A reminder for China's large model industry: The most critical battle is coming

The large model industry in China is falling into a silent cognitive trap.

In the past six months, DeepSeek has pushed API prices to rock - bottom levels. Kimi has continuously raised the bar in long - text processing, and Doubao of ByteDance has been making inroads in the number of C - end users. Everyone talks about SOTA and Agentic, and every press conference tells stories of technological progress.

These stories are all true. However, they miss the most important thing.

There are countless possible paths for large model commercialization, but there are only two sectors that can support a company with a trillion - dollar market value: programming and office work.

This judgment doesn't require complex arguments. Just take a look at Anthropic. It has only one - seventh of OpenAI's active users, yet it has captured 31.4% of the global LLM market revenue share. Earning more money with fewer users, there's only one reason: users pay for productivity. Once they've used it, they can't go back.

This is the most fundamental logic of large model commercialization. This logic has been verified across the ocean. However, the large model industry in China has remained almost silent in response.

01

Why These Two Sectors?

Programming and office work are the most valuable legacies of the electronic computer and Internet revolutions in the past six decades.

From programmers, lawyers, and editors to accountants, consultants, analysts, and investors, the underlying work methods of all knowledge workers who create value in offices have hardly changed in the past few decades.

Word, Excel, PowerPoint, and IDE are essentially digital paper and pens. For the first time, large models have enabled these tools to understand and execute tasks.

A programmer assisted by a large model may see a several - fold increase in productivity. A legal assistant who can use AI to review contracts could previously review three contracts a day, but now can review thirty. This increase in productivity is real, quantifiable, and users are willing to pay for it. Anthropic's ARR has proven the price the market is willing to pay.

More importantly, this business model naturally has a diffusion mechanism similar to the prisoner's dilemma. If a law firm uses AI - assisted review, its peers must follow suit; otherwise, they will fall behind in terms of efficiency and cost. If a software company uses AI - assisted development and its competitors don't, they will lag in the iteration speed.

With the steam engine, electricity, computers, and the Internet, each tool revolution has divided users into adopters and non - adopters. Adopters move forward, while non - adopters fall behind. Large models will be no exception.

Each of the programming and office work sectors is a market worth trillions of dollars. They cover the core workflows of global knowledge workers. Once a company enters these sectors, it means tapping into the most stable, continuous, and rigid part of corporate budgets.

Chinese large model companies need to launch a full - scale attack on these two sectors, concentrating all machines, talents, and funds in these two directions. This is not an option but a necessity. AGI is a long - distance race, and large model companies without a commercial cycle won't reach the finish line.

02

Enhancing Capabilities, Not Replacing Positions

The business models of the office and programming sectors have a core advantage that other AI sectors don't have: their value lies in enhancement, not replacement.

The logic of replacement is zero - sum. When AI replaces a position, the position disappears, and the expenditure shifts from labor costs to software costs. The total corporate expenditure may not increase and could even decrease. The logic of enhancement is positive - sum. AI enables a programmer to write more and better code, an analyst to cover more and deeper industries, and a lawyer to handle more complex cases.

Enterprises are far more willing to pay for this enhancement than for replacement.

Once a person has used a large model to assist in work, they can't go back to the state without it. Previously, it took three days to write a research report, but now it only takes three hours. Previously, it took two hours to review a contract, but now it only takes twenty minutes. Once this experience is established, going back is unacceptable.

From December 2025 to February 2026, Claude Code's ARR more than doubled. The driving force was not marketing but users' active promotion. Those who had used it told their colleagues, colleagues told their teams, and teams pushed their companies to make purchases.

This is the diffusion mechanism of the enhancement effect. When all programmers use AI for coding, programmers who don't use AI will be eliminated. When all analysts use AI for data modeling, analysts who don't use AI will fall behind. It's not AI that eliminates people; it's those who can use AI that eliminate those who can't.

For Chinese large model companies, this logic points to a clear goal: they must closely approach their competitors in terms of user experience and then leverage the inherent advantages of the Chinese market to scale up.

China has the world's largest group of programmers, with over 7 million software developers. Tens of millions of knowledge workers write documents, create spreadsheets, and hold meetings in offices every day. This is a huge and temporarily protected market. How long the window period lasts depends on geopolitics and the speed of technological evolution. But within this window period, the company that first creates a Chinese version of Claude Code will capture the largest share of profits in this market.

On paper, the capabilities of Chinese models are approaching those of the world's top - tier models (at least in the competitive arena). DeepSeek V4 and Kimi K2.6 top the open - source rankings, and their API costs are much lower than those of their American counterparts. However, the key to commercialization has never been just the technology itself. Product experience, workflow integration, user habit cultivation, and building trust in enterprise - level services are all more challenging than rankings.

Chinese large model companies need to send product managers into the real - world work scenarios of programmers and knowledge workers. They need to understand why users are willing to pay for Claude Code and what is lacking in its Chinese alternatives.

03

The First Business Lesson for Idealists

The current group of Chinese large model entrepreneurs may be the most creative and technologically idealistic group in the past two decades. They believe in AGI, open - source, and that technology can change the world. However, the operating logic of the business world has never been related to idealism.

There is a repeatedly verified law in the history of technological revolutions: the ultimate winner is not the one with the strongest technology but the one that establishes a commercial closed - loop earliest.

Wang Laboratories was a pioneer in word processors, leading in technology by a significant margin. However, it was obsessed with a closed system and refused to be compatible with the IBM PC standard. When Microsoft's Word entered the market with a lower price and a more open strategy, Wang's business empire collapsed within five years. Xerox PARC invented the graphical user interface, the mouse, and Ethernet but never successfully commercialized these technologies. Apple and Microsoft took these inventions from Xerox and each built a trillion - dollar business empire.

Today, Chinese large model companies face not a technological threshold but a cognitive threshold. Many people are still immersed in the arms race of model capabilities, believing that as long as the model is strong enough, commercialization will naturally follow. The model is just the engine, and the product is the car. Users buy the car, not the engine. No matter how good the engine is, without a steering wheel, seats, or brakes, users won't make a purchase.

The cruelest part of large model commercialization is that the window period is limited, and the margin for error is low. When Anthropic has already established user habits and brand recognition, latecomers need to pay several times the cost to enter this market. In the Internet era, Google was not the first search engine, and Facebook was not the first social network, but they were the first to establish network effects and commercial closed - loops. The same is true in the AI era.

What Chinese large model companies need most now is not just the next round of financing or the next version of the model but a sober strategic review: Temporarily put aside all non - core business lines, concentrate the best talents on the programming and office work sectors, quickly establish a closed - loop of product experience, and seize users' minds with the strongest execution ability.

This is a battle that must be won within two years. Those who can't win won't even have the qualification to participate in the next round of the competition.

04

Conclusion: The First to Enter Xianyang Will Be the King

The Chinese large model industry is at a crossroads. On one side is the pursuit and surpassing in technology; on the other side is an almost blank answer sheet in commercialization. The gap between these two aspects will be repeatedly measured by the market with real money in the next two years.

The programming and office work sectors, each worth trillions of dollars, are silently waiting for the first Chinese player to truly enter. This is the ultimate test of a large model company's strategy, execution ability, and business intuition. The company that first establishes a sustainable commercial cycle in these two sectors will have a blood bank to support the long - distance race of AGI and will become the first Chinese large model company with a trillion - dollar market value.

This reminder letter is for all Chinese large model entrepreneurs who are still dreaming. You've built a great engine; now it's time to build the car. The window won't stay open forever.

This article is written based on publicly available information and is for information exchange only. It does not constitute any investment advice.

This article is from the WeChat official account "Jinduan". Author: Mu Yang. Republished by 36Kr with permission.