Author: Simon Taylor, founder of Fintech Brainfood
Compiled by: Jiahuan, ChainCatcher
Cloud computing and mobile Internet are no longer the main forces driving changes in financial services. Taking over are smart tokens (units through which AI models process and generate information) and value tokens (on-chain tokens that represent currencies or assets).
The same ideas that made fintech different a decade ago, such as mobile-first, cloud computing, and replacing bank branches with APIs, are now basic standards for all financial companies around the world. When these practices become popular, financial technology will change from the next innovation direction to a daily routine in the industry.
New financial business and investment directions are taking shape around three changes.
Decision-making is moving from spreadsheets to AI models, and models rely on tokens to run.
Records that originally existed only within a company now exist in the form of Tokens on a shared ledger.
The customer is also becoming an Agent with a wallet.
Money is Token, and Token is money.
This article will discuss how these changes will affect finance, which companies are already taking action, and why the first trillion-dollar financial company based on modern technology will be a company with AI and Token at its core.
This competition is for the world’s largest profit pool.
In 2025, the banking industry will generate more net profits than any other industry. The financial industry has the largest profit pool in the world and is the second largest industry in the world in terms of market capitalization. The Boston Consulting Group (BCG) estimates that fintech accounts for only about 4% of this. JPMorgan Chase is on the verge of becoming the first trillion-dollar bank. No financial company based on modern technology has come close to this scale.
This article will outline the future development path of financial technology.
Fintech dies by success
Fintech has slashed the costs incurred by traditional banks that relied on branches, paper documents and phone calls. Mobile Internet puts an extra computer in customers' pockets, and cloud computing provides businesses with faster and cheaper infrastructure on which they can build services.
This allowed financial firms to make money for the first time from a customer base that would otherwise have been difficult to turn a profit on. With a mobile-first model, Nubank already has 140 million customers. It costs $1 per month to serve a customer and generates an average revenue of $17.10 per customer, according to its second-quarter earnings report.
By comparison, large banks bring in about $350 per retail customer per year, but service costs are several times higher and profits come mainly from selling other financial products to existing customers, as well as lending, investing and other overall operations.
Revolut, Nubank, Stripe, Ramp, and Robinhood are still capable of compounding for years to come, and likely will. Fintech as a separate investment category may come to an end, but the winners may still grow into the largest companies in the category's history, driven by new tailwinds.
Fintech’s share of global venture capital equity deals will fall from 14.5% in 2021 to 12.3% in 2025. At the same time, the proportion of fintech financing deals going to AI companies has risen from about 10% to 70% to 80%, depending on the source. New venture capital funds are now prioritized for investment in AI.
Decision-making, recording and service channels
Every financial product consists of three parts: decision-making, recording, and ways to reach customers. In a 19th-century bank branch, the manager used his own judgment to price your loan, wrote it down in a paper ledger, and you found him through the branch. Decision-making is a bet on risk, records record the facts that have happened, and reaching customers depends on a building.
Since then, the way decisions have been made has evolved: first by individual judgment, then by humans using spreadsheets, and now by basic models in collaboration with humans.
Records shift from paper ledgers to databases and then to tokens.
The way financial services reach customers has also shifted from branches to mobile phones, and then to agents.
Smart Token is the basic unit used by AI to participate in decision-making. It is used to complete computational tasks, but the model output is not completely certain, and the same question may get different answers. Value Token is the basic unit for recording currency or asset value. 1 USD worth of USDC or OUSD, backed by reserves in USD or USD equivalents at a ratio of 1:1.
But in the past two stages, there have always been three problems.
Decision-making hasn’t changed much. We’ve added machine learning, rules engines, and credit scoring, but the decision rules are still created and tweaked by humans. Basic models and agents allow the software to undertake deeper decision-making work.
Records are not always consistent. Bank mainframes can keep very reliable records, but these records only exist within one institution and must be replicated and synchronized elsewhere. Stablecoins and tokenized real-world assets (RWA), also known as value tokens, allow all parties to share a programmable ledger. Everyone sees the same record.
Service channels still rely on human operations. Excellent interface design reduces the operational resistance to almost zero, but in the end people still need to press the button. Giving users budget charts didn’t make them better at saving. If the Agent is authorized to automatically transfer cash to the savings account, the user no longer needs to open the interface and perform manual operations.
In the next era, these three elements will change again.
Money is Token, Token is money
On August 19, Stripe agreed to acquire OpenRouter for a reported price of just over $7 billion. OpenRouter is a gateway that distributes AI call requests to different models. The model calls scheduled every day involve 10 trillion Tokens, covering more than 400 models. On the same day, Ramp launched Router.com, which routes each request to the lowest-cost model that meets user quality requirements. Two of the best-run financial technology companies determined in the same week that AI Token routing itself is a financial services business.
Martin Casado of a16z calls it “Tokens are the new USD.” I want to take it one step further: not all Tokens are USD. USD represents a certain amount, while there is uncertainty in the output produced by the model through smart tokens. What the routing platform does is calculate pricing per unit of AI in a determined dollar amount. This pricing happens millions of times per second.
This is a financial market, and it has just been built by two payment companies.
The smart economy is built on this cycle.
Financial giants’ products reflect this shift
Among financial companies that have reached larger scale, the fastest-growing ones understand this change. It can be seen from their recently released products: they all use AI models and Agent workflows at the decision-making level, promote asset tokenization at the record level, and open services to Agents and machines at the service channel level.
Decision-making (AI model and Agent workflow)
Records (Tokenization and Stablecoins)
Service channel (Agent becomes a customer)
Robinhood has pushed toward tokenization more aggressively than either Nubank or Ramp, with Nubank going further with its own underlying model. But looking at these companies together, common directions emerge.
Financial decisions are turning to smart tokens
Assessing customers' repayment ability has always been an important source of competitive advantage in the financial industry. Traditional financial institutions’ past advantages came from balance sheets large enough to absorb losses and from decades of accumulated data on the performance of consumer loans. But ultimately, these jobs still rely on spreadsheets, credit approval committees and, at most, machine learning.
Revolut leveraged cloud computing and mobile Internet to achieve initial scale expansion at a lower cost. More than 80 million customers then provide it with a set of data that no other company has. Therefore, it trained the basic model PRAGMA based on its own transaction history to replace the rules and scoring system in the financial technology era.
The result was a 130% improvement in high-risk loan identification and a 65% increase in fraud detection recall (i.e., the proportion of actual fraud that is detected by the system). The credit approval committee has never handed over a score like this. Nubank has done the same thing with nuFormer: directly reading the raw transaction data represented by Tokens, eliminating months of manual work in designing and building data features.
The same changes are making their way into decision-making processes within companies.
Traditional automation breaks down when every workflow is a little different. Customer access and review, credit approval and month-end closing still require manual processing of links that cannot be covered by automation. The Agent does not require the process to be exactly the same every time, it only needs to be able to understand the content and steps.
The Agent used by Allica, a British small and medium-sized enterprise lender, can directly read free-written emails from loan brokers, fill in missing materials, call the decision engine, and give credit approval results. In the early days of its launch, half of the cases have been independently approved by it, which takes only 12 minutes on average, while the same process often takes weeks in large banks.
Smart Tokens are becoming a competitive advantage.
Financial records are turning to value tokens
Bank core systems are changing the way they record asset values: from balances in traditional databases to value tokens. This change is particularly important when institutions need to agree on ownership of assets, transfer value, or make assets available to wider markets.
In the past, companies that made loans to consumers or businesses had to build costly and complex back-office operations to package the loans into securities and sell them. A loan corresponds to a paper contract and accounting records in the company's own system.
Tokenization makes every asset easier to transfer.
Take Figure, for example, which uses value tokens to record newly issued home equity loans. This type of loan is a loan issued with the net value of the house value minus the outstanding mortgage loan as collateral. The loans then enter the Figure Connect trading platform for purchase by capital markets investors. This infrastructure is also increasingly being made available to other lenders, including Figure’s competitors, or companies it acquires in the future.
In the second quarter of 2026, Figure Connect's quarterly transaction volume reached $2.8 billion, accounting for 65% of the total transaction volume on its consumer loan trading platform. Its adjusted net income growth was 95% and its adjusted EBITDA margin (earnings before interest, taxes, depreciation, and amortization) was 54.6%. Calculated by adding revenue growth and profit margins, the score is about 150, reflecting high levels of growth and profitability.
Shared ledgers also solve a problem that fintech has never been able to solve: there are currently no truly global fintech companies. Nubank is a giant in Brazil and strong in Mexico, but elsewhere it still resembles a startup. Revolut is strong in Europe, but ranks between fourth and seventh in most of its larger markets. Two obstacles have prevented us from ushering in a truly global, ultra-large financial company.
Every time you enter a market, you must obtain a local license, prepare capital, and equip a legal team. In Mexico alone, it needs to prepare about $100 million in regulatory capital before it has even started serving any customers.
Fund flows will be suspended. Fedwire doesn't run on Saturdays, CLS and T2 don't run all weekend, and U.S. stocks only trade six and a half hours a day. And an AI-driven business is running every hour.
Tokens are designed to operate around the world and around the clock. Therefore, a product can use the same settlement network without having to tap into 40 local banks. Visa can now settle with U.S. card issuers via USDC seven days a week, up from just five business days a week in the past. Robinhood’s equity tokens are also listed in more than 120 countries.
AI agents built for 24/7 trading can only be truly useful if markets can be open and settled around the clock.
Agent is becoming the new entrance to financial services
I don’t want to use your Agent, I want my Agent to use your product.
If AI can effectively complete financial tasks, it will become a participant in economic activities. Agents need a budget, identity, permissions, and a way to hold or transfer value on behalf of others. This has created a new class of financial customers: service channels originally designed for human users holding mobile phones are now being called directly by software that works for users.
Human users bring attention, so the consumer Internet can be monetized in the process of attracting users and promoting purchases. The Agent brings a clear task: help me obtain X, the budget is Y, and the constraint is Z. It launches a query, gets structured results, and makes a selection in milliseconds. Your brand will either enter the model’s contextual window and become a line of information it refers to when making its selections, or it won’t appear at all.
Robinhood is the first company to treat Agents as customers at scale. It allows Agents to directly call account functions through the MCP server of transactions and banking services, without having to guess how to operate the web page or App interface. 100,000 users have given accounts with restricted permissions to Agent.
In contrast, some airlines block similar visits as bot traffic and then wonder why customers think the site is down. Stripe's Link is filling in the payment link: providing the Agent with a one-time payment card that can only be used within the approved limit, so that it can complete the payment without holding the user's original payment voucher.
Agent allows a financial firm to exponentially increase the number of customers it can serve.
What’s next after fintech?
JPMorgan Chase will reach a trillion-dollar market capitalization first, relying on deposits, licenses, and a balance sheet strong enough to withstand shocks. The traditional moat is far from gone, but it is no longer the only source of advantage.
Today, every financial company can build advantages in three areas.
decision making. Risk decision-making shifts to its own basic model, and Agent is responsible for workflows that were difficult to automate in the past.
Record. Assets can be transferred outside institutions, and funds and markets can operate around the world around the clock.
Service channels. Agent becomes a client.
The valuation list of private companies is exciting because for the first time we can see how they can go from their current size to a trillion dollars.
Stripe. With a valuation of US$159 billion, it has the strongest developer lock-in effect in the financial technology industry. Its advantages come from technology dependence and higher migration costs after access, and it has just bought a platform for pricing AI Tokens.
Revolut. It is reportedly raising money at a $115 billion valuation, has more than 80 million customers, and the underlying model for its financial business is already operational. It continues to acquire licenses, most recently adding conditional approval in the United States, as well as licenses in Australia and the United Arab Emirates.
Ramp. Valuation of $44 billion, annualized revenue of over $1 billion, and free cash flow positive.
Coatue’s “Magnificent 8” list of companies includes two fintech companies. According to Coatue's own data, once a company crosses the $100 billion valuation threshold, it has about a 31% chance of growing to a trillion-dollar size.
Don’t overlook the new crop of AI-native financial companies, either. Hebbia says institutions using its products collectively manage $30 trillion in assets. Rogo’s valuation rose from $750 million to $2 billion in just 16 weeks. If one of them reaches the scale in finance that Cursor does in programming, it won't call itself a fintech company anymore.
The first wave of financial technology put banks in your pocket.
The next wave will delve deeper into the decision-making, recording and work processes that were previously hidden behind the screen.
The companies in the first wave can lead this shift, but the category they created, “fintech,” no longer describes what comes next.
Fintech as an independent investment category has come to an end because the ideas that originally underpinned it have become common practices throughout the industry.