From Nevermined, AI payments are beginning to enter real business.

Byline: Attorney Shao Jiadian

Recently, the term “AI payments” has been getting more and more attention. The concept itself is still evolving quickly, though, and different people tend to see different aspects of it. Some are focused on user-experience upgrades like voice ordering and automated payments. Others are focused on integrating wallets for AI agents and giving them the ability to pay. But as some projects actually start running, the market is beginning to see that the core of AI payments is not only “payment” itself, but how AI services are priced, transacted, and settled.

Take Nevermined as an example. It isn’t a single-point payment tool; it’s a piece of infrastructure that enables AI services to be priced, charged for, and settled. Its capabilities include billing, access control, real-time settlement, and compatibility with agent protocols such as MCP, A2A, x402, AP2, and more. What it truly wants to solve isn’t “how money gets paid,” but how, once an AI does a job, it can immediately become a revenue item.

(The images above are screenshots from Nevermined’s official website.)

It sells not a wallet, but a billing system for AI services

Many payment products are more like cash registers—the focus is simply getting the money in. Nevermined is different. It’s more like installing a full billing system for AI services. When an AI service is called once, the system can decide whether to grant access this time, how much to charge, how to leave an audit trail after deducting fees, and what to do if disputes arise later.

This sounds very technical, but it isn’t hard to understand. Today, lots of AI products are getting things done, but relatively few can keep making money. The problem often isn’t the capability—it’s the pricing model. Traditional software can be priced by account, by year, or by package. But AI agents often don’t work that way. What looks like a simple task may already involve multiple model calls, multiple tools, and several rounds of external services. If you keep using traditional software pricing methods, the price is very likely to become distorted.

What Nevermined is catching is this step. It turns those repeated back-end calls that used to be hidden into business actions that can be billed separately. In the past, people sold “software usage rights.” Now they’re starting to sell “every unit of machine labor.”

Why this business is starting to hold up

AI payments are starting to have real-world meaning not because the concept is new, but because the way AI works has already been forcing pricing models to change. Nevermined offers different models—charging by usage, charging by outcome, charging by value, and so on. At its core, this is answering a very practical question: how should an AI service actually be sold?

Many AI products used to be hard to sell, not because nobody needed them, but because customers couldn’t figure out where their money was going. If it’s a monthly subscription, people worry they’re paying too much. If it’s priced per seat, it doesn’t match actual call volumes. Especially in agent scenarios, a single conversation behind the scenes may include dozens, even hundreds, of micro-operations. If the billing logic can’t track the actual work process of the AI, commercialization will always feel awkward.

Nevermined’s proposed solution is to convert calls, results, access, and other actions into billable events, and then put billing, authorization, and settlement into the same chain of execution. This way, what customers buy is no longer a vague “AI capability bundle,” but services they can see and calculate clearly, one by one. The real value of AI payments isn’t letting the machine make the payment—it’s that the work the machine does can finally be priced seriously.

It also has a very critical move: it hasn’t locked itself into purely encrypted payments. Public information shows that it supports bank cards, stablecoins, crypto assets, and real-time bank transfers at the same time. This choice is important because if you want to make AI payments into a business, you first need to make it easy for customers to plug in. The more open the payment rails, the more it looks like infrastructure; the higher the barriers, the more it looks like a toy for insiders.

It’s no longer just a concept project

To judge whether this kind of project is still stuck in the concept stage, you can’t just look at how it tells stories—you also need to look at what it actually puts on the product pages. Nevermined’s public information shows that it already has partnerships with platforms such as CrewAI, Olas, Naptha, Mother, Helicone, and more. Its goal is to provide payment and billing capabilities for next-generation agent transaction scenarios. On the official product page, it already lists capabilities like MCP tools, A2A services, x402 payments, as well as cards, stablecoins, bank transfers, etc., as product features—not merely leaving them in vision statements.

Even more noteworthy is that there is publicly available overlap information between it and the Olas line. Nevermined has mentioned that when Valory deploys payment and billing capabilities to Olas’s AI agent marketplace, it leverages Nevermined to reduce the deployment cycle from 6 weeks to 6 hours. Olas’s public page also mentions marketplace integration with Nevermined, enabling agents to pay and get paid, and supporting real-time, dynamic pricing in agent-to-agent transactions.

At minimum, this indicates one thing: Nevermined isn’t just living in the imagination of “future machines trading with each other.” It has already been put into actual scenarios and used there. Of course, this doesn’t mean it has already pulled off a large revenue myth. What public materials can confirm is that it has been productized, already integrated into an ecosystem, and already entered real usage scenarios. As for exactly how much money it has earned, there are no clear numbers disclosed externally at this time. That boundary needs to be kept clear.

The truly sensitive part isn’t technology—it’s legal identity

The most troublesome part of AI payments is often not whether the technology can be built, but what the platform is considered to be legally after it’s built. Many tech teams are most likely to overestimate “code neutrality,” but regulators typically don’t buy that story.

In its product introduction, Nevermined emphasizes traceability, auditability, real-time invoices, and tamper-proof logs. From a business perspective, these capabilities are obviously great because they increase trust. But from a regulatory perspective, problems also follow.

If a platform only provides billing logic, access control, and interface verification, while funds are still controlled by users themselves and payment authorization remains on the user or wallet side, then it’s more like a technology service provider. But once the platform starts collecting on behalf of others, splitting on behalf of others, performing unified settlement, or custodying funds, or deeply intervening in stablecoin transfers, its regulatory profile will change significantly. Regulators don’t usually care much what you call yourself. What they care about more is whether you are processing money for others.

This is also the biggest difference between AI agent payments and ordinary software services. Previously, software platforms mostly sold tools. Now, these kinds of platforms are beginning to run into issues close to the core of payments—“billing,” “authorization,” “settlement,” and “leaving a trail.” Step further forward, and you’ll also run into questions of responsibility: if an agent automatically calls the wrong service, or makes an incorrect decision within the scope of authorization, whose loss is it? The user, the platform, the service provider, or the model itself? The smoother AI payments are, the more clearly the boundary of backend responsibility has to be spelled out.

What it truly changes isn’t just the way payment is made—it’s the way transactions work

If you only view Nevermined as a new type of collection tool, it’s easy to underestimate it. What’s truly interesting is that it assumes a future by default: AI won’t just be used by people—AI will also be bought, called, and hired by another AI.

Once you enter that stage, the transaction structure changes. Previously, a company sold software to people. Now it might be one agent calling another agent. Previously, a purchase corresponded to a contract. Now it might be a chain of automatically triggered small transactions. Previously, payment required human confirmation. Now, payment may be completed automatically within the authorization scope by the system.

Behind all this, it isn’t just the cash-register layer that gets rewritten. It’s pricing methods, contract structures, risk-control logic, and even the perspective from which payment regulation views the situation. Nevermined is certainly not the end state yet, but at least it has put one trend on the table: many businesses in the future may not be made by people negotiating with people—they may be calculated, initiated, and settled by machines.

Understanding this matters far more than just staring at the four words “AI payments.” Because in the next stage, what’s really valuable may not be whose model chats better—but who can get the chain “machine does the work—machine gets charged—machine settles” running smoothly first. Whether AI payments are worth watching isn’t whether it will order someone a coffee. It’s whether it can truly turn machine labor into revenue units. Whoever gets this working first will be closer to the entry point of the next wave of AI commercialization.