Will Token Be the New Dollar for Stripe?

Stripe is incorporating stablecoins into the AI economic infrastructure, intertwining crypto settlements with AI Tokens, ushering in a new paradigm of value transfer in the machine age.

Written by: Will Awang

Stripe's largest acquisition to date, over $7 billion, is for a company that does not handle payments. OpenRouter is an AI model gateway that processes over a trillion tokens daily, serving more than 10 million developers and businesses, helping them route requests and optimize token usage across over 400 models from more than 80 suppliers.

Three months ago, it completed a Series B round, valued at $1.3 billion, with investors including Nvidia and Google, among other AI industry giants. At that time, OpenRouter's CEO Alex Atallah described his company as the 'Stripe of the AI world.' Three months later, it was directly acquired by Stripe.

Why now? And why is a payment company building the infrastructure for the AI economy?

For the first question, Stripe provided a rather aggressive answer in its shareholder letter: the singularity began on January 1st. The basis for this judgment is not news, but the ledger—revenue from AI companies and crypto business is growing at more than double the rate of Stripe's total revenue.

For the second question, essentially, building economic infrastructure for the internet today means building economic infrastructure for AI. Patrick Collison's statement in the official press release is even more direct: Tokens are the core currency for businesses building with AI, and the economic potential of the real world will depend on how well scarce computational resources are utilized.

This article aims to clarify Stripe's strategic focus on Tokens and the AI economy through recent public materials, the August shareholder letter, Stripe's official press release, and President Will Gaybrick's concurrent interview at a16z, indicating that this acquisition's purpose extends beyond payments.

Collison's statement is the starting point of this article and crucial for understanding the positioning of Tokens within Stripe: Tokens are not yet dollars, but the systems governing dollars are being transferred to tokens. However, the transfer is not complete, and the untransferred part is precisely the most valuable—not the transfer itself, but the pricing.

I. In the AI Era, Digital Flows Support Every Business

The most significant statement in the shareholder letter is that Stripe is rewriting its origin story for the first time in fifteen years.

Capital and intelligence are becoming the two digital flows underpinning every business. In the past, every developer needed a reliable way to manage their revenue pipeline, which is why Stripe was born; moving forward, every developer will also need a reliable way to manage their intelligence pipeline.

Stripe's original narrative was 'we make online payments simple,' and it has now transformed into 'in the AI era, we manage the financial and intelligence flows of businesses.' The revenue pipeline is where it has already established itself, while the intelligence pipeline is the entry point it has just acquired.

The question is, what qualifies intelligence as a 'flow'?

1.1 A Machine in the Middle of the Loop

The mainstream consumers of tokens are no longer humans.

OpenRouter's investors and board, AMP PBC, analyzed the one hundred trillion tokens flowing through OpenRouter, and the conclusion differs from most people's imagination: the median request is not a person asking a large model a question, but a machine in the middle of a loop. The proportion of tokens from inference models has increased from negligible to over half within a year, with the average prompt length quadrupling, and a significant proportion of requests terminating in a single tool call.

This changes the shape of consumption. An agent running overnight repeats the same type of calls thousands of times, occurring unattended and continuously. This is no longer a one-time purchase; it is a continuous consumption flow.

And Stripe, as a company, has always done one thing: charge on the flow.

It is the nature of 'flow' that brings tokens into its range, regardless of whether AI is popular or not.

1.2 Only Things with Exchange Rates Need Routing

Businesses use tokens in two types of scenarios. One type is on the cost side, using tokens to build things; the other type is on the product side, assessing whether a token-driven product is useful and which models to switch between. The demand for conversion almost entirely comes from the second type: the ability to switch between different models for the same task indicates that there is an exchange rate between tokens, and only things with exchange rates need routing.

Another thing happening simultaneously is that the total amount is still rising; the shareholder letter states that token consumption is compounding at 9% weekly, while OpenRouter claims that inference volume is increasing at least tenfold annually; however, on a unit task basis, businesses are starting to calculate each transaction. The shareholder letter clearly lists these questions: How much is this task worth? Which model should handle it? Who pays, when do they pay, and what is the time value of this delayed payment?

Collison's statement in the press release reflects this structure: the two companies together help businesses manage profitability in the AI era from both sides—maximizing revenue and effectiveness on one side while minimizing costs on the other.

Cost reduction and efficiency improvement are both goals for Stripe.

Only when the total amount is increasing does optimizing unit costs justify building an entire layer of infrastructure. Payments follow the same path: when transaction volumes are small, no one cares about failure rates; only when volumes increase do people specialize in optimizing those few percentage points.

But optimization does not equal cost-cutting. Gaybrick mentioned in the a16z interview that he has seen many companies treating the efficiency brought by agents as a means to compress cost structures, leading to layoffs and cuts in operational costs, while his judgment is that the best way to optimize cost structures is to grow more.

Capital is something that needs to be invested and returned, not saved. This is the most essential literal meaning of the word 'Capital.'

1.3 The First Time Crossing to the Other Side of the Ledger

Stripe spent over $7 billion to buy the gateway it had already been connecting with.

OpenRouter was already using Stripe for payments, and Stripe's Token Billing had already integrated with OpenRouter, automatically syncing the prices of OpenAI, Anthropic, and Google models, with businesses only needing to set their markup rates. This is not entering a new market; it is reclaiming a relationship that has already been running.

What changes is the position. All of Stripe's previous large acquisitions were on the 'collecting money' side, while OpenRouter is its first crossing to the other side of the ledger, managing how businesses spend money.

Another change worth noting: its building method has changed.

The entire suite for dollars—Payments, Connect, Billing, Radar, Atlas, Terminal—was basically built piece by piece by Stripe itself. However, for tokens, the suite—Bridge, Privy, Metronome, OpenRouter—was all acquired. The same company, the same shape, the first time built from scratch, the second time entirely bought.

When it was built for the first time, the shape did not yet exist, so it had to be explored. The second time, the shape is already clear, and the only question is who stands in that position first. Returning to the initial statement about the singularity: if you believe that a phase change has occurred, you do not have the time to build it all from scratch. What is acquired is not revenue, but time.

The two flows are not rhetoric; they are the two sides of the ledger.

Similarly, what makes tokens a currency is not their attributes, but their billing.

II. Three Tokens, All in Stripe's Hands

In the past few weeks, the industry has almost simultaneously awakened. From Ramp to Cursor, multiple companies have launched their own routers in a short time. This is a replay of the emergence of acquiring institutions: when a cost item is significant enough to warrant optimization, a middle layer will spontaneously grow. This has happened once with dollars.

But it also indicates something else. Things that can be created in a few weeks are not worth over $7 billion.

2.1 Three Layers of Meaning of Token

The term 'token' internally refers to at least three things at Stripe, all of which are its core businesses.

The first type is network tokens, substitutes for card credentials, the set that Stripe collaborates with Visa, Mastercard, and American Express to enhance authorization rates and resist transaction failures caused by card number changes. This also applies to the AI economy.

The second type is stablecoin tokens, represented by Bridge, Tempo, Privy, and Open Standard, which tokenize the currency itself.

The third type is AI tokens, represented by OpenRouter, Token Billing, and Metronome, which are units of intelligent consumption.

The three tokens perform the same action: abstracting a value into a programmable symbol. The first was done by card organizations, the second by tokenized currencies like stablecoins, and the third pushes it to scales never covered by card organizations—small, high-frequency, with trillions of units daily. Tokens have never been the exclusive domain of the crypto industry; they are what the payments industry has been doing for twenty years.

There is also a division of labor among the three: AI tokens generate bills, network tokens complete card transactions, and stablecoins handle settlements. The statement in the shareholder letter that 'tokens can easily cross borders, and global coverage capabilities are more important than ever' is an expression of the token for the AI economy that is Global by Default.

And today, all three of these elements are in the hands of the same company. The payment industry has spent twenty years repeating the same actions, and Stripe has captured the results of all three: it partners with card organizations, it acquired Bridge and Privy for stablecoins, and it bought OpenRouter for AI.

2.2 Models as Products, Measurement Layer as Cash Register

Looking at these three elements together, Stripe and OpenRouter are essentially the same; the underlying product has simply shifted from dollars to tokens.

Where are the similarities? Stripe connects different countries, banks, card organizations, and payment methods into a relatively unified interface; OpenRouter connects over eighty suppliers, four hundred models, and inference services into a unified interface. One routes funds, the other routes computing power.

Payment has never been just about "moving money"; the value lies in the multitude of activities that occur simultaneously: measurement, routing, pricing, risk control, and reconciliation. Models are products, the measurement layer is the cash register, and Stripe excels at being that cash register.

The AI economy component chart in the shareholder letter represents the completed form of this structure:

This chart should be read alongside another section of the letter. When describing the changes in current business forms, the letter lists four items: existing companies are transforming their business models, measurement and billing are on the rise while many traditional models are declining; companies are launching new products faster, with speed and flexibility becoming the norm; company registrations are accelerating; and stablecoins are rapidly gaining popularity, potentially accelerating further as they become the native currency of the AI economy.

These four items correspond to four layers of shelves. This chart is not a product list; it is a company that controls the cash flow of cutting-edge businesses, making predictions about the business forms of 2026.

The entry method is also clearly stated: these elements will not be sold separately but will be deeply embedded into Stripe's existing products and platforms, with the process "not necessarily obvious." Stripe anticipates a transfer of market share, with the AI economy gradually taking business away from the previous AI era. As for how widely it will spread, one number suffices: 88% of the companies in Forbes AI 50 use Stripe, including OpenAI and Anthropic, while the remaining 12% have mostly not yet begun monetizing.

3. When Tokens Become Stripe's New Dollar

If tokens truly are the new dollars, then the entire ecosystem that grew around the dollar should also grow around tokens. This ecosystem is layered: first comes the cash flow, then the take rate on that flow, which must guard against fraud, relying on accumulated data.

3.1 Cash Flow and Take Rate

The payment take rate has not been driven to zero in twenty years, and it has never relied solely on transportation.

Simplifying complex transactions and collecting a take rate is a shared surface for both companies. Stripe itself provided a deeper layer in the second paragraph of the shareholder letter: when it envisions a prosperous world, what truly attracts it are the underlying mechanisms that keep everything running—currency, credit, currency circulation, legal structures, and risk management.

Legal structures and risk management are included in its own list, placed on the same level as currency. This is also a recurring point made by Gaybrick in interviews: Stripe is a company that absorbs complexity and risk; that highly praised API is merely a facade for complexity.

The surface is the take rate, while underneath lies the burden. The real question is: have the two elements behind the take rate, fraud and data, moved over as well?

3.2 Fraud, Now with a Different Target

Fraud was the first to move over.

In Stripe's AI economy product system, Radar, originally used to prevent payment fraud, now prevents token fraud; Metronome and Token Billing, originally for measurement and billing, now account for token consumption; Treasury, originally for fund scheduling, now manages smart budgeting. The same system runs across both asset types.

The most straightforward evidence is the free trial. Once software has a real cost structure for computational power, abuse emerges. Cursor was one of the first companies to encounter this, with about one-sixth of free registrations being abuse, including attempts at model distillation. There are actually two types of theft: abusing free credits steals the quota, while model distillation steals the asset itself.

Similar incidents are happening at more companies. OpenCode co-founder Dax Raad shared backend data indicating that Stripe has intercepted a large number of fraudulent payment attempts for OpenCode in recent days, far exceeding the revenue the company has generated; if these transactions were allowed, they could potentially bankrupt the company. They have been battling this group recently: some have registered thousands of free accounts in bulk, rotating proxies, and then reselling the quotas by token. On August 11, Dax announced the cleanup of 7,013 fraudulent accounts, expecting to save $400,000 monthly.

Stealing quotas and assets, the shareholder letter includes both in the company's mission: the rise of AI in the economy has brought about new forms of theft and fraud that require sophisticated methods to effectively curb.

3.3 Data, Also with a Different Target

The AMP PBC study reached the same conclusion from a completely opposite direction. It begins by stating that this acquisition is neither a routing acquisition nor a billing merger, but a strategic security decision, even specifically stating "it is not that tokens are the new dollars," but adds in parentheses, "although they indeed are."

Its argument is as follows: moving funds is inherently a commoditized business; banks have been doing it for hundreds of years with low margins. What Stripe has truly built is a scalable online trust machine. Radar's moat does not lie in the model itself but in the fact that it has been training on massive adversarial transactions daily for ten years. The freshness of the corpus cannot be replicated, and the corpus can only come from sitting in the flow.

This is almost a direct replica of financial risk control. The reason credit card fraud models can be developed is that someone holds cross-institutional transaction data; a single bank's ledger cannot train a Visa risk control model. The same structure applies to tokens: cutting-edge labs can only see their own models, which is equivalent to only seeing their own accounts; cloud vendors can see the infrastructure but not the intent. The only place that can see the full picture of cross-model behavior is the routing layer.

We push from "money" to "risk control," while AMP pushes back from "risk control" to "money," and both paths converge at the same point.

Determining whether something is money is not based on who declares it to be money, but whether the systems that prevent money recognize it.

This is not our analogy. An investor sitting on the board of OpenRouter explained this transaction using the same logic.

4. Will It Be Stripe's New Dollar?

Cash flow, take rate, fraud, data—all layers have moved over. So, is the token already a dollar?

This question is reversed. The real question should be: if the token is the new dollar, what position does Stripe hold in this system?

4.1 From Transportation to Configuration to Pricing

The conclusion of the shareholder letter in the section on OpenRouter provides the answer, using very specific verbs: helping businesses effectively configure this new currency called intelligence capital.

Configuration is not transportation. Transportation asks how this money moves from A to B without incident; configuration asks whether this money should be spent here and whether it is worth the price. In terms of products, Stripe's "configuration" consists of three combined elements: knowing the true cost of each unit of intelligence (Token Billing synchronizes model prices across the network), knowing where it should be routed (OpenRouter), and knowing which consumption is real and which is fraudulent (Radar migration).

These three elements together constitute pricing power.

Interestingly, blockchain amplifies the value of this rather than diminishing it. Stablecoins address the transfer issue—how money moves, who keeps the accounts, how long it takes to settle—but they have never solved the valuation of the transfer target. A stablecoin payment can be completed in a second, but "how much should this model invocation cost" is something that blockchain and stablecoins cannot answer. Once the cost of transportation is driven close to zero, all value is squeezed to the pricing side.

This also brings the earlier statement about "two streams of digital flow" into reality. Capital flow has a price; interest rates, exchange rates, and risk premiums are all prices. Intelligent flow is now also starting to have a price, and Stripe wants to be the one quoting that price.

Tokens will be Stripe's new dollar, but not because it spends tokens as money; it wants to become the price setter of this new currency.

On the dollar side, pricing power is distributed among central banks, banks, card organizations, and markets, taking a long time to develop into what it is today. No one has yet occupied this position on the token side, and Stripe has just acquired the only observation point that can see the whole picture.

However, position does not equal qualification. For a price to become a price, it must be accepted; and acceptance requires a set of universally recognized rules: how to identify the identity of agents, how to screen cross-border payments initiated by machines, and under which license stablecoin settlements fall. Each of these has an answer on the dollar side, while none exist on the token side.

Risk control is capability, while rules are consensus. Capability can be bought, but consensus can only be waited for.

4.2 Two Cards, Two Dimensions

In the same matter, card organizations are playing another game.

Visa's Intelligent Commerce issues limited-scope tokenized credentials to agents; a ticket-booking agent receives a token that is only valid for specific airlines and specific travel windows. Mastercard's Agent Pay follows the same idea, combining limited credentials with programmable spending controls. Visa summarizes its design philosophy in one sentence: adding a layer of agent-ready solutions on top of existing infrastructure, without building a new network.

Card organizations want agents to learn how to use cards, while Stripe aims to price what agents spend.

The former changes the initiator, while the product remains the same—still USD and physical goods. It aims to solve the problem of "when the button is pressed by something other than a person," a problem it has already addressed: from offline to online, from magnetic stripes to online verification of bank cards, each time using the same method to issue a limited credential to a new initiator. The latter changes the product itself; Stripe is not satisfied with merely enabling agents to complete a payment; it wants to price, bill, route, and manage risk in this intelligent business.

For this reason, both companies are partners with Stripe rather than adversaries; they are simply not on the same dimension.

In the same shareholder letter, Stripe completed its largest acquisition ever, acquiring the only cross-model behavioral data available on the market.

Conclusion

A company that claims to be "the Stripe of AI" ultimately became part of Stripe.

Will tokens become Stripe's new USD? Yes, but not because Stripe uses tokens as money; it's because it wants to be the price setter for this new currency.

Stripe has acquired the vantage point to see the whole picture and the ability to assess how much each unit of intelligence is worth. What it cannot acquire is the set of rules that will make this price recognized.

It took decades for the USD to reach this point, while tokens are just starting out.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.