that sends a transaction from the account that pays for it. What matters for the agentic case is where that rule lives. Because it sits in the protocol it covers every transaction type, no relayer contract sits in the path to audit or exploit, and the sponsor's commitment is bound to one specific payload, so it cannot be detached and reused.

When an AI agent pays for a resource over x402, it signs a stablecoin transfer naming the facilitator as sponsor. The facilitator checks it against the scheme's verification rules, adds its signature, pays the fee and submits. The agent settles in the currency it holds and never touches CCD.

Identity is what makes that workable at scale, because a sponsor co-signs for counterparties it will never meet, and on Concordium every account is created against a verified identity. That does not prevent misuse, since the sponsor's signature is the gate, but it provides recourse, which an application-layer relayer cannot.

What It Does Not Solve

Consent does not vanish under delegation, it relocates. The human authorises the agent and its limits once, up front, and everything then rests on how well those limits hold. The sponsor has to know what it is co-signing, and the risk is not being drained at a cent a time but substitution: co-signing a payload that pays a different recipient, or amount, than the one agreed. Sponsorship also concentrates the need for a fee balance onto fewer parties, and a facilitator that lets its balance run dry fails every agent routing through it rather than one payment.

The harder version is what happens when an agent has to commit real value and hold it until a counterparty delivers. Limits set once are only worth whatever enforces them, and enforcement is a different problem from payment. Protocol-Level Locks are being designed for that case, and it is the one that decides whether delegation is something people will actually trust with money.

Builders who want to run a sponsor service can start from the sponsored transactions documentation or Concordium's AI Agent technology stack.

Join the Concordium Community, follow us on X.