Written by: Thejaswini M A
Compiled by Saoirse, Foresight News
In 2017, developer Alex Atallah recognized that the crypto industry lacked a user-friendly frontend trading portal. Seeking to address this pain point and capture commercial value, he founded OpenSea—a platform that aggregates trading of unique digital collectibles and became a favorite during the 2020 NFT boom.
When the industry downturn arrived, Alex Atallah did not succumb to disappointment. As the open-source AI wave emerged, he once again saw the fragmentation problem he had once tried to solve. The AI field was flooded with vast amounts of raw supply, leaving developers overwhelmed by fragmented APIs and inconsistent billing systems. The time had come to build a unified front-end portal.
Thus, Alex Atallah founded OpenRouter, calling it the “Stripe of AI.” Stripe provides software developers with a unified API to handle all payment processes in one place; OpenRouter does exactly the same for AI models, which is why it’s called the “Stripe of AI.”
I don’t believe in the idea that wishing makes it so, so I call it a self-fulfilling prophecy: today, Stripe acquired it for over $7 billion.
OpenRouter completed a $113 million Series B round in May this year, with a post-money valuation of $1.3 billion. Stripe's acquisition has quintupled its valuation.
This article will analyze exactly what Stripe acquired and why the real profits are generated at the interface layer.
What does OpenRouter do?
Typically, businesses that want to use AI models from OpenAI, Anthropic, and Google simultaneously must register separate accounts and write custom code for each provider. By integrating with OpenRouter, businesses and developers only need to connect to a single API to access all models. When an application makes a request, OpenRouter automatically routes it to the most cost-effective, fastest-responsing, most privacy-secure, or currently available model among its 500 supported AI models.
The platform's transaction volume has increased fivefold in six months, currently routing 25 trillion tokens per week for 8 million users.
OpenRouter charges developers the base cost of the models themselves, a 5.5% fee for credit card payments, and a 5% fee for cryptocurrency payments. Research firm Sacra estimates that, as of March 2026, the platform's $900 million in total transaction volume translated to an annualized revenue of $50 million. Since then, traffic has continued to rise, prompting Stripe to pay a premium of 50–140 times the sales multiple to acquire the business.
Stripe can easily afford this expense. The payment giant generated $6.8 billion in revenue last year, holds $3.2 billion in cash, and was valued at $159 billion in its February equity offering. According to its 2025 public annual report, the platform’s total payment volume reached $1.9 trillion, a 34% year-over-year increase, accounting for approximately 1.6% of global GDP. Patrick Collison and John Collison stated that Stripe serves over 5 million businesses, “including all the leading AI companies,” naming ChatGPT, Claude, Cursor, Replit, and Vercel explicitly. Stripe has not publicly disclosed its overall revenue or broken down the revenue contribution from its AI business.
Source: stripe.com
Stripe has completed its full product lineup, filling in all the components required for AI agents to make autonomous payments.
At the infrastructure level, it spent $1.1 billion to acquire the stablecoin infrastructure company Bridge. Since AI agents require solutions for fund storage and management, Stripe acquired the wallet service provider Privy, with the transaction amount not disclosed.
Next is Metronome, a usage-based billing platform used to track compute consumption, adopted by major players such as OpenAI, Anthropic, and Nvidia to account for resource usage. Stripe acquired it in December 2025 for approximately $1 billion.
Subsequently, Stripe partnered with Paradigm to incubate and launch the payment blockchain Tempo, which officially went live in March, along with a machine payment protocol that enables automatic service payments without manual approval. Stripe also introduced the x402 protocol, allowing AI agents to purchase computing resources using cryptocurrency.
Stripe's strategic deployment of a full-chain AI agent system
However, Stripe also wanted to see the request flow to determine which model should handle the task, and the OpenRouter acquisition perfectly solved this issue.
Since December last year, Patrick Collison, CEO of Stripe, has publicly stated that the AI industry is moving away from fixed monthly subscription models toward pay-as-you-go billing based on actual usage. Stripe already had a backend system for tracking and billing usage; this new transaction gives it control over the frontend portal through which developers access models.
Digital transaction settlement, a dead end in business
Cryptocurrency networks can facilitate asset transfer and settlement, but where does the profit actually come from?
The underlying technology bears the heavy workload and incurs significant costs, but only user-facing applications can generate profits.
Verizon and AT&T have invested billions of dollars to lay cables and build 5G base stations to ensure data transmission, but they are essentially low-margin utilities. Profits are always generated at the interface layer—such as Apple selling smartphones that use the network, or Meta earning advertising revenue through apps on its platform.
Such cases are everywhere. Circle generates 94% of its revenue by investing customer deposits in U.S. Treasuries, even though its routing protocol handles 60% of cross-chain USDC transfers, generating almost no direct income because the underlying settlement blockchains belong to third parties. Circle is launching Arc (mainnet launching September 16) to move up the stack, monetizing through interfaces rather than letting transaction fees go unused. Simply handling transfers and settlements is hard to scale unless a monopoly is formed. Visa and Mastercard also only manage data routing and settlement, yet generate massive profits because they operate a closed system. Unlike public blockchains and the open internet, Visa is a private network; merchants wishing to accept cards must use this infrastructure and pay a 2–3% fee on every transaction.
The numbers for AI payments appear massive, but they hold no real significance. Since May 2025, the x402 protocol has processed over 153 million transactions totaling $40 million, averaging just $0.26 per transfer. Even on low-cost blockchains like Base and Solana, paying network fees for a $0.26 payment is economically unviable. In the Web3 space, hundreds of millions of tiny transactions are often indicative of Sybil attacks—bots repeatedly executing micro-transactions to inflate metrics for future token airdrops, or developers running stress-test scripts in loops.
Compare with OpenRouter: $900 million in annual volume, reliably taking a 5% cut. The underlying payment network handles more work but captures less actual value, earning only pennies in marginal profit.
Even specialized blockchains like Tempo did not issue a native token upon launch. This means they deliberately forgo the substantial token speculative gains typically associated with crypto networks. This trade-off was made to provide enterprise customers with a stable and predictable fee structure.
Although Stripe’s founders aimed to build infrastructure capable of processing millions of transactions per second, Tempo’s current actual utilization is extremely low. Since its launch in March, the network has averaged fewer than one real transaction per second. Stress tests have demonstrated that the network can handle over 18,000 TPS, yet daily fees generated from real business activity amount to only $5.34, totaling approximately $620 in fees over thirty days. While locked-up assets on-chain have reached $40 million, the economic returns from settlement activities remain negligible.
Data source: DefiLlama
Only platforms that directly connect with customers can reap massive profits. The original vision of the crypto industry was for crypto protocols to aggregate AI models and offer unbeatable low prices through token discounts—but this path is now failing. While crypto protocols can access cheap computing power, it is centralized companies that control users and capture the profits.
Chutes, a decentralized compute network built on Bittensor, has processed a total of 9.1 trillion tokens across 400,000 users. During the peak business period in May–June, Chutes generated nearly 22 billion tokens per day via the OpenRouter interface, but still relied on OpenRouter for distribution of up to a quarter of its daily traffic. io.net generates 4 billion tokens per day and similarly depends on this centralized platform for traffic distribution.
Data source: openrouter.ai
Decentralized networks like Chutes do not bear the high costs of data centers and corporate personnel; computing power is provided by ordinary individuals, and participants are rewarded with crypto tokens. As a result, pricing can be significantly lower than that of large tech companies. The cost advantage is clear, but the weakness lies in trust—and trust is precisely the entire value of the interface layer.
Quality control is a major challenge for OpenRouter. When an application makes a request, OpenRouter forwards it to the lowest-cost available independent server. However, servers vary widely in quality—some reduce hardware specifications or alter model parameters to cut costs. A survey of 42 servers revealed that nearly one-third completely concealed their configuration details. To address this, OpenRouter plans to launch a "Certified Zone" for trusted service providers. This issue is even more acute in encrypted networks: malicious nodes, driven by profit motives, actively run downgraded versions of models to increase margins.
Projects like DGrid and Bittensor are exploring theoretical approaches to detect model output degradation, but large-scale real-time verification remains an unsolved challenge.
Service availability is another major practical obstacle. Unlike large tech companies with massive, resilient server clusters, networks like Chutes are highly prone to outages during traffic spikes. To ensure stability, developers still rely on centralized aggregation platforms as a fallback—ultimately paying the very transaction fees they hoped to avoid.
On the revenue side, there is a discrepancy between the project’s publicly reported income and actual market demand. Bittensor publicly claimed $43 million in subnet revenue for the first quarter, but independent analysis verified that actual external paid revenue was at most $15 million, with Chutes’ real revenue falling below $2.5 million. Chutes receives newly minted crypto tokens worth $54,000 daily, sustaining its operations through token inflation rather than a steady stream of genuine paying users.
DGrid has built a gateway compatible with OpenAI standards, integrating over 200 models and settling transactions via x402, while auditing outputs through a "Proof of Quality" mechanism. The platform claims that in the six months prior to launch, 13,000 users generated $20 million in revenue, paving the way for the DGAI token issuance. However, this revenue fundamentally stems from the sale of lifetime membership memberships priced at $1,580 each, granting members future token allocations—effectively a disguised token presale rather than ongoing consumption of API calls.
Decentralized networks can also handle real-world traffic, as demonstrated by Dippy. This AI character app with 8.6 million users abandoned traditional cloud providers and migrated entirely to Bittensor’s Targon network. Although on-chain financial revenue is difficult to fully verify, this represents a major breakthrough, proving that consumer-grade applications can run seamlessly on decentralized infrastructure.
When the platform operates independently, it can fairly allocate traffic. Once taken over by a company like Stripe, conflicts of interest will emerge. Stripe, with its own payment tools, wallets, and blockchain products, will naturally favor directing traffic to its own ecosystem, squeezing competitors.
The biggest selling point of crypto narratives is the absence of centralized control. Decentralized systems cannot be acquired by corporations or have their rules secretly altered to favor proprietary applications. Stripe may develop excellent software, but it can never provide a truly neutral, unbiased underlying pathway.
The business logic at the interface layer is entirely sound, but the crypto industry is still very far from mastering the interface layer.
This Stripe transaction sets a public market valuation for businesses positioned at the intersection of traffic. Reports indicate it paid $7 billion to acquire a business generating approximately $50 million in annual revenue, precisely because this platform connects 8 million developers with around 500 AI models—making it an immensely valuable venture. If you're developing an open-source model routing project, you can now use this valuation as leverage when engaging with venture capitalists.
But the situation has become awkward. In the past, decentralized compute networks grew by leveraging OpenRouter to avoid the hassle of customer acquisition. Now, this gateway to traffic belongs to Stripe, a company with strong commercial incentives to prioritize its own products. If you originally chose OpenRouter for its neutrality—perhaps to access affordable domestic models without interference—you’re likely to feel concerned.
Only a router truly neutral to any party will not deliberately favor traffic. However, to achieve commercial deployment, three major challenges must be overcome simultaneously: proving the trustworthiness of the actual code running on nodes; achieving high availability without relying on centralized backups; and building a sustainable business model that doesn’t depend on token subsidies. The industry has not yet accomplished this.