The Token Revolution: Ribbit Capital's $41 Trillion Thesis That Nobody Is Talking About

Eight months ago, in June 2025, Ribbit Capital distributed a 41-page letter to its limited partners. Not about a portfolio company going public. Not about a new fund. About tokens.

Not cryptocurrency tokens. Not NFTs. Tokens as the fundamental unit that will reorganize how money, identity, expertise, and memory move through the global economy. The letter argues that every business on the planet is becoming a token factory. That your most valuable digital asset will soon be the memory tokens you share with AI agents. That the distinction between data and money is dissolving. And that whoever controls the infrastructure for issuing, verifying, and routing these tokens will capture trillions of dollars in value over the next decade.

Since then, the thesis has been quietly validated across every vector. Robinhood launched a blockchain. Coinbase shipped agent wallets. Every major bank rushed into stablecoin issuance. AI coding startups crossed billions in revenue. Prediction markets went mainstream. The eight months since the letter shipped read like a scorecard of Ribbit's own predictions coming true.

When a firm managing $12 billion writes 41 pages about a thesis, the checks have already been written. The letter is the receipt.

Ribbit publishes these letters openly, and plenty of people have read them individually. But no one has connected the dots across all of them: cross-referencing the Token Revolution thesis against the identity framework from January 2024, stress-testing the claims against what has actually happened since, and tracing how the firm is already executing on every prediction it makes.

This is Part 1 of a 7-part series. This overview maps the complete thesis. Over the next six days, we go deep on each major argument, pulling from every relevant letter and cross-referencing against what is actually happening in the market right now.

The Firm Behind the Thesis

Before the argument, the track record. Ribbit Capital was founded in 2012 by Micky Malka, a Venezuelan entrepreneur who had already built and sold a fintech company before most VCs knew what fintech meant. The firm's entire identity is built on a single belief: software entrepreneurs will replace legacy financial institutions.

That belief has produced extraordinary returns. Ribbit was an early investor in Coinbase (now worth over $60 billion), Robinhood, Nubank (the largest digital bank outside of Asia), Affirm, Revolut, and Brex. The firm does not spray capital across hundreds of deals. It writes concentrated checks behind conviction, then publishes detailed thesis letters explaining why.

Previous letters predicted the insurance technology wave years before Lemonade went public. They identified China's fintech explosion when most Western VCs were still cautious. They flagged the crypto opportunity early enough to back Coinbase before it was a household name.

The June 2025 letter is their most ambitious yet. And unlike most VC thought pieces, it is not a prediction about what might happen. It is a description of what is already underway.

The Prequel: Ribbit's 2024 Identity Letter Laid the Foundation

The Token Revolution letter, published in June 2025, did not appear out of nowhere. A year and a half earlier, in January 2024, Ribbit published a separate 23-page identity letter that reads, in hindsight, like the intellectual scaffolding for everything that followed.

That earlier letter opened with an Alan Kay quote: "Context is worth 80 IQ points." It then laid out what Ribbit called the "Digital Identity Frankenstein":

A system so broken that 5% of the $60 trillion in US ACH volume is still estimated to be fraudulent. The financial services industry spends over $40 billion annually on compliance alone, and yet the fraud keeps growing.

The 2024 letter introduced a concept that would become central to the 2025 thesis: the "data backpack."

That data backpack is what the 2025 letter would rename "memory tokens."

The 2024 letter also made a prediction about economics that shapes the entire Token Revolution thesis:

Card issuers earn over $185 billion in annual revenue for doing the work of verifying and credentialing customers. Ribbit asked: what if identity issuers could capture similar economics? What if JP Morgan could issue a KYC credential to every customer and be paid by other financial services providers to verify that credential later? Identity as a profit center, not a cost center.

Perhaps most importantly, the 2024 letter drew a historical parallel that illuminates the current moment. In 1958, Bank of America launched the "Fresno Drop," mailing unsolicited credit cards to 60,000 customers. Within a year, 2 million cards circulated across California. The payments system worked. The identity system did not. Fraud spiraled, delinquencies hit 20%, and the executive who created the program resigned. But the infrastructure survived and eventually became Visa.

What the Letter Actually Says

The core argument unfolds across a series of interlocking theses, each building on the last. Here is the architecture, stripped of the prose and reduced to its structural logic:

Premise: Tokenization is the process of rendering the world for computers. When you hear "tokenizing data" or "tokenizing money" or "tokenizing identity," it means making what people know and do across human systems safely accessible to software, including increasingly autonomous software.

Key insight: Not all tokens are equal. Their value depends on two axes: scarcity (how rare or expensive the token is to produce) and secrecy (how proprietary it is to a specific person or company). Generic web data is abundant and public. Your medical history is scarce and deeply private. The delta between these two extremes is where fortunes will be built.

The Token Taxonomy: What Ribbit Sees That Others Miss

Most analysis of tokenization treats it as a single phenomenon. Ribbit breaks it into at least nine distinct token types, each with different supply dynamics, different buyers, and different competitive moats. This granularity is what separates the letter from every other "tokenization is coming" thinkpiece:

Identity tokens (who you are), memory tokens (what you tell AI about yourself), expert tokens (specialized knowledge refined through feedback loops), access tokens (keys to sensitive data), context tokens (behavioral patterns and revealed preferences), asset tokens (stablecoins, tokenized treasuries, programmable money), payment tokens, credential tokens, and attention tokens. Each has its own economics, its own competitive moat, and its own set of winners. The sections that follow examine the most consequential of these in detail.

What makes this taxonomy powerful is the reframe it forces on every major technology company. Google's $264 billion ad business is built on inferring context tokens from search queries. Meta's $164 billion ad business runs on identity and context tokens extracted from social behavior. But the hundreds of millions of people now using ChatGPT and Claude daily are handing over memory tokens directly, in plain language, voluntarily. The gap between inferred intent and stated intent is the gap between a $264 billion business and whatever comes next.

The Memory Race: Why Your ChatGPT History May Be Your Most Valuable Asset

The letter's most provocative section concerns memory tokens. Ribbit observes that ChatGPT launched its memory feature in early 2025 and that daily users are already sharing "highly personal information, about our health, our jobs, what we want to buy, and more, with sometimes staggering outcomes."

This is not abstract. Consider what a heavy ChatGPT user has shared over the past year: career goals, financial anxieties, relationship dynamics, health concerns, creative ambitions, parenting questions. No platform in history has collected this kind of voluntary, high-context, deeply personal data at this scale.

Ribbit then poses a thought experiment that stops you cold:

This is the logical endpoint of the "data backpack" concept from Ribbit's 2024 identity letter. What started as a portable credential wallet has become something far more valuable: the richest representation of a person that has ever existed in digital form.

The race for memory tokens is still open. ChatGPT has pole position with its user base, but Ribbit sees multiple entry points. Could Stripe, with its position in business identity, become a memory layer? Could Plaid, trusted with financial data, extend into broader personal context? Could identity wallets like ID.me or Persona offer portable memory that travels across applications? India's Aadhaar system, now processing over 2 billion authentications per month across 1.3 billion enrolled citizens, offers a template for what government-scale identity infrastructure looks like. The question is whether memory tokens follow a similar path or remain in private hands.

The letter explicitly predicts a "Login with ChatGPT" button. Whatever emerges, Ribbit's bet is clear: memory tokens will be the most fought-over resource of the next decade, and the firm is positioned across every potential winner.

Token Factories: Every Company Is Becoming One

The central metaphor of the letter is the token factory: any technology or organization that transforms tokens of one type into more valuable tokens of another type. This is borrowed from an NVIDIA concept (the letter credits Jensen Huang), but Ribbit applies it far beyond chips and compute.

A medical AI company like Abridge starts as a tokenizer, converting doctor-patient conversations into structured clinical notes. Through feedback loops (physicians correcting outputs, administrators adjusting billing codes), it refines those raw tokens into expert tokens. Over time, it becomes trusted to write to electronic health records, billing platforms, and insurance systems. What started as note-taking becomes the operating system for clinical data.

Ribbit sees this pattern repeating across every industry:

The letter introduces a term for the most successful of these: Vertical Token Systems. These are "BPO-dismantling, ERP-commoditizing platforms" that will train and orchestrate agents to deliver real business automation. Unlike traditional SaaS, they do not force companies to replace existing systems. Instead, they deploy agents that operate on top of whatever is already in place, "like a trusted deputy who takes on more and more responsibility until one day they are running the company."

The evidence that token factories are already scaling is hard to ignore. Coding, the first vertical to be fully tokenized, now consumes 50% of all AI inference tokens (up from just 11% in May 2025). AI coding startups have reached $3.75 billion in combined annualized revenue, with Cursor exceeding $1 billion and Anthropic's Claude Code approaching the same threshold. The time horizon for coding tasks AI can complete autonomously is doubling every four months, with Opus 4.5 now handling tasks that take nearly five hours. These are not demos. These are production token factories generating billions in revenue.

The ambition is explicit. The letter asks: what would PwC look like if rebuilt from scratch by an AI-native team? Not 370,000 employees in 149 countries, but "a tightly knit network of domain experts paired with fleets of agents doing 95% of the repetitive work."

The B2B opportunity may be even larger. Ribbit's 2024 identity letter devoted an entire section to this:

The average invoice costs over $10 to process. The reason for all of this friction is identity: businesses still transact on trust, handshakes, and manual credit checks. Token factories that solve B2B identity, companies like Middesk for KYB, Rutter for cross-platform data access, and Stedi for system integration, are building the infrastructure for tokenizing the world's largest commerce market.

KYA: Know Your Agent Will Be Bigger Than KYC

Perhaps the single most investable insight in the entire letter is Ribbit's argument about agent identity. The firm observes that over 50% of internet traffic is already non-human. As AI agents proliferate, the current identity infrastructure will break completely.

The problem is not just stopping bad bots. It is enabling good ones. Consider a medical agent managing your chronic condition. It needs to access health records, schedule appointments, refill prescriptions, and manage insurance claims. How do you authorize it? How do providers verify its permissions? How do you ensure it can refill a prescription but cannot unilaterally change your address?

Ribbit coins the term KYA (Know Your Agent) and argues it will become a larger market than KYC (Know Your Customer):

The letter identifies multiple companies positioned to win this market: Persona and ID.me for linking agents to human identities, Okta and CyberArk for enterprise agent access control, HRIS platforms like Rippling and Deel as potential systems of record for "agentic workers," and new entrants like Oasis Security focused specifically on machine-first identity.

The scale becomes clear when you consider that cybercrime is now estimated at $8 to $12 trillion annually, making it the third-largest economy in the world behind only the United States and China. And the problem is accelerating: AI-powered synthetic identities, deepfake voice cloning, and automated phishing are making current defenses obsolete. Sift reported an 800%+ year-over-year increase in account takeover attacks in fintech as early as Q2 2023.

The evidence of agents going mainstream is now overwhelming:

Every one of these agents needs identity. Every one needs to prove what it is authorized to do, who it represents, and what data it can access. Ribbit predicts every person will run tens or hundreds of AI agents, and every business will deploy agents per human employee. The number of identity verification events grows exponentially. This is not a gradual shift. It is a step function.

Stablecoins Were Just a Prototype

The letter devotes an entire chapter to stablecoins, and the argument is counterintuitive enough to be worth its own section.

Critics point out that most stablecoin volume is bots. Visa's 2024 report found only 10% of trailing 30-day volume was human-initiated. Ribbit's response: that is exactly the point. Stablecoins are the first dollars built "machine-first," dollars that software can natively hold, send, and transact without intermediaries.

Tether earned $13 billion in profit in 2024, roughly 100 times the profit per employee of JPMorgan or Blackstone. And it is expanding far beyond crypto. Tether purchased 26 tons of gold in Q3 2025 alone (more than Kazakhstan, Brazil, or Turkey), holds approximately 96,000 BTC, and even received a $1.2 billion offer to acquire Juventus Football Club. This is not a stablecoin company anymore. It is a sovereign-scale financial entity.

The issuance wave Ribbit predicted is now unmistakable. JPMorgan, Visa, BlackRock, Fidelity, PayPal, Klarna, Revolut, Stripe, and Cash App are all moving into stablecoin infrastructure. Over 85% of the market is still USDT and USDC, but new entrants are proliferating across wallets (Phantom, MetaMask), exchanges (Binance, Hyperliquid, Kraken), and DeFi protocols (Ethena, Sky, Ondo). Emerging market users are already treating crypto cards as primary accounts for cross-border payments and remittances.

Ribbit sees this as the first data point in a much larger trend: as agents need programmable, composable, globally accessible money, stablecoins are the natural starting point, not the destination.

The destination is a world where AI agents tokenize GPU capacity, negotiate supply contracts, manage corporate treasuries, and trade compute futures. All autonomously. All using asset tokens that encode permissions, terms, and compliance directly into the instrument. Coinbase is already offering $1 billion in crypto-backed loans powered by Morpho, a DeFi lending protocol. The infrastructure Ribbit described in theory is becoming operational at scale.

The letter describes a scenario where a procurement agent spots discounted GPU capacity, executes a tokenized prepayment contract, deposits the token to the company treasury where other agents can interact with it, and later a working capital agent lists it on a financing marketplace for liquidity. A process that today requires dozens of hours is fully automated by agents with programmable assets.

The Attention and Truth Deficit

The letter's final major thesis borrows from Charlie Munger: "Invert, always invert." If AI creates an abundance of intelligence, what becomes scarce?

Ribbit identifies two answers: attention and truth.

When any AI agent can generate compelling content 24 hours a day, human attention becomes the most contested resource on earth. When deepfakes, AI-generated misinformation, and synthetic media flood every channel, verified truth becomes infrastructure, not luxury.

The letter sees prediction markets, oracle networks, and even memecoins as primitive early versions of attention and truth markets. Since June 2025, the prediction has been validated at stunning speed.

Kalshi, a regulated prediction market platform, hit $6.3 billion in trading volume in December 2025 alone, with 83% driven by sports betting. Robinhood integrated Kalshi's prediction markets directly into its app and is now generating over $100 million in annualized fees from the product. Phantom, a crypto wallet, launched perpetual futures trading and has already generated $13 million in cumulative fees. The infrastructure for pricing attention and truth is not theoretical. It is generating nine-figure revenue.

The most provocative framing: memecoins as proto-attention markets. Strip away the speculation and what remains is a mechanism for attaching a price to cultural relevance. Ribbit is not endorsing memecoins. They are observing that the underlying behavior, people paying to signal what matters, is a real economic primitive that agents will eventually need in machine-readable form.

How Ribbit Is Already Positioned

This is the part most commentary misses. The letter is not speculative. It is a rationalization of bets Ribbit has already placed. Every thesis chapter maps directly to existing portfolio companies:

The timeline of Ribbit's public moves confirms the thesis was in development for years before the letter shipped:

The Open Questions (and Why Ribbit's Position Is Unique)

Every thesis this ambitious leaves open questions. But the answers may matter less than most people think, because of the way Ribbit has structured its bets.

Timing. Ribbit acknowledges that "AI agent adoption may be years away" for some use cases. Autonomous agents managing treasuries and negotiating contracts require levels of reliability that are still maturing. But Ribbit does not need to be right about the timeline. Their portfolio companies (Stripe, Coinbase, Plaid, Persona) generate revenue regardless of whether agents arrive in 2026 or 2030. The token thesis is an accelerant on businesses that already work.

Regulation. The letter spends less time on regulation than you might expect. Cross-border compliance varies wildly, and the "tokenize once, verify everywhere" vision assumes convergence that will take time. But this is precisely where Ribbit's portfolio depth becomes a moat. Having investments across Stripe, Coinbase, Persona, Revolut, and Nubank in dozens of jurisdictions means the firm has real-time visibility into regulatory shifts everywhere. No other investor has this breadth of signal.

Competition. Every major cloud provider is building agent infrastructure. Every identity company is adding machine identity features. But as the portfolio map above shows, no other firm on earth has a portfolio spanning every layer of the token stack simultaneously. The token thesis is not just an investment framework. It is a description of the network effects already forming between Ribbit's own portfolio companies.

Why This Matters if You Are Not a Ribbit LP

You do not need to be an accredited investor to think in tokens. The letter's frameworks are useful for anyone building, investing, or working in technology:

If you are a founder: Ask what tokens your product creates, refines, or consumes. The most defensible companies will be those that build continuously improving token loops, where better data creates better outputs, which attract more users, who generate more data. Ribbit calls these "trust flywheels."

If you are in crypto: The institutional thesis for tokenized assets is accelerating, not because VCs love blockchain ideology, but because agents need programmable money. But note the risks: BTC ETF flows have flipped negative, Bitcoin dropped from $130K to $85K in Q4 2025, digital asset treasury companies are trading at massive discounts to NAV, and gold (+43%) and silver (+128%) are outperforming Bitcoin (+12%) since April. The token economy is expanding even as crypto prices contract. That divergence is the thesis.

If you work in financial services: The letter predicts that "everyone will have access to a world-class personal financial advisor" through AI agents. The $100B+ that banks earn from customers keeping cash in low-interest accounts is directly threatened. If you work at a bank, this is the most important document your competitors are reading.

If you use ChatGPT or Claude daily: You are generating memory tokens right now. These may become among your most valuable digital assets. The question of who controls them, whether you can export them, whether they are portable across AI platforms, will become a defining consumer rights issue of the next five years.

The infrastructure supporting this transformation is scaling at historic rates. Hyperscaler capital expenditure reached $399 billion in 2025, up 68% year-over-year, with consensus estimates of $533 billion for 2026 and $614 billion for 2027. The bottleneck has shifted from compute to memory: DRAM inventory at suppliers crashed 80% in a single year, from 15 weeks to just 3 weeks. Memory and storage stocks led the entire S&P 500 in 2025, with SanDisk up 871%, Seagate up 308%, and Micron up 283%. The physical infrastructure for the token economy is being built at the pace of a wartime mobilization.

The letter ends with a nod to the transformer paper that started the current AI wave. Its title, Ribbit notes, was "Attention Is All You Need."

In the token economy, attention is just the beginning.

For more on the infrastructure powering the token economy, see our coverage of the five tokenization waves converging in February 2026 and our data-driven analysis of Ribbit's portfolio convergence. For a deeper look at how AI agents are reshaping software, explore our complete guide to AI agents in 2026.