In today's booming era of new consumption, "self-pleasing consumption" was once a buzzword. The term, in essence, is about pleasing oneself. As it turns out, this notion of "pleasing oneself" is not confined to the new consumption sector—it looms just as large in AI. This is especially conspicuous among large-model players.

In the past, when many large-model players rolled out version iterations and updates, they tended to stay in their own lane, looking inward rather than outward. When one player shipped a new version, others on the same track would follow suit. This approach certainly accelerated the evolution of large models, but it also boxed them into a narrow scope, limiting what they could actually do.

As AI development enters deep waters—and especially as more and more large-model players turn their attention to application and implementation—we are seeing a shift. Players are no longer simply competing over models; they are pivoting to the application and implementation level, beginning to prove their worth by "getting tasks done." In moving from "competing on models" to "getting tasks done," AI is no longer a solo performance confined to its own track. It has become a symphony performed in concert with industry.

AI Has Already Become Embedded in Industry

Today, the shadow of AI falls on every detail of every industry—e-commerce platforms, social platforms, traditional industries alike. No corner is untouched. The fundamental reason is that AI is no longer something confined to AI itself; it has become deeply fused with a wide range of business formats. Using AI to search for information, handle office work, and socialize is becoming the new default for more and more people.

This "embedding" is not a conceptual flourish. It has already landed in the depths of specific industrial processes.

Take the home-textile industry. Nantong Silicon-Based Jingwei, a company with 634 home-textile stores on cross-border e-commerce platforms, has only 11 human employees—but it is supported in its work by 932 "digital employees": 30 for pattern design, 60 for virtual shooting, 30 for AI programming, and most of the rest handling product listing and customer service. A human designer needs at least a day to complete a single drawing; an AI designer needs only 10 seconds. Image creation, copywriting, and product listing across all 634 cross-border stores are handled entirely by AI, with over 100,000 new product links listed daily. Last year, the company's revenue climbed to 2.5 times the previous year's figure.

On the e-commerce operations side, Taobao Factory's "Spark" system equips industrial-belt merchants with an intelligent working model of "1 AI store manager + N agent assistants." It has boosted the hit rate of merchant product selection by 2.5 times, raised the ad-investment signing rate to 90%, and achieved a comprehensive cost reduction of about 50%.

In industrial procurement, JD Industrial's vertical large model "Gong Xiaozhi" completes tasks across 21 agent scenarios with just 8B parameters. By estimate, end-to-end digital-intelligent procurement can save 20%–30% in costs and cut transaction time by 30%.

By penetrating deep into industries, AI achieves a thorough transformation of their inner workings. With AI, upstream supply becomes more precise and efficiency gains more rapid. When AI begins to embed itself in industry, what it sets off is a deep transformation running from the front end to the back end.

Through such transformation, what changes is not merely how an industry connects with its downstream—more importantly, the industry itself begins to be reshaped from within. As AI becomes more extensively and deeply embedded, players no longer compete merely on model capabilities within their own track. They increasingly compete on the depth and breadth of their embedding into industry, and on how much real change AI brings. When AI is deeply embedded in industry, it becomes new-type infrastructure for industrial development, truly triggering a profound transformation of industry from the inside out.

AI Has Already Built a Full Industrial Chain

Looking back at past AI development, players of different stripes were all focused on their own tracks—a key manifestation of AI players "pleasing themselves." Large-model players focused on building large models; embodied-intelligence players focused on building embodied intelligence; computing-power players focused on building computing power. Players in each segmented field kept to their own lanes, standing out within them and winning the favor of investors.

As AI development enters deep waters—and especially as it shifts from "pleasing oneself" to "collaboration"—sticking solely to one's own track and field is beginning to fall short of what it takes for AI to penetrate deeply into industries and scenarios. So how to build a strong industrial chain and create a more mature AI "complex" through mutual collaboration has become the new choice for more and more players.

This trend has already become quite clear in 2026.

At WAIC 2026, Blue Core Computing's self-developed LX500 CPU chip can independently support inference of trillion-level-parameter domestic open-source large models, improving comprehensive AI inference performance by about 30% at the same power consumption and reducing comprehensive hardware procurement costs by about 20% at the same business throughput. It has already entered the supply-chain systems of manufacturers such as Lenovo.

Meituan disclosed that its LongCat-2.0 trillion-parameter large model completed the entire process—from pre-training initialization to final release—on domestic chips, calling more than 50,000 domestic chips at peak during training. The overall training cycle was not significantly lengthened by the use of domestic computing power, and performance improved by nearly 50% compared with the initial training version.

At the computing-power scheduling layer, Infinigence's autonomous infrastructure platform has served 3.7 million P of computing power, covering 16 different chips. It has jointly launched the "Tianwen Plan Ecological Alliance" with GPU manufacturers such as Biren, Enflame Technology, and Iluvatar CoreX, connecting the full chain from "silicon-based computing power" to "model intelligence" and then to "industry implementation."

Today's AI no longer fights alone. It achieves larger goals through collaborative coordination among the elements, processes, and links within the industrial chain. Collaboration between AI software and hardware, and between AI models and applications, has become the mainstream of AI industry development.

For AI players, no longer being confined to their own industry—but building an industrial chain that spans upstream and downstream, and using it to incubate more mature AI products—is becoming the new choice for more and more of them. Beyond collaboration across processes and links, collaboration across traditionally defined tracks has also emerged as a new trend. In the large-model track, collaboration among vertical models in each segmented field, and their collaboration with general models at a larger level, is reshaping how AI is supplied.

AI Has Begun to Generate Its Own Revenue

A look back at the development of past AI players makes it clear that growth was mainly based on capital infusion. This was almost universally true—for AI players born from giants and newly emerging players alike. As investment in AI from capital markets enters a new cycle, more and more AI players are shifting from relying solely on capital infusion to a new path that balances self-generated revenue with capital infusion.

At this point, merely "pleasing oneself" rather than "pleasing industries" will certainly make self-generated revenue impossible.

A landmark change is that the revenue structure of leading large-model companies is undergoing a "track switch."

Zhipu's revenue in the first half of 2026 reached 954 million yuan, a year-on-year increase of 399.7%. Of that, open-platform and API business revenue reached 825 million yuan, a year-on-year surge of more than 27 times—rising from 15.2% of total revenue in the same period last year to 86.5%. As of the end of August 2026, ARR reached 1.6 billion US dollars; while token call volume grew 40 times compared with the beginning of the year, average API pricing rose by about 101% instead of falling. Management's judgment at the earnings conference was telling: "Once model capabilities are proven, the business begins to generate its own revenue."

Kuaishou's Kling AI demonstrated the path to revenue generation on the application side: operating revenue in the first quarter of 2026 exceeded 650 million yuan, a year-on-year increase of more than 300%. The annualized revenue run rate rose from 100 million US dollars a year earlier to nearly 500 million US dollars, with API calls from B-end enterprise customers and paid member subscriptions on the C-end forming a dual-wheel drive.

In film and television production, "Taiping Nian," in which Kling AI participated, compressed what was originally a roughly two-month special-effects production task to within two weeks. Hollywood's "House of David" used Kling to support the generation of hundreds of high-quality shots, at production costs only one-third of quotes from traditional studios.

Overseas markets are verifying the same logic: Anthropic's revenue in the third quarter of 2026 reached 16.956 billion US dollars, a year-on-year surge of 1,127%. Adjusted operating profit turned positive at 2.463 billion US dollars, and revenue output per dollar of computing-power input rose from 0.40 US dollars to 1.72 US dollars.

At the same time, embodied intelligence has begun to move into scenarios such as transportation, logistics, and education, and AI has moved from chat dialogue into fields such as short videos, short-drama production, and office work. In the final analysis, everything tells us that AI is moving from external infusion to self-generated revenue, from "pleasing oneself" to "pleasing industries."

This is, in fact, a completely new upgrade. Through it, AI players are no longer just the "minority." They have begun to connect with people's production and daily life through a variety of products, penetrating into the details of both—and becoming the "majority" within them.

Closing Thoughts

Today's AI is no longer merely competing on parameters, no longer merely "pleasing itself." It has begun to penetrate deep into industries, build its own industrial chain, and achieve a comprehensive transformation. As AI moves from "competing on parameters" to "competing on tasks," from walking alone to operating collaboratively, a brand-new development cycle is opening. On such a new journey, the players that can achieve full commercialization capabilities and truly resonate with industry at the same frequency may be the ones who laugh last.