Jordi Visser Is Buying Past Bitcoin: Why AI Agents Could Outrun AI Stocks in 2026

Can crypto actually outrun AI stocks? Jordi Visser thinks it can. And he's putting his own book behind the call.

The former Morgan Stanley executive, who later ran money at Weiss Multi-Strategy Advisors, has been expanding his crypto holdings past the big three: Bitcoin, Ether, and Solana. Not because of a halving. Not because of ETF inflows either.

Because of AI agents.

His argument goes like this. Software agents are about to do real economic work. They'll buy compute. They'll rent storage. They'll pay each other for datasets and API calls and inference time. And not one of them can walk into a bank branch and open a checking account.

Machines need wallets. That's the whole trade.

Visser spent two decades reading macro tape for institutions. He's not a guy who chases tickers on Crypto Twitter at 2am. So when someone with that résumé says he's rotating out of the safest crypto assets and into the riskier ones, it's worth ten minutes of your morning.

The numbers behind the thesis

Start with settlement speed, because that's where this argument either holds or falls apart.

Solana runs 400 millisecond slots. Median transaction fees sit under a cent. Ethereum mainnet will charge you more in gas than the asset you're swapping and take around twelve seconds to confirm it. The speed difference isn't theoretical. You feel it.

Now put an agent on that rail. Ten thousand micro-payments an hour, each one conditional on the last. That agent can't queue. It can't wait for a block to finalize while its decision loop idles. It needs sub-second settlement, every single time.

Very few chains clear that bar. That's not a knock on Ethereum. It's just arithmetic.

Then there's the float. Stablecoin supply has pushed well past $200 billion and keeps climbing. That's working capital for machines. An agent negotiating a contract for compute needs a unit of account that won't swing 8% while the invoice clears. Dollars on a fast chain solve that. A volatile token doesn't.

And the rails themselves are shipping. Payment standards built for software, not humans, are live on mainnet right now. Not testnets. Not whitepapers with a roadmap slide.

This isn't the 2021 playbook

Compare it to the last cycle and the difference is glaring.

In 2021 crypto sold retail speculation and 400% APY that lasted eleven days. In 2024 and 2025 the AI trade sold datacenter capex and a handful of hyperscalers printing cash. Both worked. Neither one needed the other.

But here's the thing about the AI trade nobody wants to say out loud. The revenue is lopsided. Nvidia sells the shovels. Almost everyone else is still figuring out how to charge for the gold.

Crypto isn't competing with Nvidia. It's selling the payment layer that all that compute eventually runs on. Smaller story. Much bigger surface area. Every agent that transacts becomes a customer.

Hot take one. Most AI tokens are vaporware, and Visser's thesis doesn't save them. Buying a coin with AI in the ticker because agents are coming is like registering a domain in 1998 and calling yourself a company. The trade isn't AI coins. The trade is the rails agents actually touch.

Hot take two. If you're benchmarking crypto against Nvidia's chart, you're already losing. Different asset, different job. Nvidia sells growth. Crypto rails sell settlement. Stop asking one to be the other.

And a third, less popular one. Expanding past Bitcoin and Ether means Visser isn't hedging. He's rotating into beta. That's a much sharper risk profile than the digital gold crowd wants to admit, and it only pays off if agent volume shows up on chain.

What traders are watching

According to people running this play at size, three signals matter more than any narrative.

Agent payment volume, counted in settled transactions rather than press releases. Priority fee revenue, which tells you whether blockspace is genuinely scarce or just memecoin loud. And validator economics once Firedancer finishes its full rollout across the network.

That last one matters more than it sounds. Firedancer is the independent validator client built by Jump. When it hits full share, Solana's throughput ceiling stops being a talking point and starts being a number you can measure. More headroom means cheaper machine payments. Cheaper machine payments mean more agents can afford to show up.

Priority fees are the honest scoreboard. If they climb steadily through the fourth quarter without a memecoin mania attached, that's agents paying. If they spike and collapse, it was retail wearing a robot costume.

Solana doesn't wait for permission. It also won't wait for you to get comfortable.

What's next

Put three things on the calendar.

Firedancer's full mainnet share, which is the single biggest throughput unlock on the roadmap. The Alpenglow consensus upgrade, targeting finality near 150 milliseconds, which would make the machine-payment case obvious to anyone still arguing about it. And stablecoin issuance, which is the closest thing to a leading indicator this thesis has.

Every new billion in stablecoin supply is working capital for software. Watch that line, not the price chart.

If both move together, agent payments are real. If stablecoins flatline while Solana fees spike, you're watching people trade memes and calling it infrastructure.

I tested this so you don't have to. The agent payments story isn't loud. It's quiet. And it's measurable in fee revenue, which is more than almost anything else in crypto can claim right now.

That's the only version of this trade I'm willing to put money behind.