Will AI Agents Push Up Crypto Prices? The US$18,850 Answer
AI agents really are paying each other, millions of times a month, and a lot of it really does run on crypto rails. So why did every one of those coins fall in 2026? The answer is a single number, and once you see it you cannot unsee it.
Part 2 of a series. Part 1 explained agentic payments and who really decides what gets bought. This part answers the money question. You do not need to read Part 1 first — everything here stands on its own.
First, the words — all of them, in plain English
If you already know these, skip ahead. Nobody will mind. But this whole article turns on the difference between a few terms people use loosely, so let us be precise.
An AI agent does not just answer you — it does things for you. It can search, compare, book and buy.
A blockchain is a shared record book that many computers keep at the same time, so no single company owns it. Ethereum and Solana are two of them. To use one you pay a tiny fee, and that fee is the thing people are really betting on when they buy the coin.
A stablecoin is digital money designed to hold a steady value, usually about one US dollar. It is not a bet on a coin going up. It is a dollar that software can move. USDT and USDC are the big two.
A layer-2 is a faster, cheaper network built on top of a blockchain. It bundles up transactions and posts them down to the main chain for safekeeping. Base is a layer-2 built on Ethereum, and it is owned by Coinbase.
Gas is the small fee you pay in a chain's own coin to move anything on it. An agent holding US dollars on Ethereum still needs a little ETH to move them.
One last thing before we start, because it is the hinge of the whole argument.
A blockchain earning fees and a coin going up are not the same event. A road can be jammed with traffic and still make its owner nothing, if the toll is set near zero. Keep that picture in your head. We are about to watch it happen.
The argument, and why it breaks
Here is the argument you will hear. AI agents need to pay each other constantly and in tiny amounts — a few cents for a search, a data lookup, a translation. Card systems cannot handle thirty-cent payments economically. Stablecoins can. Therefore blockchains get busy. Therefore their coins go up.
The first three steps of that are correct. The last step does not follow, and here is the number that shows why.
The calculation nobody has done. Solana processed 75.4 million agent payment transactions in a recent 30-day window. Solana charges about US$0.00025 per transaction.
75,400,000 × $0.00025 = about US$18,850 for the month.
Call it US$226,000 a year — against a coin worth tens of billions. As a driver of the price, that is a rounding error on a rounding error.
And it gets starker. x402, the main protocol agents use to pay each other, charges no fee at all. It is free by design. Roughly 165 million transactions have moved about US$50 million in total — about 30 cents each.
This is not a flaw. It is the whole point. Cheap fees are the product. Every chain is competing to be cheaper than the others. A rail that competes on being cheap cannot also be a money mine. Those two things are the same lever pointed in opposite directions.
"But everyone says the agents run on Ethereum"
This is the version of the argument I get asked about most, so let me take it seriously, because the factual part of it is completely true.
Over 90% of agent stablecoin payments settle on a network called Base. Base is an Ethereum layer-2 — a faster, cheaper network built on top of Ethereum that bundles up its transactions and posts them down to Ethereum for safekeeping. So yes: when people say the agents are running on Ethereum's rails, they are right.
Ethereum's position genuinely is central. It hosts more than half of all stablecoins in existence, holds roughly 68% of all money locked in decentralised finance, and is the settlement layer big institutions have chosen for tokenising real-world assets.
So why has that not shown up in the price? Because of one number.
Since March 2024, Base has earned about US$98 million in fees from its users — and paid Ethereum about US$4.9 million.
That is roughly 5%. Base kept about US$94 million.
And who owns Base? Coinbase — a listed company. So the agent activity flows to Base, and Base's profit flows to Coinbase's shareholders. Ethereum gets a nickel on the dollar.
Think of it as a landlord and a tenant. Ethereum owns the land. Base built the shop on it. The shop is busy — genuinely, roaringly busy with exactly the agent traffic everyone is excited about — and the rent is about five percent of the takings.
Now here is the part that makes it worse rather than better: the landlord keeps voting to cut the rent.
In March 2024 Ethereum made it far cheaper for layer-2 networks to post their data down. It worked — layer-2 fees collapsed, which was the whole point. But Ethereum's own monthly income fell by 60% to 80%. Daily network fees went from around US$40 million in early 2025 to a low near US$10 million in 2026. Standard Chartered estimated that Base alone removed roughly US$50 billion from Ethereum's market value by pulling activity off the main chain.
There is a knock-on effect most people miss. Ethereum destroys — "burns" — a portion of the fees it collects, which shrinks the supply of ETH. Fewer fees means less burning. ETH supply grew 0.18% in 2025 and has been slightly inflationary through much of 2026. The famous "ultrasound money" argument, that ETH gets scarcer as the network gets busier, is not currently operating.
And now the decisive point, which ties back to the test I set out earlier.
Ethereum is deliberately engineering the scarcity away
I said usage only becomes value when space on the chain becomes scarce. So is Ethereum's space getting scarce? The opposite — and on purpose.
The Fusaka upgrade of 3 December 2025 began expanding capacity from 6 data slots per block towards 128. It is being raised in steps: to 10 on 9 December 2025, to 14 on 7 January 2026, with developers aiming for 48 by mid-2026.
Capacity is being expanded faster than demand is growing. That is not an accident or an oversight — it is the roadmap. The entire goal is to make layer-2 networks cheaper. Cheaper is the product.
Which means Ethereum is actively working to ensure that the one mechanism that could turn agent activity into a higher ETH price never switches on.
To be fair to the other side: as agent traffic grows, those data fees could become a meaningfully larger share of what Ethereum burns — some estimates put it at 30% to 50% of total burn. And Ethereum's grip on stablecoins, on decentralised finance, and on institutional tokenisation is real and durable. If you want to own ETH for those reasons, they are respectable reasons.
But that is a bet on Ethereum being important. It is not the same as a bet on agent payments making ETH expensive. Those are two different claims, and the second one currently has a five-percent rent cheque behind it.
The strongest version of the argument against me
I should deal with the best case for the other side, because plenty of thoughtful people make it — and one person makes it better than anyone else.
Raoul Pal, a former hedge fund manager who now runs Real Vision and who a great many Malaysian investors follow, argues we are entering what he calls the Exponential Age, arriving at an "economic singularity" somewhere between 2030 and 2032. He defines it as "the moment when the economic models we have used for the last two hundred and fifty years simply stop working."
His argument about agents is the sharpest part of it. He does not see AI as replacing workers. He sees it as "adding billions of new economic participants, AI agents and robots, each one producing output, consuming energy, transacting, paying for services, generating data." And on payments specifically: "Money is about to start moving at silicon speed."
I think he is right about all of that, and my own research says so independently. Agents genuinely cannot open bank accounts. Card rails genuinely cannot carry thirty-cent payments. Machine-to-machine settlement is genuinely real and compounding while almost nobody watches. That is the same conclusion I reached earlier — I simply arrived by counting transactions rather than by modelling the economy.
So where do we disagree?
Almost nowhere — and that is the genuinely interesting part. Because here is what he also said:
Read both halves carefully, because most people quote the second one and skip the first.
He is separating importance from price. That is precisely the distinction I drew a moment ago about Ethereum: being essential to the plumbing and being a good thing to own are two different claims, and the first does not deliver the second.
Now look closely at what "owning the operating system" actually describes. That is an ownership idea. It is an equity idea. You cannot own an operating system with a gas token — a gas token is a toll on a road, and we have just spent several pages watching every road operator race to push that toll towards zero deliberately, as published policy.
So who owns the operating system in this story? Coinbase, which owns Base and takes roughly half the economics of USDC. Circle, earning interest on a float measured in hundreds of billions. Visa and Mastercard, selling identity and trust. Stripe and OpenAI, owning the checkout itself. Every one of those is a company. With shares. With earnings.
Follow his logic all the way to the end and it lands on equity, not tokens. Which is exactly where the money went in 2026, while the tokens fell 29% to 40%.
I do not think he is wrong about the world. I think his audience stops reading one sentence too early.
2026 was the perfect experiment, and it already gave us the answer
You do not have to take my reasoning on faith, because this year ran the test for us.
In 2026 the agentic payments story went completely right. The protocols shipped. Visa and Mastercard joined an alliance. Stablecoins settled US$7.2 trillion in February 2026 alone — passing America's ACH bank transfer network for the first time in history. Stablecoin supply hit a record US$315 billion.
And the coins fell anyway.
| What was tested | What happened | Verdict |
|---|---|---|
| The big chains in 2026 | Bitcoin −29%, Ethereum −37%, Solana −40%. Solana is about −77% from its January 2025 high | Every milestone was hit. The coins still fell |
| The pure "AI agent" coins | ai16z peaked at US$2.4bn; in August 2026 its founder declared the successor token dead and told holders to sell. VIRTUAL went from about US$5bn to roughly US$414m — around −92% | That trade already ran, and already failed |
| Stablecoin supply as a signal | Held near records through the downturn, then shrank for the first time in four years, losing US$15bn — the biggest drop since the Terra collapse | Supply has come loose from price. It is no longer a useful indicator |
| Solana's own economics | Network revenue down about 90% from its January 2025 peak; share of all network fees fell from 26.6% to 17.3% between Q1 and Q2 2026 | As 21Shares put it: scale is proven, value capture is not |
So where did the money go? To company shares and to interest on cash.
Circle earns 94% of its revenue from interest on the dollars backing USDC. If you hold USDC, you earn none of that — Circle's shareholders do. Coinbase takes roughly half of all USDC economics through its agreement with Circle, worth US$305 million in the first quarter of 2026 alone. Circle and Tether together sit on a float of around US$260 billion, which is, in plain terms, a very large bond portfolio.
The honest summary: agentic payments are real, useful and growing. But the value lands in equity — company shares — and in Treasury bill interest. It does not land in tokens, because the tokens are attached to rails that are deliberately almost free to use.
Two fair points on the other side, and one threat with a date on it
In fairness: an agent holding USDC on a chain still needs a little of that chain's own coin to move it — real, permanent demand, but tiny per agent, and the industry is actively building ways to remove even that.
One honest warning about every number in this section. The figures in agent payments disagree wildly depending on who is counting. Coinbase says over 90% of agent stablecoin payments settle on Base. A separate count says Solana took about 82% of genuine agent transactions in February 2026 once test traffic was stripped out. Both cannot be the headline.
They are measuring different things — different definitions, different windows, and one of them is published by the company that owns the network being praised. You will also see this protocol's yearly volume quoted anywhere between US$19 million and US$600 million.
When an industry's own numbers disagree by thirty times, that tells you something on its own: it is early enough that nobody has agreed what to count yet. Treat every figure here as a direction, not a measurement — and be suspicious of anyone who quotes one without saying whose it is.
But here is the threat almost nobody prices in. The banks are coming for the rail itself. JPMorgan, Citi and Bank of America are building a shared tokenised deposit network through The Clearing House, targeted for the first half of 2027. JPMorgan's own system already moves more than US$7 billion a day. Twenty-four of the fifty largest US banks are working on it.
If agent payments end up settling in tokenised bank deposits rather than stablecoins, crypto captures none of it. Programmable money does not require a public blockchain coin. It only requires money that software can move — and banks can build that themselves.
The one test that would change my mind
Usage only becomes value when space on the chain becomes scarce — when there is more demand than room, so people bid the fee up. That is the only mechanism that turns activity into a rising coin price. Right now it is nowhere close, and Ethereum is the clearest proof: it is expanding capacity from 6 data slots per block towards 128, on purpose, because cheaper layer-2s are the entire point of its roadmap. Watch network revenue and fee per transaction, not transaction counts. A count is meaningless when each one is worth thirty cents and costs a fortieth of a cent.
And on Bitcoin specifically: there is essentially no connection. Bitcoin is not a payment rail in any part of this story. Anyone selling agentic payments as a Bitcoin catalyst is reaching.
So what should you actually take from this?
Let me be clear about what I am not saying. I am not saying AI agents are a fantasy. They are real, they are paying each other now, and that will grow. Part 1 of this series makes the case that agentic commerce matters enormously — just not in the way most people are told.
What I am saying is narrower and, I think, more useful:
The traffic is real. The toll booth is not.
Every piece of this story points the same way. The busiest agent payment protocol charges nothing. Solana earns eighteen thousand dollars a month from seventy-five million transactions. Base keeps ninety-five cents of every dollar and hands Ethereum a nickel. Ethereum is widening its own road on purpose so that nickel gets smaller. And the banks are building a rival road that would pay crypto nothing at all.
Meanwhile the money is landing somewhere very ordinary and very visible: on the income statements of companies. Circle earns interest on the cash behind its stablecoins. Coinbase takes roughly half the economics of USDC and owns the busiest agent network outright. Visa and Mastercard sell identity checks. Stripe and OpenAI own the checkout.
None of those are coins. All of them are shares.
And that is not a prediction — it is what already happened. Through 2026 this industry hit every milestone it set for itself. Stablecoins settled more money in February than America's main bank transfer network. The protocols shipped. Visa and Mastercard joined an alliance. And Bitcoin fell 29%, Ethereum 37%, Solana 40%.
When a story goes completely right and the asset goes down, the story was never the thing driving the asset.
There is an honest version of the crypto case, and it is worth stating fairly: Ethereum really does hold more than half of all stablecoins and most of decentralised finance, and it really is where large institutions are choosing to settle tokenised assets. If you want to own it because you think that position is durable and valuable, that is a serious argument.
But notice that it is a completely different argument. It is a bet on Ethereum being important. It is not a bet on AI agents making it expensive. People keep using the second to justify the first, and the numbers simply do not connect them.
Be careful with anyone who shows you a transaction count. Seventy-five million sounds enormous. It is worth eighteen thousand dollars.
Bottom Line
AI agents genuinely are paying each other on crypto rails — but those rails are free by design, and getting freer. Solana earned about US$18,850 from 75.4 million agent transactions; the main agent payment protocol charges nothing; and Base, where over 90% of agent payments settle, passes Ethereum only about 5% of what it earns. The value is landing in company shares and interest income instead, which is exactly what 2026 showed when the story went right and the coins fell 29% to 40%. Watch network revenue and fee per transaction, never transaction counts — and watch whether space on any chain ever becomes genuinely scarce, because that is the only thing that would change this answer.
Will AI agents push up crypto prices?
Mostly no, on the evidence so far, and the reason is structural rather than a matter of timing. Solana processed 75.4 million agent payment transactions in a 30-day window and earned roughly US$18,850, because it charges about US$0.00025 per transaction. The x402 protocol agents use to pay each other charges no fee at all — it is free by design. The money instead reaches company shares and interest on the cash backing stablecoins. Through 2026 the agentic payments story went right while Bitcoin fell 29%, Ethereum 37% and Solana 40%.
Will AI agents boost the price of Ethereum, since they run on Ethereum's rails?
The factual claim is true but the money does not reach ETH. Over 90% of agent stablecoin payments settle on Base, which is an Ethereum layer-2. Since March 2024 Base earned about US$98 million in user fees and paid Ethereum about US$4.9 million — roughly 5%. Base is owned by Coinbase, a listed company, so the profit reaches shareholders rather than ETH holders. Ethereum's own daily fees fell from around US$40 million in early 2025 to near US$10 million in 2026, ETH supply has been slightly inflationary through much of 2026, and Ethereum is deliberately expanding data capacity from 6 slots per block towards 128 — which prevents the scarcity needed to lift the price.
Does any of this affect Bitcoin?
Essentially not at all. Bitcoin is not a payment rail in any part of this story. Agent payments settle in stablecoins on fast, cheap networks such as Base and Solana. Anyone presenting agentic payments as a Bitcoin catalyst is reaching.
What would change this conclusion?
Scarcity. Usage only becomes value when space on a blockchain runs short, so users bid the fee up — that is the only mechanism that turns activity into a rising coin price. Today every chain is competing to be cheaper, and Ethereum is expanding capacity from 6 data slots per block towards 128 on purpose. Watch network revenue and fee per transaction rather than transaction counts. If blockspace ever becomes genuinely scarce because of agent volume, this answer changes.
Which companies actually capture the value?
On the evidence so far: Circle, which earns interest on the cash backing USDC and takes 94% of its revenue that way; Coinbase, which owns Base and takes roughly half of all USDC economics; Visa and Mastercard, whose value-added services grew 28% and 22%; and the checkout owners, Stripe and OpenAI. Every one of those is a company with shares and earnings, not a token. This is educational information about where revenue is landing, not a recommendation to buy anything.
This article is for educational and informational purposes only and does not constitute financial advice. Past performance is not indicative of future results. All investments carry risk, including the possible loss of principal. Cryptocurrency and digital assets are highly volatile and may result in total loss. Consult a licensed financial adviser before making investment decisions. Adezeno is a licensed unit trust consultant with Eastspring Investments Berhad.