Technology•

Solana Foundation Says AI Agents Now Drive 76% of OpenRouter Token Use and Builds Payment Rails to Serve Them

Solana Foundation AI lead Rishin Sharma told the Breakpoint conference that agents generate 76% of OpenRouter token use, and unveiled payment channels, private inference and a paid API server.

Written By :MINRK
Solana Foundation Says AI Agents Now Drive 76% of OpenRouter Token Use and Builds Payment Rails to Serve Them

Photo by Franck V. on Unsplash

Mentioned in This Article

SOL
SOL
Solana
$74.88
2.86%

1. Solana AI agents strategy takes the stage in London

The Solana Foundation used its Breakpoint conference in London in mid-November 2026 to argue that Solana (SOL) should become the payment layer for autonomous software. Rishin Sharma, who leads artificial intelligence at the foundation, said agents generated 76% of the tokens processed on OpenRouter in October 2026, up from 8% a year earlier, and presented new tools for Solana AI agents to pay for services.

Sharma summed up the foundation's position in a single line, telling the audience that "the biggest opportunity in crypto is already underway." The claim rests on a simple chain of reasoning: software that acts on its own is fast becoming the main consumer of online services, that software will need to pay for what it consumes, and a blockchain is well suited to carrying those payments.

To give the effort a visible home, he showcased Solana.ai, a new front page for the network's artificial intelligence ecosystem that went live in the week before the presentation. According to Sharma, more than 120 teams are building AI products on Solana, and their work falls into three broad areas: payments, trading and infrastructure.

2. How much AI usage now comes from agents?

Roughly three-quarters of it, on the measure Sharma chose. API keys classified as agentic were responsible for 76% of the tokens processed on OpenRouter in October 2026, compared with 8% twelve months before, a figure he attributed to OpenRouter's monthly recap. He described the shift under the banner that agentic use is the new majority.

Some background helps explain why that number carries weight. In this context a token is not a cryptocurrency but the unit of text that a language model reads and writes, and it is the basis on which model providers bill their customers. OpenRouter is a service that routes requests to many different AI models, so the mix of traffic passing through it offers a window into who, or what, is actually using those models. An agentic key is one tied to software that carries out multi-step tasks on its own rather than a person typing prompts one at a time.

Sharma paired that statistic with a broader one about the web itself. Citing Imperva's Bad Bot Report, he said bots and other automated software made up more than 53% of global web traffic in 2025, up from 51% in 2024. Taken together, the two data points support his central argument that people are no longer the majority of internet users in terms of raw activity, and that the gap is widening.

3. Falling prices, soaring consumption

Cost is the other half of the picture Sharma drew. The price of processing one million tokens has dropped from $37.50 with GPT-4 to $0.14 with GPT-5 nano, a decline of about 270-fold. Over the same span, monthly token consumption has climbed roughly 330-fold.

The relationship between those two figures is the point. When a resource becomes dramatically cheaper, it tends to be used in ways that were previously uneconomic, and long-running automated tasks that call a model thousands of times fit that description. An agent that would have been prohibitively expensive to run at the older price can operate continuously at the newer one, which helps explain how automated keys moved from a small minority of usage to a clear majority within a single year.

From there Sharma shifted the focus away from raw model quality. In his telling, how intelligent a model is matters less than what it is able to do. He framed the decisive questions as practical ones: which tools the model can use, what data it can reach, and whether it can handle money. The last of those is where the foundation sees its opening, because a capable agent that cannot pay for a data feed, an API call or a unit of computing power is limited in what it can accomplish without a human stepping in.

4. Why would AI agents need a blockchain to pay?

Agents need a way to pay for online services directly, in small amounts and at machine speed, and the foundation's case is that blockchain-based payments already do this at scale. Sharma pointed to x402, a standard that lets agents pay for online services, which he said has processed more than 400 million payments and $100 million in volume, starting from zero a year ago.

Those two totals also say something about the nature of the activity. Hundreds of millions of payments adding up to $100 million implies that the typical transaction is very small, which is the pattern one would expect from software paying per request rather than from people making purchases. Conventional payment systems were designed around human buyers, with accounts, cards and sign-up flows; a piece of software requesting a single piece of data has little use for any of that, and a fee structure built for larger purchases fits poorly with payments of this size.

Sharma also highlighted a specific project to show that the activity is concentrated on his network. BlockRun, which is built on Solana and which he described as the largest x402 server, handled more than 60 million transactions in two months. The standard itself is being developed by the x402 Foundation, which is backed by Coinbase and the Linux Foundation.

5. Payment channels, private inference and Pay.sh

Three launches formed the core of the foundation's announcement. The first is a set of payment channels that the foundation claims can handle 1 million payments per second. Payment channels are a familiar blockchain technique in which many transfers between parties are handled without each one being recorded individually on the main network, and the stated capacity is aimed squarely at the per-request payment pattern that agents produce. The throughput figure is the foundation's own claim.

The second is a private AI service offering what the foundation calls attested private inference on Solana. Inference is the act of running a model to produce an answer, and attestation refers to a verifiable assurance about how that computation was carried out. For agents that handle sensitive instructions or financial data, the ability to use a model privately while still being able to confirm the conditions under which it ran addresses an obvious concern.

The third is Pay.sh, a server that integrates more than 100 APIs that agents can pay to use. Rather than requiring a developer to register for each service separately and manage a collection of credentials, the idea is that an agent can reach a wide catalog of services through one place and settle the bill as it goes.

Sharma added that the foundation is working with Google, Alibaba and ByteDance on charging for AI models through agent payments. He did not present those efforts as finished products, but the names matter: all three are large technology companies that operate AI models of their own, and their involvement suggests that the model providers themselves are examining whether per-request payments from software could sit alongside traditional subscriptions and billing accounts.

6. Raoul Pal offers a more measured timeline

Not every speaker at Breakpoint described the agent economy as already arrived. Macro investor Raoul Pal, the founder of Real Vision, appeared earlier at the conference and argued that agents are not yet meaningful economic actors. "They're like kids," he said, a comparison that casts today's agents as participants who are present and active but not yet operating with real independence or resources.

Pal's longer-term outlook nonetheless points in the same direction as the foundation's. As more assets move onchain, he expects agents to trade and settle on blockchains. And if that happens, he said, "the value accrues to the underlying layer one," a reference to base networks such as Solana on which applications and assets are built.

The difference between the two views is mainly one of timing. Sharma's figures describe agents as the dominant users of AI models today, measured by the volume of tokens they consume. Pal's remark concerns their standing as economic participants, which depends on how much value they control and direct rather than on how many requests they send. Both positions can hold at once: software can account for most of the traffic while still handling only a modest share of the money.

7. Groundwork laid earlier in the year

The London announcements build on steps taken months ago. In May, Solana partnered with Google Cloud on a stablecoin payments service for AI agents. Stablecoins, which are tokens designed to hold a steady value against a currency such as the U.S. dollar, are the natural unit for this kind of commerce, since a service provider selling API access wants to be paid in something that will be worth the same amount an hour later.

Interest in the theme extends beyond the Solana ecosystem. Asset manager Franklin Templeton has described agentic AI as a killer use case for crypto, and the backing of the x402 Foundation by Coinbase and the Linux Foundation places the payment standard within reach of developers well outside any single blockchain community.

That breadth cuts both ways for Solana. A standard supported by several major organizations is more likely to be adopted widely, which enlarges the market the foundation is targeting. It also means the network cannot assume that agent payments will settle on Solana by default, which helps explain why the foundation is emphasizing throughput, privacy tooling and a ready-made catalog of paid APIs as reasons for developers to choose it.

8. What comes next for the foundation's AI effort

Two further releases are on the foundation's roadmap. It plans to publish benchmarks that test how AI models perform on financial tasks, and to launch an on-ramp that converts cash into stablecoins for AI assistants. No dates were given for either.

Each addresses a gap in the picture Sharma laid out. Benchmarks focused on financial tasks would give developers a way to compare models on the specific skill that matters for an agent entrusted with money, which fits his argument that practical ability counts for more than general intelligence. The on-ramp tackles a more basic obstacle: before an assistant can pay for anything in stablecoins, someone has to fund it, and for most people that starting point is ordinary cash.

The figures presented in London, including the 76% share of OpenRouter token use, the more than 400 million x402 payments and the 1 million payments per second claimed for the new channels, come from the foundation and the third-party sources it cited. How far they translate into lasting demand for the Solana network will depend on whether the more than 120 teams building there, and the large model providers now in discussions with the foundation, turn agent payments into a routine part of how AI services are bought and sold.

Related Articles

NEWSLETTERS

Don't miss another story.

Subscribe to the MINRK Newsletter today.

Tick the newsletters you want. No selection? You'll get MINRK Headlines.

By signing up, you will receive emails about MINRK products and you agree to our terms of use and privacy policy.

MINRK Daybook

Market analysis for crypto traders and investors.

EVERY WEEKDAY

MINRK for Advisors

Defining crypto, digital assets and the future of finance for financial advisors.

EVERY THURSDAY

MINRK Protocol

Exploring the tech behind crypto one block at a time.

WEEKLY

MINRK Long & Short

A must read for institutions. Insights, news and analysis delivered weekly.

EVERY WEDNESDAY

MINRK Headlines

The biggest crypto news and ideas of the day.

EVERY WEEKDAY

MINRK Policy

Examining the intersection of cryptocurrency and government.

WEEKLY

MINRK Research

Join thousands of readers who rely on MINRK for data-driven insights on the latest digital asset trends.

MONTHLY