Zhipu's earnings surge as AI commercialization breaks new ground! What are the opportunities?

ME News

joined discussion ·

As APIs capture the traffic tax, what remains of the moats for ByteDance and Google?

Author and source: Leiphone

In June 2026, Cloudflare, the global giant in network infrastructure and content delivery networks (CDN), released a set of data:AI bot traffic accounted for 57.4% of global HTML page view requests, officially surpassing human traffic.

As of today, according to Cloudflare's latest data,

the share of AI bot traffic has risen to 60.4%.

At the time, this seemed like just another technical brief drowned out by the wave of model releases. But by August, the impact of these "digital natives" taking center stage began to emerge:

First, the traditional digital ecosystem is shrinking: as AI agents that do not click on ads sweep across the internet, numerous small and medium-sized websites reliant on ad revenue are experiencing unprecedented "bleeding";

Second, identity verification has become absurd: to block these AI crawlers, CAPTCHAs on major websites have evolved to an "anti-human" level, leaving real users trapped in the ironic dilemma of "how to prove I am me."

Another change is occurring among service providers: recently, ByteDance's Doubao platform officially began charging transaction service fees for orders in specific industries (such as hotels).

Behind this digital reversal lies a restructuring of the internet's underlying logic:Previously, web pages were designed for humans to "view," with dopamine as the core asset; now, they are designed for Agents to "read," with execution rights as the core asset.

This is not merely a simple change in figures, but an upgrade in the logic of internet commerce, particularly forthe two biggest "traffic toll collectors" of the internet era—ByteDance and Googleit could bring about disruptive shocks.

ByteDance's commercial empire is fundamentally built on"human dopamine".

ByteDance's core moat is its"emotional feedback loop": by leveraging algorithms to precisely push short videos, it continuously stimulates users' browsing interest and extends their engagement time. This massive pool of attention resources has further spawned a mature monetization chain encompassing feed-based advertising, live-stream e-commerce, and local lifestyle services. This classic B2C paradigm has established ByteDance's global, phenomenon-level content and consumption ecosystem.

However, the advent of the Agent era has fundamentally shattered this underlying logic.

Third-party AI programs can scrape video scripts, product information, and group-buying discounts from Douyin around the clock in bulk. They freely appropriate the content data that ByteDance has accumulated at enormous cost, yet they contribute neither genuine user watch time nor live-stream engagement to the platform, let alone generate actual consumer orders.

After all, Agents do not produce dopamine.

As AI-driven web traffic grows while human traffic peaks, the "attention tax" that ByteDance has historically relied on for survival is facing exhaustion.

With the logic of traffic flow changing, transaction models must adapt accordingly. AI cares only about task completion and process smoothness. Faced with unemotional algorithmic nodes, ByteDance's course of action isto upgrade its content platform, previously designed for human browsing, into a service network directly callable by AI, thereby re-establishing AI dependency.

From both technological and commercial perspectives, this involves two steps: building external moats and creating internal closed loops.

First, regarding building external moats, this refers to blocking unauthorized AI scraping.

On mobile apps, products such as Douyin, Jinri Toutiao, and Dongchedi have implemented stricter identity verification. When real users browse videos or check products, the security chips in the mobile operating system (Apple's Secure Enclave and Android's TEE) generate an encrypted "device identity credential." Regardless of how realistically cloud-based AI programs simulate human behavior, they cannot obtain the exclusive signature from authentic mobile chips, causing the platform gateway to directly discard their requests.

On the web end, ByteDance has employedDynamic code obfuscation technologyWith every page refresh, the underlying code structure and text layout are randomly shuffled. While human users browsing via standard browsers see everything normally, AI scrapers directly accessing the code retrieve only a jumble of invalid garbled text. To extract meaningful content, one must first render the page as an image and then perform optical character recognition (OCR), which multiplies computational costs several-fold.

However, building walls is merely a defensive measure; the true core lies in how to convert traffic into new transaction "tolls."

Want to leverage Douyin’s vast network of local merchants and e-commerce supply chains? No problem—but you cannot stealthily scrape data on your own. You must integrate your AI within ByteDance’s ecosystem.

ByteDance has fully encapsulated inventory, pricing, and ordering capabilities for all products and services—including Douyin E-commerce and travel/hospitality—into standardized API interfaces.External developers seeking to access these resources must do so through the Doubao large language model or the Coze platform.All AI applications operate within a closed loop inside ByteDance’s ecosystem. From transaction identity verification and fund settlement to revenue sharing, this effectively mandates payment of a "toll," leaving external parties with no opportunity to bypass the platform.

In August 2026, Doubao officially began charging a 12% transaction service fee on travel and hospitality orders. This implementation marks ByteDance’s formal transition from a traffic-based model to a data-driven model:It no longer relies on user ad clicks for revenue but instead takes a cut directly from every transaction facilitated by AI.

ByteDance’s most formidable asset was originally its "massive consumer user base + comprehensive offline service supply chain." In the era of AI agents, this asset has been seamlessly restructured into a new transaction-service model comprising a "proprietary content ecosystem + complete transaction supply chain + on-device security protection."

ByteDance is well aware: while AI agents lack dopamine, the developers behind them are certainly mindful of revenue and expenses. The ultimate goal of this upgrade to data sovereignty is singular: to block all avenues for free-riding, preventing their "dopamine assets" from leaking away without compensation.The only way to tap into the content dividends accumulated by Douyin (or other products) over the past decade is to pay via the API billing meter.。

As a result, ByteDance's moat has not disappeared; rather, it has deepened.

Unlike ByteDance's "dopamine tax" logic, Google's core business in the past was built on"users proactively searching for information"as its foundation.

As the world's largest search engine, Google guards the internet's traffic gateway. When people encounter problems, they open Google to search, browse results, and click on ad links. This advertising bidding model, charged based on impressions and clicks, has been Google's trump card for maintaining its position as a global revenue leader for years.

In plain terms, this is also a form of "traffic tax," levied on all merchants seeking exposure."Click-based rent"。

However, the widespread adoption of AI agents has directly bypassed Google Search’s web interface.

Users simply issue commands directly to the agent, which automatically scours the entire web in the background and compiles the answers. In this process, search ads that rely on clicks for monetization lose their audience, and the commercial chain that Google depends on for its livelihood is severed at the root by AI.

Nevertheless, while the upper-layer advertising business has been impacted, Google still holds a fundamental necessity in the AI era that no one can bypass, namelyreal-time information from the physical world. This is also Google’s new commercial anchor point.

The internal knowledge bases of large language models are static and inherently suffer from timeliness limitations. Any truly functional AI agent, without external real-time search capabilities, risks becoming obsolete due to outdated information. Meanwhile, the world’s largest real-time information database, updated at millisecond speed, is held precisely by Google.

From this perspective, Google can establish three insurmountable moats:

The first layer lies in thereal-time information databaseGoogle has bundled its search and mapping services into tools that AI can directly access. Whenever third-party AI systems require real-world data, they automatically tap into Google’s real-time information repository, effectively providing AI with an external "real-time information source."

The second layer lies ingeospatial dataHigh-precision location data, real-time traffic conditions, business operating statuses, and three-dimensional urban spatial data are essential requirements for AI applications in transportation, local lifestyle services, and logistics. Globally, no AI startup has the capacity to build a global mapping infrastructure of comparable scale from scratch. Consequently, the codebases of such AI systems almost invariably rely on Google Maps API integrations.

The third layer lies inknowledge graph calibrationWithin its AI service platform, Google has upgraded traditional web search to return structured factual data. When an AI system submits a query, Google does not provide a cluttered list of web pages; instead, it returns verified, standardized data that the AI can immediately use for judgment and decision-making.

As a result, Google’s new business model involves offering paid API access to its information services to AI companies and developers worldwide. Regardless of how upstream players like OpenAI, Anthropic, or domestic large-model vendors compete in the agent application space, any intelligent agent requiring access to real-time internet information must pay Google API service fees.

No matter how powerful large language models’ reasoning capabilities become, they cannot execute tasks in a vacuum. The barrier Google has established essentially controls the physical anchor points between Artificial General Intelligence (AGI) and the objective world. No matter how capable an agent is, if it needs to resolve real-world issues involving "when" and "where," it must interface with Google’s APIs. This represents Google’s new "traffic tax" model in the To-Agent era:"I may not need a user interface, but you cannot do without an anchor point."

Whether it is ByteDance's closed-loop transaction ecosystem or Google's information infrastructure, the implementation of these new business models relies heavily on underlyingtechnical support for "human-bot identification."The industry has currently formed a mature combination of strategies, with only one core objective:to distinguish whether visitors are humans or AI, manage data access permissions, and transform the previously "open internet with free crawling" into segmented "paid territories" controlled by platforms.。

Hardware key-based identity authenticationis currently the most reliable method for distinguishing humans from bots. Essentially, it issues an unforgeable "physical access card" to each genuine device.

Industry-standard protocols like WebAuthn rely on secure chips within devices for identity verification. Human users can pass verification via fingerprint or facial recognition; however, AI programs running on cloud servers lack the hardware chips of real phones or computers, cannot obtain this exclusive "access card," and are thus blocked at the gate.

Platforms can also implement refined management: allowing compliant AI partners through official cooperation while blocking all unidentified crawlers.

On the application side, ByteDance focuses more on client-side defense, leveraging apps like Douyin and Toutiao to call upon the mobile phone's secure chip to generate encrypted device identities, thereby intercepting third-party agents.

As the architect of web standards and owner of the Android ecosystem, Google is rolling out hardware-based security keys across all platforms. When AI attempts to perform critical user actions such as executing searches or watching YouTube videos, the server issues a "challenge" that can only be correctly signed by the device's secure enclave. This effectively blocks third-party AI agents.

If hardware authentication is akin to checking an ID card, thenbehavioral fingerprintingis like "identifying individuals by observing their behavior."

Human web browsing behavior is fragmented and irregular: casual scrolling, random clicks, and lingering on pages. In contrast, AI crawlers leave highly regular traces, often navigating pages at millisecond-level intervals, bulk-loading large volumes of pages in short periods, and following rigid data extraction paths. By analyzing metrics such as access frequency, dwell time, and click trajectories, platforms can identify automated programs with high probability.

A more advanced approach involves monitoring the "rhythm" of network connections. For instance, Google employs

JA4

+ technology, which does not inspect transmitted content but rather examines connection establishment details: the order of encryption methods, packet transmission timing, and other low-level environmental signatures that are impossible to conceal. Google's gateways can determine within one millisecond whether the requester is an AI, then immediately impose rate limits, increase pricing, or block access entirely.

A previous test project called Project Lanterna identified 14 well-known AI crawlers, including official crawlers for Claude and GPT, in just 31 hours using this method. Service providers like Cloudflare have compiled global "blacklists" of AI crawlers, allowing websites to enable protection with a single click.

Additionally,Content obfuscation technology, rendering web scraping efforts futile.

The most common technique is dynamic page obfuscation, with ByteDance being a典型 player in this space. While pages appear normal to human users, they present as garbled code to web crawlers. For AI to extract meaningful content, it must first screenshot the entire page and then use optical character recognition (OCR) to extract text, multiplying computational costs several times over. This forces developers to pursue official partnerships.

After blocking unauthorized web crawling, platforms have launched official "authorized channels."

Increasingly, platforms are no longer allowing AI to scrape web pages at will; instead, they are introducing standardized official APIs. Platforms can now control call frequency, charge API fees, and define data access scopes, transforming the previous passive state of "being scraped" into a new business model characterized by "active authorization and pay-per-use" in the API era.

The various service interfaces on ByteDance's Coze platform, as well as Google's search and map data APIs, fundamentally operate on this same logic.

We are crossing a clear industry dividing line. The coordinate system of internet commerce has officially moved beyond the familiar binary framework of To-B and To-C, entering the new epoch of To-A.

Many are offering new definitions:In "To-A," the "A" can stand for AI or Agent. It can also refer to enterprises that adopt AI as their native mindset, core productivity, primary workflow, or even organizational structure—often referred to today as "AI-native" or "AI-pilled" companies.

Behind this new era lies a profound transformation on the demand side. Business logic, once centered on "humans," is shifting toward an "AI agent"-centric model. Direct code-to-code interaction breaks through the commercial ceiling imposed by human attention and cognitive limits, achieving exponential improvements in operational efficiency. "Human-not-in-the-loop" is becoming the norm.

In the B2B era, the shift from human reliance to AI dependence has made data sovereignty, agent ecosystems, underlying information infrastructure, and API access rights critical to business success. ByteDance is consolidating its ecosystem to create a closed-loop transaction system, while Google is digging deep to establish an information foundation. Though their paths differ markedly, both are essentially vying for the new "taxation rights" of the AI age.

The era of the traffic tax may be over, but the war over the API tax has only just begun. As tech giants race ahead once again, what opportunities remain for us?

54K Views