Jordi Visser: The $900 Trillion Unlock

↓ Jump to the video and timestamps

Jordi Visser has been right about this AI moment longer than almost anyone I've talked to, and this conversation was one of the more clarifying ones I've had about where things are actually heading. Not where the mainstream narrative says they're heading, where the underlying data points.

The framing I keep coming back to: most people who think they're watching the AI adoption story are measuring the wrong thing entirely. They're counting ChatGPT downloads and model subscriptions. Meanwhile, agent-to-agent token usage has already gone exponential and far exceeds human-to-LLM usage by a wide margin.

That gap between what the public thinks is happening and what's actually happening is the arbitrage. And I think it's one of the most significant mispricings I've seen since the early Bitcoin days.

I came into this conversation as a real-world asset tokenization skeptic. I'm leaving it as a convert. Not because Jordi talked me into something fuzzy, but because the logic of agents needing liquid, programmable collateral clicked in a way it hadn't before. This post lays out that argument, along with where I think Bitcoin fits in the endgame.

Key takeaways

• Agent-to-agent token usage has already gone exponential, and the vast majority of people tracking "AI adoption" are measuring human-to-LLM usage and missing the real story entirely.
• Crypto infrastructure was always built for agents, not humans. The ghost city has users now. They just aren't human, and they don't care about Visa and Mastercard.
• Tokenization of real-world assets is the mechanism that converts illiquid equity, houses, private businesses, locked credit funds, into programmable collateral. Jordi's framing: roughly $120 trillion in physical hard currency supports roughly $900 trillion in assets. Tokenization changes what that ratio means.
• Stablecoins are winning the early agentic commerce phase. I'm not going to pretend otherwise. But Bitcoin is the only terminal reserve in this economy, the place capital parks when everything else hits its Schumpeter ceiling.
• The Bitcoin ecosystem is being stress-tested by AI-initiated attacks right now. The ColdCard entropy bug and AnchorWatch's code-inspection sprint are a preview of what's coming. The ecosystem will be more resilient in a year because of it.
• The people building the AI layer disproportionately came from Bitcoin. OpenRouter, a16z crypto, the mining industry. The psychographic overlap is a tell.

The arbitrage nobody is seeing

The Citrini post Jordi and I talked about this morning put it plainly: a guy's Delta flight gets canceled on the tarmac, he points an agent at it, and within minutes he has a $250 credit and a rebooking. All those pennies consumers have left on the ground for decades, they're about to get picked up by agents operating at a speed and thoroughness no human would bother with.

That's a real use case. Not a demo, not a whitepaper. It's happening now.

The bigger story is the agent-to-agent layer underneath the consumer-facing agents. The public's mental model is still ChatGPT. When people think about AI usage metrics, they think about token consumption between a human and an LLM. What they're not tracking is the volume of agent-to-agent token usage, which has gone exponential and already dwarfs the human-facing side.

That fact alone, in my read, highlights the mispricing. The people calling this a bubble are largely the same people not using the tools, and there's a high correlation between those two populations.

I tested Muse that morning before we sat down. Tried to use it to book Amtrak, ran into a wall because Amtrak is a walled-off bot and won't budge. But within ten minutes of playing with it I sent it to my wife: here's your calendar agent, here's the boys' sports schedule, just download it and use it. The consumer agent moment has arrived.

The night before, at the Freedom Tech Week party at PubKey, I ran my Hermes agent live. John Arnold and I from Ten31 had recorded a video that morning about oil markets. I told Hermes: pull the clip via the Fountain MCP, find the sharpest quote, write a one-sentence summary, post it to my X with subtitles. Done in under ten minutes.

That same task would have taken someone on our team two hours, find the clip, cut it, write the tweet, add subtitles. That efficiency gain is compressing across every industry right now.

Jordi's line on it: use it or get left behind. That's the plain simple thing.

Why Bitcoin is the last stop, not the first

I want to be honest about where I actually stand on this, because I think a lot of Bitcoiners are going to resist the framing and be worse off for it.

Stablecoins are winning the early agentic commerce phase. That's just true. The infrastructure is there, agents can use it, and it's already happening.

Stripe's acquisition of Bridge, valued by Stripe at roughly $1 billion, was a clear signal that stablecoin transaction rails for the agentic economy are not a future bet. They're a present one. Marty noted the premium paid on the deal, roughly $1 billion on $10 million in revenue, while Jordi's view was that Bridge had no revenues at the time and the company had been around for less than two years. Either way, Stripe going on to get involved in OpenRouter is the next step in that same stack. They're building for agentic commerce end to end.

But stablecoins are not the terminal destination. Jordi's Schumpeter frame is the right one here. In a world where AI allows anything to be instantly replicated and competition compresses to near-zero time, every growth story for every token eventually plateaus. You get your 10x to 1,000x window, and then competition kills the growth. Ford stock stopped going up for a reason.

In that world, when the growth stops and you need to park capital somewhere, where does it go? Bitcoin is the answer. Not because of tribalism but because of logic. It's the only asset where the terminal value argument holds. No competition can replicate the fixed supply, the security model, or the eleven-year history of that network.

Jordi put it this way: Bitcoin is the diversification layer in an agentic economy the same way real estate and bonds are the diversification layer in the current one. A decade from now, in his read, it's the only store of value left standing.

My own anchor example of this thesis in action: I can go to Strike right now, post Bitcoin as collateral, and have cash in my bank account within minutes. That's the $900 trillion thesis in miniature, liquid access to value that used to be locked up. AI agents need that same kind of frictionless, programmable collateral, and Bitcoin is the asset that doesn't decay, doesn't get competed away, and doesn't depend on any company staying solvent.

Tokenization and the $900 trillion, what that actually means

I'll be straight: I was a real-world asset tokenization bear for a while. It never clicked for me why you'd need to put assets on-chain when the existing settlement infrastructure, clunky as it is, mostly works. Then Jordi walked me through the house example and the logic landed.

You own a house. Your equity has doubled. You're trapped, you don't want to reset your mortgage rate, you can't easily sell a fraction of it, and borrowing against it through traditional channels means weeks of paperwork and a trip to seven different banks.

Now imagine that equity is tokenized. You can sell a slice to any of eight billion potential buyers, settle in minutes, and buy it back in six months when your cash flow recovers. Tokenization makes exactly that possible.

Jordi's framing from a paper he was releasing that week: roughly $120 trillion in physical hard currency currently supports roughly $900 trillion in total assets. That ratio is what makes the system fragile. Every time there's a liquidity crunch, central banks have to print because assets need to be reliquified from a tiny base of actual money.

Tokenization changes that equation by making the $900 trillion itself liquid. The collateral can move. Which means nobody has to panic-sell during a crunch because they can already access the value.

His Long Island beverage business example made the private company case. A friend owns a business with real estate worth $5 million but declining revenues and a license that kicks in a year out. Banks won't touch it because the cash flow is down.

With tokenization, he can sell a piece of the business token, bridge the gap for six months, then buy it back when the revenue comes. The whole thing, negotiation, documentation, settlement, handled by agents working through crypto rails. Without agents, humans couldn't manage the complexity. Without crypto, agents can't replace the function.

Vlad Tenev and Robinhood's move toward tokenization rails is the TradFi institution taking this seriously in public. The DTCC settling equities is already essentially a digital ledger doing what blockchain does, just a closed, permissioned one. When I thought about stocks through that lens, the argument for on-chain settlement of real-world assets stopped being abstract. The guardrails are just better.

The Dallas Fed has already flagged that tokenized deposits and AI agents could fundamentally reshape bank balance sheets. The institutions are paying attention even if most retail investors aren't.

AI agents are the missing piece, and crypto was built for them

Jordi has been using the ghost city image for months and it's the right one. Crypto built the infrastructure. Layer 2s, smart contracts, stablecoin rails, privacy-preserving payment channels, all of it sat there, waiting for users who never showed up in the numbers people expected. The conventional critique was that humans found wallets too hard, custody too confusing, and the UX too hostile.

That critique was correct. Humans were never going to use this stuff at scale.

Agents are different. They don't care about UX. They operate at the speed of code. They will negotiate fee structures, find the most efficient payment rail, and settle cross-border transactions in the time it takes a human to read a contract clause.

The Uber analogy Jordi used is right: Uber didn't work before smartphones. Not because the idea was wrong but because the enabling layer didn't exist. Agents are the enabling layer for crypto. The ghost city finally has occupants.

The enterprise adoption signal I've been watching is Marty's account of Latham & Watkins deciding to buy NVIDIA GPUs and self-host their AI infrastructure. Law firms don't make capex bets like that on hunches. They're uploading decades of institutional knowledge into a system they control, making sure their case history and research doesn't feed someone else's model.

Eli Lilly moved earlier, building out a private AI data center. The Fortune 500 layer is already deep into this.

But the more interesting signal is the consumer agent layer arriving now. Muse dropping, consumer agents gaining rapid adoption, enterprise adoption hasn't even fully played out yet and we're already entering the next phase. Enterprise agents replace human employees. Consumer agents run 24/7 doing the tasks humans find frustrating or simply never get around to.

Stack consumer demand on top of enterprise demand and compute needs don't plateau, they accelerate.

John Gray from Blackstone gave a talk at his event, available on YouTube, walking through five concrete ways AI was already growing margins across his portfolio companies. That's a capital allocator reporting ROIC. The productivity gains are here.

The agrarian-to-industrial analogy Jordi used for compute stuck with me. When economies urbanized, the question was how you feed eight billion people when everyone's leaving the farm. The answer was that innovation in agriculture kept pace with the population explosion.

Agent economies have the same dynamic: every new agent is a mouth to feed, and their food is compute. Anyone telling you we're going to be oversupplied in compute is making the same mistake as someone in 1880 predicting the cities would starve.

Hardening Bitcoin in the age of AI attacks

There was a BitDevs meetup here in Austin the night before this conversation, and the discussion that stuck with me was around what Rob Hamilton from AnchorWatch has been doing in response to the ColdCard entropy bug.

The short version: ColdCard, which many of us considered the gold standard in Bitcoin hardware wallets, had an entropy vulnerability being exploited by AI-initiated attacks. Hamilton and a group of developers asked the obvious follow-up question, if this can happen to ColdCard, who else is exposed? They built a harness to inspect the code of every Bitcoin project they could find a GitHub for and found hundreds of vulnerabilities across projects within days.

Responsible disclosure followed. Many of those vulnerabilities were patched before they could be exploited.

Bitcoin is the first target because it's a bearer asset. If you get it, it's gone. That reality is forcing faster, more rigorous security iteration in the Bitcoin ecosystem than anywhere else in software.

The pain of the current window is real. Some projects got hit. But the ecosystem will be more hardened in a year because of this work than it would have been after a decade of ordinary development cycles.

There's something almost poetic about it: agents are doing the code inspection work, filing the patches, and securing the infrastructure. The same technology creating the attack surface is also the best tool for closing it.

I've been running my own experiment here too. I gave my agent an eCash wallet, double-blinded signatures, fast settlement, genuine privacy, and just let it run. Sent it to LN Markets to trade, move funds in and out in eCash, convert back.

It's a small experiment but it illustrates what the velocity layer for agent transactions actually looks like. I pushed back on Jordi's Zcash focus here: eCash gives you comparable privacy with better speed for agent-scale transactions. The L2 privacy conversation matters, but I think eCash is more suited to the agent use case right now.

The people convergence, Bitcoin minds building AI

This is the thing I don't think gets enough attention: the people building the AI layer disproportionately came from Bitcoin.

OpenRouter, a16z crypto, the mining industry. I'm recording next with someone who came out of Foundry and is now deep in AI research. The personnel flow runs one direction, and it's from Bitcoin to AI infrastructure, not from traditional tech to either.

Jordi introduced me to Ali Yahya, one of the original partners in the a16z crypto group. He went to a16z because he was trying to convince the Google X team to invest in crypto and couldn't get it done internally.

He's the person at a16z who understood the agentic value of crypto rails before almost anyone in either world. The fund a16z raised for crypto during the depths of the bear market, Jordi estimates around $2.2 billion, was them betting on the eventual convergence while everyone else was writing obituaries for the space.

The psychographic overlap between early Bitcoiners and early AI adopters is not a coincidence. There's something about the Bitcoin psyche, the willingness to hold an unconventional thesis through ridicule, to operate in adversarial environments, to think in systems, that translates directly to early AI adoption. The people who spent years arguing that a fixed-supply digital bearer asset would eventually be worth something were already wired to see the agentic economy coming before the mainstream did.

I've been in Bitcoin mining for eight years. From 2018 to 2021, we did flare-gas mining on well pads in North Dakota, take the gas, scrub it, run it through a generator, mine Bitcoin. We were arguing then that Bitcoin miners were demand-response assets, synthetic grid participants that could absorb excess power and turn off when the grid needed it. Watching the energy conversation catch up to where we were six years ago is gratifying in the mildly annoying way that being early always is.

Jordi's claim is that the grid runs at roughly half its actual capacity because we can't dispatch it efficiently. Bitcoin miners figured out how to turn that into an advantage.

Data centers are starting to figure out the same thing. The Navier-Stokes fluid dynamics result Jordi mentioned, a math problem that had resisted solution for over a century, worked through by an AI system, has downstream implications for turbine efficiency, battery technology, and grid storage that compound into real capacity gains. The energy constraint people are extrapolating linearly is going to look different in three years than it does today.

About Jordi Visser

Jordi Visser is a longtime macro investor and the founder of Visser Labs, where he publishes research and video content focused on the intersection of AI, crypto, and global markets. He spent decades running macro portfolios at major institutional funds before pivoting full-time to research and education around the AI transition. He began covering AI markets in early 2024 and has been building out a crypto research layer on top of his existing AI subscriber base. His work focuses on how tokenization, agentic commerce, and the convergence of AI and crypto rails change the investment landscape over the next decade.

Sources mentioned

• AI Agents Can't Open Bank Accounts. Crypto Infrastructure Was Always Built for Them (TFTC): the infrastructure argument for why crypto was always built for agents, not humans
• Dallas Fed: Tokenized Deposits and AI Agents Could Drain $700B in Bank Duration (TFTC): institutional acknowledgment that tokenization and AI agents reshape bank balance sheets
• PwC Projects $31.6 Trillion in AI Data Center Capex Through 2050 (TFTC): the compute and power build-out scale
• ECB Admits Central Bank Money Must Go Onchain to Survive (TFTC): official institutions acknowledging the on-chain settlement direction

Watch the conversation

Timestamps

• 0:07 - Bitcoin wins in a world of devaluing fiat
• 4:19 - Agent wallets, Bitcoin rails, and the incumbent problem
• 8:45 - The bubble worry vs. the expected value case
• 12:04 - Enterprise adoption: Latham & Watkins, Eli Lilly, and the on-prem turn
• 23:43 - The order of operations: stablecoins, L2s, tokenization
• 29:24 - Stripe, Bridge, and the $1B signal
• 37:49 - The $900T opportunity: what tokenization actually means
• 51:06 - The Trump administration and why the regulatory window matters
• 57:41 - Compute bottlenecks, consumer agents, and what's next
• 1:05:57 - Energy, the grid, and why Bitcoin miners were right
• 1:10:26 - Ali Yahya, a16z, and the Bitcoin-to-AI personnel pipeline
• 1:15:25 - ColdCard, AnchorWatch, and hardening Bitcoin under AI attack

Sponsors

• Cash App: For a limited time, new customers can get $21 added to their balance. Just use code TFTC10 when you sign up, and send at least $5 to a friend in the first two weeks. cash.app/app/TFTC
• Square: For up to $200 off eligible Square hardware. square.com/go/tftc
• Bitkey: Use code TFTC10 for 10% off the new Bitkey. bitkey.world
• Aven: aven.com/bitcoin
• CrowdHealth: joincrowdhealth.com/tftc
• Simple Mining: simplemining.io/tftc
• Salt of the Earth: drinksote.com/tftc