Is AI going to dismantle the toll booths of the US Internet?
Earlier this year, the U.S. stock market was plagued by an "AI panic trade".
A week after Anthropic launched its AI plugin, the market value of U.S. stocks in the software and services sector evaporated by nearly $1 trillion. The unspoken subtext among traders is straightforward: the core business of SaaS can be easily handled by AI.
But to attribute the panic solely to SaaS is to take a too narrow view. The impact of AI on the internet business model has become tangible after Muse was launched.
Personal Agent represented by Muse has already shown its fangs to replace super traffic entrances. After the market saw this substitution possibility, entrance-type platforms generally declined, and Booking and Airbnb, which saw the largest drops, both lost nearly 20% in less than a month.
Muse bypasses the product interface, checks goods for users, orders takeout, hails taxis, and completes the whole process of a task from start to finish. When the entrance is bypassed, the underlying right of traffic distribution is also taken away, and the share of core monetization methods such as advertising monetization and transaction commissions will naturally be eroded.
However, the same impact does not weigh equally on Chinese and American companies.
The U.S. internet is open and decentralized. Traffic platforms such as Meta and Google are only responsible for diverting traffic, and users eventually complete transactions on independent websites. When the scenarios are scattered, it is difficult for giants to block the path, and Agents can bypass the giants to place orders directly on public independent sites.
The situation in China is different. China's internet has long formed closed loops in all scenarios, and platforms themselves complete the entire process from traffic entry to transaction completion. This kind of closure can directly block Personal Agents.
It is far more difficult for AI to subvert Chinese internet companies than their American counterparts. But high difficulty does not mean no risk, and many domestic companies are facing hidden worries of "being turned into suppliers".
The more users rely on AI for decision-making, the looser the traffic entrance will be, and platforms will only be left with heavy asset links such as logistics, warehousing, instant delivery, payment and clearing, and after-sales services. "Being turned into suppliers" will weaken the bargaining power of platforms, and their profit structure will naturally decline.
After all considerations, Apple is still the one least likely to lose.
/ 01 / The U.S. Internet's Business Model Has Been Shaken
On the eve of Anthropic's listing, the market's highest valuation expectation for it was $2 trillion. For comparison, Meta's market capitalization is $1.58 trillion. A company that has just started commercialization not long ago has a higher valuation than the world's leading social media giant.
The gap in valuation multiples is even more obvious: when annualizing the revenue of the first half of this year, Anthropic has a 61x PS ratio, Meta has 8x, and Google has less than 10x.
Obviously, capital is not just betting on its intrinsic value. What it is buying is the possibility of substitution: new technologies will overthrow the old entrances, large models will take over search, matching, and content production, and gradually erode the most core profit pools of internet giants.
Especially after the emergence of Personal Agent, the winning rate of this bet seems to have increased somewhat.
On September 8, Meta's personal Agent product Muse was launched, and it was only open to two markets: the United States and Canada. Even so, it quickly topped the free list of the App Store, with downloads exceeding 10 million in 40 days, faster than ChatGPT in its early days.
What Muse does is to take over the links that users originally had to do by themselves: price comparison, order placement, payment, booking flights, hotels and restaurants, filling out forms, sending emails, scheduling, and communicating with customer service. Zuckerberg designed two revenue sources for it: transaction commissions and subscriptions.
As soon as Muse was launched, Meta's identity changed: from the "American Baidu" that had repeatedly encountered setbacks, it became the leader of Personal Agents. Its market value increased by 290 billion U.S. dollars within a week.
The decline on the other side came quickly. After the market understood the direction of Muse, entrance-type platforms generally fell, with Google and Amazon falling 3.80% and 2.24% on a single day respectively, which was a slight decline; Booking and Airbnb both fell by nearly 20% in September.
Why can the launch of a product cause a group of platforms to fall at the same time? The reason does not lie in Muse itself, but in what it points to.
The foundation of the internet business is the attention economy. DAU and user duration are the tax sources, and the take rate is the tax rate. All products are striving to acquire new users and compete for user time. In the final analysis, they are enclosing land first, bringing users into their own interfaces, and then converting traffic into money through advertising and e-commerce.
But Muse is cutting off the traffic entrances of the tax source land. When Personal Agent bypasses the product interface, checks goods for users, orders takeout, books tickets, hails taxis, and completes the whole process of a task from start to finish, the most underlying traffic distribution right of the internet will also be taken away.
a16z is indeed prescient on this matter. It wrote a sentence in its 2025 annual report: "The KPI of screen time is coming to an end." Perhaps in the future, the daily active user indicator will be replaced by another indicator, namely Task Completion Rate (TCR).
Once the measurement standard changes, the business model will be updated to a new generation. However, this generational replacement does not progress synchronously in the Chinese and American markets.
/ 02 / Domestic Internet Is Facing Hidden Worries of Deflation
Muse is not new. Doubao can even say: what you have, I have long had.
As early as December 2025, Doubao Mobile had almost fully implemented all the functions of Muse. But it did not cause any splash, and the stock prices of domestic internet assets did not react at all.
The reason is not complicated: as soon as it enters a super App, it will encounter "abnormal login environment" and the account will be forced to log out. The same product leads to completely different stock price reactions in China and the United States, and the root cause lies in the ecosystem.
U.S. internet services are open and decentralized. Meta and Google play the role of traffic diversion, and users eventually complete transactions on independent websites. Even as strong as Amazon, its market share in U.S. e-commerce is less than 40%. More often, when people buy something or book a room, they need to open different official websites, register accounts, fill in addresses, enter card numbers, and complete payments.
The scenarios are not closed loops, and each platform is only responsible for a part of the whole process, so no one can close the door for the entire network. Therefore, Agents have operating space in the open ecosystem, bypassing traffic platforms to make them lose a sum of advertising revenue; even transaction websites such as Expedia may be downgraded from an entrance to a supplier, with commissions and bidding advertising revenue affected accordingly.
China has a different ecosystem. China's internet has long formed closed loops in all scenarios, and platforms themselves complete the entire process from entrance to transaction completion. This kind of closure can directly block Personal Agents outside. In other words, it is far more difficult for it to subvert Chinese internet companies than American ones.
However, high barriers do not mean no hidden worries. AI is bringing hidden worries of deflation to internet platforms.
Advertising is the core monetization model of the internet. This business can be simplified into a multiplication formula: number of ad slots × unit price of ad slots. The risk of AI is that both factors are changing at the same time.
First look at the number. The physical basis of ad slots is user stay time, and information feeds, splash screens, and recommendation slots on detail pages are all sold based on exposure and duration. Once users get used to delegating their decisions to AI, the exposure volume will drop, and the ad inventory will be compressed accordingly.
Then look at the unit price. The price of advertising is built on overload of choices and information asymmetry. When users are overwhelmed by too many options and merchants cannot fully explain their products, there is a premium in between. What AI does is exactly to eliminate the information gap, so the unit price will move towards convergence.
There is a deeper layer, which is the change of identity.
The high profitability of internet giants in the past was built on the super traffic entrance and the resulting right of traffic distribution; once users increasingly rely on AI for decision-making, this entrance will become loose, and platforms will only be left with heavy-asset, high-cost links such as logistics, warehousing, instant delivery, payment and clearing, and after-sales services. Being turned into suppliers will weaken the bargaining power of platforms, and their profit structure may decline.
While the internet industry is in a precarious state, terminal devices represented by Apple have become a safe haven instead.
/ 03 / Only Apple Is Hard to Lose
This year, no matter how U.S. internet assets fluctuate, Apple has remained firmly in a stable position.
Since the beginning of the year, Apple's stock has risen by 26%, with its market capitalization reaching 5 trillion U.S. dollars, and this "elephant" is still dancing. In the same period, Google's stock rose by 1%, Meta's fell by 4%, and Amazon's rose by 7%.
Apple can become a safe haven because it is the one least impacted by AI.
Mobile phones are currently irreplaceable personal computing terminals, and they are also a more underlying entrance than AI and the internet. No matter how well the model runs, it will eventually have to be deployed on this device. Therefore, no matter who wins the AI race in the end, Apple will not lose.
This position also allows it to avoid a capital consumption war.
Internet giants do not have this choice. In order not to be eliminated by AI, they can only spare no effort to develop AI. If they succeed, they will still be the core assets of the next era; if others succeed, they will be subverted.
The harder they work on AI, the faster they burn money. Many internet giants have already exhausted their free cash flow. According to Wall Street forecasts, the total free cash flow of Amazon, Alphabet, Microsoft and Meta in the third quarter is expected to drop to about 4 billion U.S. dollars, while since 2022, the average free cash flow of these companies per quarter has been 45 billion U.S. dollars.
Tech giants that do not have enough cash flow to burn have also become large debtors. This year, Google issued 31.5 billion U.S. dollars in bonds, Amazon issued 37 billion U.S. dollars in bonds in March, and raised about 25 billion U.S. dollars in financing in July.
In comparison, Apple, which does not need to scramble to invest heavily in AI, lives a very comfortable life. While internet companies are frantically purchasing GPUs, Apple's capital expenditure is only one-tenth of that of internet giants, and its free cash flow is still as high as hundreds of billions of U.S. dollars.
No one knows who will win the AI battle now, but it is a consensus of many people that Apple will most likely not lose.
Disclaimer: This article (report) is written based on publicly available information or information provided by interviewees, but Read Finance and the author of the article do not guarantee the completeness and accuracy of such information. Under no circumstances shall the information or opinions expressed in this article (report) constitute investment advice to anyone.
This article is from the WeChat official account "Read Finance", author: Yang Yang, published with authorization from 36Kr.