Pricing for the Internet: The fundamental issue being addressed by x402

Source: decentralised

Written by: Sumanth Neppalli, Nishil Jain

Translation: Good Oppa, Golden Finance

There are two completely different schools of thought in the field of cryptocurrency. As a media outlet, we have the privilege of observing these two viewpoints up close. One faction believes that everything is market-driven, and pricing is key to achieving transparency; the other firmly believes that cryptocurrency is a superior financial technology infrastructure. Our publishing plans flexibly adjust between the two viewpoints, because, like all markets, there is no single truth — we are merely integrating all possible models.

In this issue, Sumanth will delve into how a new payment standard is evolving online. In short, the core question is: what would happen if articles could be read on a pay-per-article basis? To find the answer, we will trace back to the early 1990s to see the experiences of America Online (AOL) when it attempted to price internet access by the minute; explore how Microsoft prices its SaaS subscriptions; and ultimately focus on the case of Claude, which prices conversations based on the amount of text.

In this process, we will elucidate the essence of the x402 protocol, its core participants, and its significance to platforms like Substack. The smart agent network is an increasingly focused theme within our organization.

The disconnect between internet business models and user behavior.

In 2009, Americans visited over 100 websites on average each month; today, users open fewer than 30 applications on average each month, but the time spent has significantly increased — from about half an hour a day in the past to nearly 5 hours now.

Winners (Amazon, Spotify, Netflix, Google, and Meta) have become aggregators, gathering consumer demand, turning occasional use into habitual behavior, and pricing these habits through a subscription model.

The reason this model works is that human attention follows a fixed pattern: we mostly watch Netflix at night and shop on Amazon weekly. Amazon Prime members bundle delivery, returns, and streaming services at a price of $139 per year, and the subscription model eliminates the hassle of frequent payments. Today, Amazon has started pushing ads to subscription users to increase profit margins, forcing users to either watch ads or pay higher fees. When aggregators cannot justify the subscription model, they turn to an advertising model, like Google, profiting by monetizing attention rather than user intent.

The composition of internet traffic has changed dramatically today:

Robots and automation programs now account for nearly half of internet traffic, primarily due to the rapid adoption of artificial intelligence and large language models (LLMs), making it easier and more scalable to create robots.

60% of the dynamic HTTP requests processed by Cloudflare come from API calls - in other words, machine-to-machine communication has taken up most of the traffic.

Our current pricing model is designed for pure human use of the internet, but the current traffic is mainly machine-to-machine and sporadic. Subscription models are based on habitual behavior (listening to Spotify on the way to work, using Slack while working, watching Netflix at night), while advertising models rely on attention economics (people scrolling, clicking, considering purchases). But machines have neither habits nor attention – they only have trigger conditions and task objectives.

Content pricing is not only constrained by the market but also depends on the underlying distribution infrastructure. The music industry has been selling in units of albums for decades because physical media needs to be sold in bundles— the cost of burning 1 song or 12 songs on the same CD is almost the same, retailers need high profit margins, and shelf space is limited. In 2003, when distribution media shifted to the internet, iTunes changed the pricing unit to singles: purchasing any song from iTunes for $0.99 on a computer and then syncing it to an iPod.

Single releases have improved the efficiency of music discovery, but they have also eroded revenue — most fans only purchase hit songs rather than 10 filler tracks, leading to a decrease in per capita income for many artists.

Subsequently, with the advent of the iPhone, distribution infrastructure changed once again. Cheap cloud storage, 4G networks, and global content delivery networks (CDNs) made accessing any song instantaneous and smooth. Mobile phones are always online, allowing users to instantly access an almost infinite library of songs. Streaming services have re-integrated all music at the access level: for just $9.99 a month, users can listen to all recorded music.

Today, music subscription revenue accounts for more than 85% of the total revenue in the music industry — this is something Taylor Swift is not satisfied with, as she was forced to return to the Spotify platform.

Enterprise software follows the same logic. Since the product is digital, vendors can charge based on the actual resources used. B2B SaaS vendors typically offer predictable service access on a “per seat” basis, billed monthly or annually, and limit features through tiered packages (for example, $50 per user per month, plus $0.001 per API call).

Subscription models cover predictable human usage, while metering systems handle the burst usage demands of machines.

When AWS Lambda runs your function, you only pay for the resources actually consumed. B2B transactions often involve bulk orders or high-value purchases, so the transaction scale is larger, and substantial recurring revenue can be obtained from a smaller but concentrated customer base. Last year, B2B SaaS revenue reached $500 billion, which is 20 times that of the music streaming industry.

If most consumption is now machine-driven and sporadic, why are we still using the pricing model from 2013? Because our current infrastructure is designed for occasional human choices. Subscription models have become the default choice because making a decision once a month is more convenient than a thousand micropayments.

It is not that cryptocurrency has created the underlying infrastructure to support micropayments (although that is also true), but rather that the internet itself has evolved into a behemoth that requires new usage-based pricing models.

Why did micro payments fail?

The dream of paying a few cents for content is as old as the Internet itself. In the 1990s, Digital Equipment's Millicent protocol promised to enable sub-cent transactions; Chaum's DigiCash conducted bank pilots; Rivest's PayWord addressed cryptographic issues. Every few years, someone rediscovers this clever idea: what if we paid $0.002 for each article and $0.01 for each song, just paying for the actual value of the items?

They all failed in the same way: humans hate measuring their pleasure.

America Online paid a high price to understand this in 1995.

They charge dial-up internet fees by the hour. For most users, this is objectively cheaper than a fixed subscription, but customers despise it because it creates a mental burden. Every minute online feels like a timer is running, and every click comes with a small cost. People involuntarily perceive every tiny cost as a “loss”, even if the amount is small. Each click becomes a tiny decision: Is this link worth 0.03 dollars?

In 1996, when America Online switched to unlimited plans, usage doubled overnight.

People would rather pay more than think more. “Pay-as-you-go” sounds efficient, but for humans, it often means anxiety with a price tag.

Odlyzko summarized in his 2003 paper “Reasons Against Micropayments” that people are willing to pay more for fixed-rate packages not because of rationality, but because they crave predictability over efficiency. We would rather pay an extra $30 a month for Netflix than optimize each $0.99 rental fee. Later attempts (such as Blendle and Google One Pass) tried to charge $0.25 to $0.99 per article, but ultimately failed. Unless a large proportion of readers convert to paying users, unit economics do not hold, and the user experience imposes a cognitive burden.

Subscribe to Hell

Isn't life just endless troubles? Perhaps the gods have also adopted a subscription model for human existence.

If we crave the simplicity of subscriptions, why are we complaining about “subscription hell” nowadays? A simple pricing reasoning method is: how often does the problem that the product solves occur?

The demand for entertainment is infinite. The black line in the chart represents this persistent pain point — an ideal state for both users and companies: a smooth, predictable pain point curve. This is also the reason why Netflix has transitioned from a quirky DVD rental service to joining the elite FAANG club — it offers endless content and eliminates billing fatigue.

The simplicity of the subscription model is reshaping the entire entertainment industry. When Hollywood studios saw the soaring stock prices of Netflix, they began to reclaim their film libraries and create their own subscription empires: Disney +, HBO Max, Paramount +, Peacock, Apple TV +, Lionsgate, and more.

The fragmentation of content libraries forces users to purchase more subscription services: to watch anime, one needs to subscribe to Crunchyroll, and to watch Pixar movies, one needs to subscribe to Disney +, turning content consumption into a “portfolio building” issue for users.

Pricing depends on two factors: whether the underlying infrastructure can accurately measure and settle usage, and who must make the decision each time value is consumed.

One-time payments are suitable for rare, unexpected events: buying a book, renting a movie, paying for a consultation. The pain points erupt and then disappear after a single occurrence. This model is applicable in scenarios where tasks are infrequent and the value is clear; sometimes, even the pain points themselves are desirable — we look forward to the experience of going to the cinema to watch a movie or visiting a bookstore to buy a book.

Accurate measurement of usage will tie pricing to work units. This is why you wouldn't pay for half a movie (its value is ambiguous). Figma cannot extract a fixed percentage fee from your monthly output (the value of creation is difficult to measure).

Even if it is not the most profitable method, charging on a monthly basis is easier to operate.

The computing resources are different: the cloud can observe usage every millisecond. Once AWS is able to measure execution time with such fine granularity, renting an entire server becomes unreasonable — the server starts only when needed, and you only pay for its running time. Twilio adopts the same approach for telecommunications services: one API call, one text message segment, one charge.

Ironically, even in fields where we can measure perfectly, we still bill like cable TV. Usage is measured in milliseconds, but funds flow through monthly credit card subscriptions, PDF invoices, or prepaid “credit limits.” To achieve this, each provider makes you go through the same process: create an account, set up OAuth/SSO authentication, generate API key authorization, bind a bank card, set a monthly limit, and then pray that you won't be overcharged.

Some tools require you to pre-charge a credit limit, while others (like Claude) restrict you to lower-tier models when you reach your quota.

Most SaaS products fall into the green “predictable pain points” range: they are too frequent for a one-time purchase and too stable to require precise per-use measurement. Their strategy is tiered packages - you choose a plan that fits your typical monthly usage, and upgrade when your usage exceeds the limit.

Microsoft's “1TB storage per user” limit is an example - it distinguishes between light and heavy users without needing to measure every file operation. The Chief Financial Officer restricts the number of users needing access to higher-tier plans by allocating permissions.

The chaotic middle ground

A simple pricing model classification method is a two-dimensional chart: the X-axis represents usage frequency, and the Y-axis represents usage variance (i.e., the degree of fluctuation in usage patterns of a single user over time). For example, watching Netflix for two hours most nights falls under low variance; whereas an AI agent making 800 API calls in 10 seconds and then stopping work falls under high volatility.

The lower left corner is the one-time payment area: when the task is rare and predictable, a simple “buyout” pricing model is effective, as you only need to incur a one-time cost to proceed.

The top left corner is chaotic “casual browsing network”: irregular news binge reading, link jumping, and low payment willingness. Subscription models are too cumbersome, while micro-payments per click collapse due to decision-making and transaction friction. Advertising has become a financing layer, aggregating millions of tiny, inconsistent views. Global advertising revenue has surpassed $1 trillion, with digital advertising accounting for 70%, indicating that a large part of the internet exists in this low-commitment space.

The bottom right corner is the ideal area for subscription models: Slack, Netflix, and Spotify align with human daily habits. Most SaaS products are here, distinguishing heavy users from light users through tiered packages. Most products offer free tiered packages to encourage users to start, and then gradually shift their usage pattern from the top left corner to the bottom right corner through daily stable habits. The global annual revenue for subscriptions is approximately $500 billion.

The upper right corner is the focal point of the modern internet: LLM queries, proxy operations, serverless burst traffic, API calls, cross-chain transactions, batch jobs, and IoT device communication. The usage is both continuous and volatile. A fixed fee based on seats cannot accurately reflect this reality, but it lowers the psychological barrier to paid initiation — light users pay more, heavy users receive subsidies, and revenue is disconnected from actual consumption.

This is why seat-based products are gradually shifting towards a metered system: retaining the foundational plans for collaboration and support while charging for heavy usage. For example, Dune offers a limited credit allowance each month, with small simple queries being inexpensive, while larger queries that take longer consume more credits.

Cloud services have made millisecond-level billing for computing, data, and API platforms the norm, with credit sold scaling with actual workloads - revenue is gradually becoming tied to the smallest observable units by the network. In 2018, less than 30% of software adopted usage-based pricing; today, that figure is close to 50%, while subscription models still dominate with a 40% share.

If spending is gradually shifting towards a consumption-based model, the market is telling us: pricing needs to align with the pace of work. Machines are quickly becoming the largest consumers on the internet - half of consumers use AI-driven searches, and the content created by machines has already surpassed that created by humans.

The problem is that our infrastructure still operates on an annual account basis. Once you sign a contract with a software vendor, you gain access to their dashboard, including API keys, prepaid credit limits, and month-end invoices. This is fine for users who are accustomed to it, but it can be cumbersome for sporadic software usage. Theoretically, you can set up monthly automatic billing using ACH, UPI, or Venmo, but these methods require batch processing, and their fee structures do not hold up in microtransactions or high-frequency trading scenarios.

This is the significance of cryptocurrency for the internet economy. Stablecoins provide a programmable, global, and precision payment method that can settle in seconds, operate around the clock, and can be held directly by agents instead of being trapped behind bank interfaces. If usage is event-driven, settlements should be too — and cryptocurrency is the first infrastructure that can truly keep up with this pace.

The essence of the x402 protocol

x402 is a payment standard compatible with HTTP that makes use of the 402 status code reserved for micropayments decades ago.

x402 is essentially a way for sellers to verify whether a transaction is completed. Sellers who want to accept on-chain payments with no Gas fees through x402 must integrate with service providers such as Coinbase and Thirdweb.

Imagine Substack charging $0.50 for a paid article: when you click the “Pay to Read” button, Substack returns a 402 code, containing the price, accepted assets (such as USDC), network (like Base or Solana), and related policies, formatted as follows:

Your Metamask wallet authorizes a payment of 0.50 dollars by signing a message and passes it to the service provider. The service provider puts the transaction information on-chain and notifies Substack to unlock the article.

Stablecoins simplify the accounting process, allowing for settlement based on network speed and small denomination amounts, without the need to set up separate accounts with each supplier. With x402, you don't need to pre-fund five credit limit accounts, switch API keys between different environments, or discover quota triggers causing task failures at 4 AM. Human billing can continue using the most suitable credit card method, while all ad-hoc machine-to-machine interactions are automatically and cheaply handled in the background.

You can feel this difference in the smart agent checkout process. Suppose you are trying a new fashion style on the AI fashion chatbot Daydream: nowadays, the shopping process redirects you to Amazon so you can pay using your saved credit card information; whereas in the world of x402, the agent is able to understand the context, retrieve the merchant's address, and pay directly from your Metamask wallet without leaving the chat interface.

The interesting aspect of x402 is that it is currently not a single entity, but rather consists of layers commonly found in real infrastructure. Anyone who builds an AI agent through the Cloudflare Agent Kit can create bots that are priced by operation. Payment giants like Visa and PayPal are also adding x402 as supported infrastructure.

QuickNode provides a practical guide on how to add an x402 paywall to any endpoint. The direction is clear: unify the “Smart Agent Checkout” function at the SDK layer, making x402 a way for agent payment APIs, tools, and ultimately retail procurement.

Integrate x402 protocol

Once the network supports native payments, an obvious question arises: in which areas will it be adopted first? The answer is in scenarios with high-frequency use and transaction values below $1—where subscription models would charge light users excessively (with a minimum monthly subscription fee becoming a barrier). As long as blockchain fees are feasible, x402 can settle each request at machine speed, with an accuracy of up to $0.01.

Two forces make this transition imminent:

Supply Side: The Explosive Growth of Tokenization in Work - LLM Tokens, API Calls, Vector Search, IoT Signals. Every meaningful operation on the modern internet has been appended with a small, machine-readable unit.

Demand Side: SaaS pricing leads to significant waste - about 40% of licenses are underutilized because finance teams prefer to pay per seat (easier to monitor and predict). We measure work at the technical level but bill humans at the seat level.

Event-native billing with limits is a way to align two worlds without scaring off buyers. We can set soft limits and ultimately settle at the optimal price: news websites or developer APIs charge per use, then automatically refund to the published daily cap.

If The Economist sets “$0.02 per article, with a daily limit of $2”, curious readers can browse 180 links without the need for mental calculations — at midnight, the protocol will automatically settle to $2. This model also applies to developer platforms: news organizations can charge for each LLM crawl to sustain future AI browser revenue; search APIs like Algolia can charge $0.0008 per query, totaling $3 for daily usage.

You can already see that consumer-level AI is developing in this direction: when you reach Claude's message limit, it doesn't just display “Limit reached, come back next week,” but rather offers two options on the same screen: upgrade to a higher subscription plan or pay per message to complete the current action.

What is currently lacking is a programmable infrastructure that allows agents to automatically make a second choice - pay per request, without the need for UI pop-ups, bank cards, or manual upgrades.

For most B2B tools, the actual end state is

“Subscription Bottom Line + x402 Burst Billing”: The team retains a basic plan linked to the number of users for collaboration, support, and daily backend usage; occasional heavy computation demands (build minutes, vector search, image generation) will be billed through x402, without requiring a mandatory upgrade to a higher tier.

Better network services can also be accessed: Double Zero aims to provide faster and purer internet services through dedicated fiber optics — by routing proxy traffic to its network, it can offer pricing at x402 per gigabyte, along with clear Service Level Agreements (SLA) and caps. Proxies that require low latency for trading, rendering, or model transitions can temporarily switch to the fast lane, paying for specific burst demands before switching back to the regular lane.

The SaaS industry will accelerate the transition to a usage-based pricing model, but will set up safeguard mechanisms:

Lower customer acquisition and activation costs: Revenue can be generated from the first call, and temporary developers who have not completed the OAuth or card binding process can still pay $0.03 to use the service; agents are more inclined to choose vendors that offer immediate payment options.

Revenue grows in sync with actual usage, rather than relying on seat expansion: this will address the issue of 30%-50% of seats being idle in most enterprises, with core billing shifting towards capped burst usage scenarios.

Pricing becomes a competitive advantage at the product level: “An extra $0.002 for each request to use the fast lane” “Half price for bulk mode” - Startups can increase revenue through such flexible pricing experiments.

The lock-in effect weakens: vendors can be tried without complex integration and time investment, and switching costs decrease.

A world without advertisements

Micro-payments will not completely eliminate advertising, but will narrow the application scope of advertising as the only viable model. Advertising will still perform well in “casual intention” scenarios, while x402 will price the scenarios that advertisements cannot cover - occasionally, users may be willing to pay for a quality article without subscribing to a monthly package.

x402 reduces payment friction and may change the industry landscape after reaching a certain scale:

Substack has 50 million users, with a conversion rate of 10%, which means 5 million paying subscribers, each paying about $7 per month. When the number of paying subscribers doubles to 10 million, Substack could potentially earn more revenue from micropayments — lower friction would encourage more casual readers to switch to paying per article, accelerating the revenue growth curve.

This logic applies to all sellers with “high variance, low frequency” sales: when people occasionally use a product rather than forming a habit, paying per use is more natural than a long-term subscription.

This is somewhat like my experience playing badminton at the local courts: I play two or three times a week, usually with different friends at different venues. Most courts offer monthly memberships, but I prefer not to be tied to a specific one—I enjoy the freedom to choose which court to go to, how often to go, and to skip when I’m tired.

Of course, I know this varies from person to person: some people prefer to regularly go to the nearest court, some enjoy the habit encouragement brought by a subscription model, and others may want to share their membership with friends.

I cannot comment on offline payments, but through x402, such personalized needs can be reflected in the digital world. Users can set their payment preferences through policies, and companies can offer flexible pricing models to accommodate everyone's habits and choices.

x402 The truly shining scene is the smart agent workflow. If the past decade was about transforming humans into logged-in users, then the next decade will be about transforming agents into paying customers.

We are already halfway there: AI routers like Huggingface allow you to choose among multiple LLMs; OpenAI's Atlas is an AI browser that uses LLM to perform tasks for you; x402 integrates as the missing payment infrastructure in this ecosystem – it enables software to settle small bills with other software at the moment the work is completed.

However, infrastructure alone is not enough to constitute a market. Web2 has built a complete support system around bank card networks: banks' KYC verification, merchants' PCI compliance, PayPal's dispute resolution, freezing of cards for fraudulent transactions, and refund mechanisms when issues arise. Smart agent commerce currently lacks these safeguards. Stablecoins + HTTP 402 allow agents to make payments, but they also remove the built-in recourse that people are accustomed to.

How do you recover funds when your shopping agent buys the wrong flight, or your research robot exceeds the data budget?

This is exactly the question we are going to explore in depth next: how developers can use x402 without having to worry about potential failures in the future.