DAU is dead, TPD lives forever.

Written by | Huahua

At the beginning of 2026, while many people were indulged in the data of the red - envelope war, the Silicon Valley tech circle was shocked by a rumor:

OpenAI is quietly abandoning a core indicator that has dominated the Internet for 20 years: DAU (Daily Active Users).

Everyone's first reaction was disbelief.

You know, ChatGPT's weekly active users have just exceeded 800 million, making it the fastest - growing consumer product in human history.

According to the traditional logic of the Internet, this should be the most worthy - of - showing - off data.

However, according to foreign media reports, OpenAI's product manager said at a meeting: DAU tells us how many people open ChatGPT, but it can't tell us how much value these people create.

OpenAI has set its sights on a new indicator:

TPD, which stands for Token Per Day, the daily token consumption.

Because they found that a user who opens ChatGPT only once a day but drives 20 Agents to work automatically is more valuable than 100 users who chat 10 sentences a day.

This is not a numbers game. Instead, the underlying logic of the AI era is undergoing fundamental changes.

1. Why has the indicator that dominated for 20 years become ineffective?

What does DAU measure? It measures people's attention.

In the past two decades, the essence of the Internet has been the attention economy.

Whoever grabs more of users' time can sell more ads, promote more products, and collect more membership fees. So the KPIs of product managers are always these few words: DAU, user time, and retention rate.

WeChat has succeeded. All of users' daily social interactions are completed here. Douyin has succeeded. Every user spends a lot of time on it and can't stop. Today, Doubao has also succeeded because more than 100 million people complete AI interactive conversations every day.

There is an implicit premise behind this logic: people's time is limited, and attention is a scarce resource. Whoever can capture more time wins.

But this premise is being broken through by AI Agents.

Imagine a scenario.

At 9 a.m., you say to your phone: Help me organize the meeting minutes of today's three meetings, analyze the dynamics of competitors last week, and generate a draft of the weekly report.

Then you go to meetings, drink coffee, take a lunch break, and go for a run. When you get home at night, all tasks are completed. The Agent has consumed 500,000 Tokens in the background and finished the work that used to take you a whole day in front of the computer.

What is your DAU contribution on this day? Maybe it's just the time of that one sentence in the morning, counted as one opening. But the productivity you drive is 500,000 Tokens.

Which indicator is more reasonable to measure your value? The answer is obvious.

2. When the interface starts to disappear

The essence of traditional software is the human - machine interaction interface.

If you want to book a plane ticket, the general path is like this:

Open the Ctrip App → Enter the departure and destination → Select the date → Filter flights → Compare prices → Fill in passenger information → Pay → Wait for confirmation.

In this process, every step consumes your attention. Product managers have spent countless hours optimizing the interface to make you more willing to stay for a long time.

But in the Agent era, these steps are no longer necessary.

You just need to say: Help me book a plane ticket to Shanghai tomorrow, economy class, departing before 8 a.m., with a budget of less than 1,000 yuan.

The Agent calls the Ctrip API, compares prices, places an order, and sends a confirmation email, consuming 100,000 Tokens, and the task is completed.

You don't open any App.

Peter Steinberger, the founder of OpenClaw, said something more radical: Agents may kill 80% of applications.

80% is not an exaggeration.

His logic is that when AI can replace humans to complete tasks, most Apps that require manual operation will disappear. You no longer need to open 50 Apps, but drive 50 Agents.

Today, this is already a reality.

What did the marketing team used to do?

Open the email marketing tool, design templates, import the user list, set the sending time, and click send. In the future, you may only need to say: Send a recall email to users who registered last month but haven't paid. The theme is Spring Festival discount, and the tone should be warm but not promotional.

The Agent automatically generates the copy, filters users, and sends emails in batches, consuming 300,000 Tokens. Your DAU contribution may be 0 because you don't open any marketing tools.

The same goes for data analysis. What did product managers used to do every morning? Open the data dashboard, refresh yesterday's data, take screenshots, write analyses, and send them to the work group. Now? You just need to let the Agent automatically collect data, generate reports, and push them to Feishu every morning, consuming 50,000 Tokens. People don't even need to open the BI system.

When tasks can be automatically completed by Agents, the software interface is no longer necessary.

When the interface disappears, DAU loses its meaning.

3. TPD, a new indicator for measuring leverage

TPD, Token Per Day, is the daily token consumption.

It measures not how many times you open an App, but how much computing power you drive.

A more precise definition is that it measures how much computing resources a user mobilizes through AI to complete tasks every day.

The underlying logic of this indicator is that in the AI era, a person's value no longer depends on how much work they can do, but on how many Agents they can drive to work.

A comparison will make it clear.

A traditional programmer writes 200 lines of code a day, DAU = 1 (opens the IDE), and productivity = 200 lines of code.

An AI programmer lets Cursor and GitHub Copilot write 2,000 lines of code a day and only takes charge of reviewing and adjusting. DAU = 1 (opens the IDE), but consumes 500,000 Tokens, and productivity = 2,000 lines of code.

Who is more valuable? Obviously, it's the latter.

Let's look at another set of enterprise - level examples.

Midjourney has 80 employees and a valuation of $10 billion.

Why?

Because their TPD is extremely high. A small number of engineers drive a large number of image - generation tasks. Millions of users around the world consume billions of Tokens every day, creating outputs that used to require tens of thousands of designers.

In other words, 80 people do the work of tens of thousands of people in the past.

The same goes for Cursor. It has 250 employees, a valuation of $29.3 billion, and an annualized revenue of $500 million.

Users consume billions of Tokens every day, driving the coding work of millions of developers around the world. According to the logic of traditional software, how many customer service representatives, how much operation, and how many servers would this require?

But in the AI era, a small number of people + high TPD can support a unicorn.

The total number of GitHub Copilot users has also exceeded 20 million, becoming an indispensable programming assistant for global developers.

This is the value of TPD: It measures leverage, not time.

4. Three rules of the new world

If TPD becomes the new measurement standard, many things will change accordingly.

And the changes will be earth - shattering.

On an individual level, the definition of competitiveness has changed.

In the past, your value = your time × your efficiency. With 24 hours a day, how much work you can do depends on how hard you work and how smart you are.

Now, your value = your judgment × the computing power you drive. A person who can use Agents can do the work of a team in a week in the past in one day.

This is most obvious among programmers. A traditional programmer writes 200 lines of code by hand a day. An AI programmer writes 2,000 lines of code with the help of Cursor and Copilot. The former's TPD is close to 0, while the latter's TPD is 500,000.

In ten years, programmers who can't use AI to write code may be eliminated, just like accountants who can't use computers today.

The same goes for product managers. In the past, to do a competitor analysis, they had to manually open more than a dozen Apps, take screenshots, record, and organize one by one. Now? The Agent automatically collects the update logs, user reviews, and function changes of competitors and generates a complete report in half an hour, consuming 100,000 Tokens.

Your job has changed from collecting information to judging information.

On an enterprise level, the growth formula has changed.

The growth logic of traditional enterprises is to recruit more people and improve labor efficiency. For a company with 100 employees, to double its growth, it either needs to recruit 200 people or double each person's output.

But in the TPD era, the growth formula is: Improve the computing power driven by each individual × the density of Agents in the organization.

Suppose a company with 100 employees, each employee drives an Agent to consume 1 million Tokens a day, and the total output = 100 million Tokens/day. How many people is this equivalent to? If a person who doesn't use AI can only complete 1,000 Tokens of work a day, it is equivalent to the productivity of 100,000 people.

Today, these data are the real results of companies like Midjourney and Cursor. How can a company with less than 100 people support a valuation of billions of dollars? It's not because of a large number of people, but because of high TPD.

On a platform level, the competition rules have changed.

In the past, super - platforms competed for more users. WeChat has 1.4 billion users, and Douyin has nearly 800 million. These are their moats. Newcomers compete for download volume, user growth, and app - store rankings.

But in the future, super - platforms will compete for who can let users drive more computing power. OpenAI doesn't pursue DAU but the API call volume and token consumption. Because they know that the value of a high - TPD user is equivalent to that of 1,000 low - TPD users.

What does this mean?

It means that in future platform wars, it will no longer be a war for users but a war to improve the computing power of each individual. Whoever can let users drive Agents more efficiently will win.

5. The era of charging by the number of people is over

The business logic in the DAU era is simple: acquire customers for free and monetize through ads or memberships.

WeChat, Douyin, and Xiaohongshu all follow this model. The more users, the higher the advertising revenue. Even in the membership model, it charges by the number of people. One membership costs 15 yuan a month, and 1 million members mean 15 million yuan.

But in the TPD era, the business model is completely different: Charge by consumption and pay for value.

According to official disclosures from OpenAI and reports from multiple media, OpenAI's Annual Recurring Revenue (ARR) in 2025 has exceeded $20 billion (approximately RMB 144 billion). A large part of this revenue is calculated based on token consumption.

Enterprise customers consume hundreds of millions of Tokens every month and pay at a price of $0.01 per 1,000 Tokens. A high - value customer can contribute hundreds of thousands of dollars a month.

The same goes for Claude of Anthropic. Their enterprise customers consume hundreds of millions of Tokens on average every month, thousands of times more than ordinary users. The DAU of these customers may be only a few hundred, but their TPD is astronomical.

Let's look at future SaaS companies. Traditional SaaS charges by the number of seats. One employee costs 100 yuan a month, and 100 employees mean 10,000 yuan.

But in the AI era, SaaS will charge by the computing - power package. You buy as much computing power as you need Tokens. For a team of 10 people, if each person drives an Agent to work, they may consume 100 million Tokens a month, contributing 100 times the revenue of the traditional model.

What does this mean? It means that the future value of a platform doesn't depend on how many users it has but on how much computing power users drive.

Suppose two platforms:

Platform A has 100 million DAU, each user opens it 10 times a day, and the average online time is 30 minutes.

Platform B has only 10 million DAU, but each user drives an Agent to consume 10 million Tokens every day.

Which one is more valuable?

The business model of Platform A is advertising and membership, and its revenue ceiling is limited by DAU and ARPU. Suppose each user contributes 100 yuan a year, and 100 million users mean 10 billion yuan in revenue.

Platform B charges by Tokens. If the token price is $0.01 per 1,000 Tokens, the daily revenue = 10 million people × 10 million Tokens × 0.01/1000 = $1 million/day = $365 million/year.

To achieve this revenue, Platform A needs each user to contribute $26.3 in ARPU, which is almost impossible for most Internet products. But Platform B only needs to improve users' computing - power mobilization ability, and its revenue can grow exponentially.

This is why OpenAI doesn't track DAU but the API call volume. Because they know that the future growth point lies not in the number of users but in the computing power driven by users.

6. How fast is this transformation?

Faster than you think.

In 2024, ChatGPT was just popularized, and most people still used it as a chatting tool.

In 2025, Agents began to explode. The number of paid users of GitHub Copilot exceeded 5 million, the monthly active developers of Cursor exceeded 2 million, and enterprises began to deploy AI workflows on a large scale.

During the Spring Festival in 2026, the AI red - envelope war involving billions of people pulled ordinary people in. Your parents started using Doubao, Qianwen, and Yuanbao. They may not know what an Agent is, but they are already using AI to complete tasks.

What about in another year?

Silicon Valley investors are betting that in 2027, software will die on a large scale.

Their logic is that when Agents can directly call APIs to complete tasks, most software that requires manual operation will disappear. You don't need to open 50 Apps; you only need an Agent scheduling center.

This sounds crazy, but look back. It took 5 years for smartphones to become popular, 3 years for mobile payment, and 2 years for short - videos (Extended reading: It took 5 years for smartphones to become popular, 3 years for mobile payment, and only one Spring Festival for AI)

What about