AI is a mixed bag: 2C customers are reaping the benefits, but 2B customers are stuck in a dilemma

ByteDance continues to increase its investment in the field of AI, with capital expenditure reaching 80 billion yuan in 2024 and is expected to double to 160 billion yuan in 2025, of which 90 billion yuan will be invested in AI computing power and 70 billion yuan will be used for IDC and network equipment. Its "Doubao" large model has covered 16 types of AIGC models such as text, voice, image, music, and video, with nearly 60 million monthly active users, ranking among the leading AI applications in China.

Alibaba has also increased its investment, announcing that it will spend more than 380 billion yuan on AI and cloud infrastructure construction in the next three years. Its hybrid reasoning model "Qianwen 3" has been ranked first in many authoritative lists. Tongyi has open-sourced more than 200 models, with more than 300 million downloads worldwide and more than 100,000 derivative models, forming one of the world's largest open-source model families. AI products have been implemented on a large scale in industries such as retail, manufacturing, and media, and revenue has maintained triple-digit growth for seven consecutive quarters.

In the first quarter of 2025, NetEase deeply applied AI technology in its flagship game "Ni Shui Han", significantly improving the game's development efficiency and user experience. By introducing AI to generate plots and art materials, the development cost of a single character dropped by more than 90%, making content production more efficient. At the same time, the number of daily active users of the game increased by 47%, driving the overall user engagement.

This series of innovative applications is also reflected in the financial performance. Netease's net profit for the quarter increased by 34.9% year-on-year to 10.3 billion yuan, showing the huge potential of AI technology in the gaming industry. After the first quarter report came out, Netease's stock price almost returned to its historical high.

Tencent quickly switched its strategy after connecting to DeepSeek, actively embracing AI with 100 times the enthusiasm. Its capital expenditure in Q1 surged 91% year-on-year to 27.5 billion yuan, of which 90% was invested in AI chips and cloud infrastructure. Its self-developed "Hunyuan Large Model" significantly improves the performance of the advertising system, reducing the response delay to 1/10 of the original. The AI ​​assistant "Yuanbao" was launched in 2024, integrating DeepSeek-R1 with the Hunyuan model. In February 2025, an advertising investment of 281 million yuan pushed it to second place on the iOS free list. Thanks to the deep integration with the WeChat ecosystem, especially the content of official accounts, Yuanbao's DAU increased more than 20 times from February to March, and the daily active volume exceeded 20 million.

Advertisements tailored to each individual, games automatically generated, short video clips, copywriting, customer service Q&A, live streaming scripts... AI is rewriting the logic of content production with over 10 times the efficiency. These ToC scenarios are the "natural home" of AI: processing unstructured data, supporting personalized generation, and realizing the large-scale implementation of creativity.

ToC starts, ToB is restructured: AI impacts the foundation of traditional enterprise software

In the past, enterprise software such as BI, CRM, and ERP built a moat with "high complexity + high customization". However, with the advent of the mobile Internet era, the agile development, high concurrency, and rapid iteration brought by the Internet's native technology stack have long broken the technical monopoly of traditional enterprise software at the architecture layer. At the same time, 2B software is collectively silent on "traffic narrative" and "technical paradigm", and ERP, which is serious about reducing costs and increasing efficiency, can't compete with various fancy middle platforms. Faced with Internet competition, enterprise software companies are lagging behind in market attention, technological attractiveness, and talent competition, and it is difficult to continue to obtain high premium returns from customers.

Today, AI may repeat this change, completely impacting the basic design and delivery logic of enterprise software. This round of AI reconstruction of B2B first penetrates from the "interface layer" to the "base layer": Initially, AI capabilities were only used as tool plug-ins and embedded in the original system of the enterprise, such as using AI assistants to generate reports, polish emails, and classify bills. But soon, more and more AI Agents are being directly embedded in the main business flow of the enterprise, such as AI employees embedded in IM chat windows, smart assistants connected to approval processes, and even "AI process agents" that connect to financial systems to automatically process contracts, payments, and budgets.

The essence of this penetration is that AI begins to replace people, not just tools . In the traditional enterprise software logic, the goal of the system is to "assist people to complete tasks better"; in the AI ​​Agent architecture, the goal of the system has become "to directly complete tasks." This is not an enhancement of efficiency, but a transfer of execution rights. It challenges the enterprise management paradigm, process control rights, and even the logic of organizational division of labor.

At the same time, the deployment method of AI systems is also breaking down the engineering boundaries of traditional software. The delivery cycle of traditional enterprise software such as ERP, CRM, and BI is often measured in "quarters", and requires multiple steps such as solution design, parameter configuration, permission definition, and process testing. Today, the deployment of AI applications can be completed in "hours". With just a prompt, a RAG interface, or a dify process, you can build an AI agent that can read documents, understand context, and respond automatically.

This means that the "complexity moat" on which traditional enterprise software relies for survival is being rapidly filled by AI's "general intelligence + modular calls".

What is more impactful is the change at the business model level .

The profit model of enterprise software has always been to combine license authorization with service fees. The premise of this model is "re-delivery + re-binding": if customers want to use your system, they must accept your standards, your integration method, and your annual maintenance contract.

The popularity of AI has made this binding no longer necessary - because AI no longer relies on huge system integration, it relies on calling and customized reasoning, on-demand, result-oriented, plug-and-play. More and more one-person teams or small entrepreneurs use AI tools to automate the entire business process and complete complex software delivery at a very low cost.

For example, some SaaS entrepreneurs have proposed:

"If an AI can help me with reimbursement, financial reconciliation, and audit reports, why should I buy a lengthy and complicated ERP system? I only need to pay for the results."

This new paradigm of "Result-as-a-Service" is challenging the business logic of the entire enterprise software industry. It also forces traditional vendors to face up to the fact that what customers really want is not a system, but an intelligent service closed loop - a system that can understand business intent, process process logic, and provide real-time feedback on results, rather than a tool that can only be operated by users themselves.

In this context, AI is not only an efficiency tool , but also an engine for the liberation of digital workers. It will guide the entire enterprise software industry from "human-driven system" to "system self-driven", and from "process orchestration" to "intention-driven".

Hidden behind this trend is a profound industry stratification effect:

• For small and medium-sized enterprises , AI is unlocking software capabilities that were “unaffordable to buy or use in the past”, allowing an entrepreneur to outsource a full set of capabilities in finance, legal affairs, customer service, and operations;

• For large enterprises , AI is fundamentally changing the coupling relationship between "information flow, decision flow, and execution flow," pushing enterprises from a labor-intensive process-driven model to a more lightweight, automated operating system.

• For software vendors , they must make a choice between "tool provider" and "intelligent service provider": whether to continue to maintain traditional license logic, or to actively move towards a new track of AI native and service delivery.

This impact is not just an iteration of underlying technology, but also a redistribution of digital control rights in enterprises : Who can dominate AI agents? Who can define business intent? Who can continuously train the organization's "AI application muscles"? Who will be able to control pricing and platform rights in the next round of software industry reconstruction?

AI transformation: the fateful choice of traditional enterprise software

Although many 2B enterprise software vendors have publicly announced that they are cooperating with mainstream large-scale model platforms such as DeepSeek, and claiming that they are actively entering the AI ​​era, these "AI strategies" are more at the marketing level, and they have not yet established AI products and engineering teams with modeling, deployment and productization capabilities. Most projects are still in the exploratory stage, lacking a clear talent introduction mechanism, technology stack planning and resource investment path.

More companies choose to "wait and see" - on the one hand, they are worried about missing the window bonus brought by AI, and on the other hand, they cannot afford the high cost of trial and error. This state of "slogans first, capabilities lagging behind" makes it difficult for them to seize the initiative in the next round of technological paradigm shift.

The more realistic dilemma is that the frantic competition among Internet giants for AI talent and computing resources is significantly raising the threshold and cost of AI research and development . AI talent is being snatched at several times the premium, and model training resources and inference costs are becoming increasingly tight. For most 2B software companies, it is no longer as difficult to establish competitive AI product capabilities in the short term as it was in the past when they just "made a functional module."

As Tencent, Alibaba, and ByteDance use hundreds of billions of dollars in capital expenditures to build AI moats, traditional enterprise software vendors are facing a life-or-death decision:

• Either actively embrace AI and restructure their own products and platforms (such as SAP launching Copilot for ERP);

• Or retreat to highly compliant vertical industries such as military industry, aerospace, and finance to maintain limited advantages.

  
  

Gartner's prediction may not be an exaggeration: by 2027, 60% of enterprise software functions will be replaced by AI native solutions, and those manufacturers who successfully transform will receive more than 3 times the market premium. Faced with this fierce new round of Internet AI impact, can traditional enterprise software manufacturers still stick to their original intentions and maintain their position?