Models Are No Longer a Limited Resource

On April 10, Anthropic unveiled the public beta of Claude for Word, marking full integration with the Office suite. From an investment standpoint, this milestone isn’t just about Claude securing another entry point—it signifies a new era in AI office competition, where the focus shifts from model capability showdowns to a contest over scenario understanding and ecosystem barriers.

For the past two years, the AI office landscape has been straightforward: stronger models equal smarter tools. But this narrative is quickly evolving. Top labs now update their models every few months, with performance gaps closing rapidly. Claude, GPT-4o, Gemini, and DeepSeek now exhibit minimal differences in general capabilities. For enterprise users, models have a remarkably short “shelf life”—a model leading the pack today might be average in just three months.

Models are fast becoming a foundational element, much like infrastructure. When computing power and models are abundant, where does true enterprise competitiveness lie? The answer is clear: data governance and scenario understanding. Models are essential but insufficient on their own. Model capabilities set the baseline for AI office tools, while scenario data accumulation defines their upper limits.

The Data Flywheel Powered by 80 Million Monthly Active Users

By the end of 2025, WPS AI’s domestic monthly active users had surpassed 80.13 million, a 307% year-on-year increase, with enterprise users making up 42% of the total. Meanwhile, WPS Office’s global monthly active devices reached 678 million.

The true value of these numbers lies in the data flywheel being built by 80 million users. Every day, users draft documents, review contracts, create PPTs, and revise official documents with AI, generating high-quality Chinese office scenario data with each interaction. This data, in turn, continuously refines AI performance in Chinese contexts, enhancing the product experience.

Once the data flywheel gains momentum, the cost for newcomers to catch up rises exponentially. Claude can train models to understand Chinese, but it can’t replicate the real-world usage scenarios of 80 million Chinese users. This scale barrier is time-sensitive—it’s not that model capabilities can’t catch up, but that scenario data accumulation can’t be rushed. Take contract review, for instance: WPS AI’s ability to spot local legal risks like “excessive penalties,” “overly broad non-compete clauses,” and “probation periods exceeding legal limits” is rooted in long-term learning from vast Chinese contract corpora and the legal system, something general large models can’t achieve through short-term training.

Data Security: The Gateway to the Enterprise Market

In the consumer market, users might choose AI tools based on a single criterion: usability. But in the enterprise market, there’s a non-negotiable line: data security.

Claude for Word highlights “operating within enterprise security frameworks” and supports gateways like Amazon Bedrock, Google Vertex AI, and Microsoft Azure. This setup works in the U.S. and European markets, where large enterprises can manage data flows through these channels. However, the Chinese market demands stricter controls: data must not leave the country.

Industries like finance, government, energy, and telecommunications face stringent data security regulations. No matter how useful an AI office tool is, if it requires sending data overseas, it will fail compliance checks. This isn’t about enterprise preference—it’s a regulatory requirement. The 2023 Interim Measures for the Administration of Generative Artificial Intelligence Services explicitly mandate providers to prevent data security risks and prohibit illegal handling of personal information. For overseas AI tools, data export itself is an insurmountable compliance hurdle.

WPS AI supports private deployment, letting enterprises run AI capabilities on their own servers with data staying entirely within internal networks. Meanwhile, WPS is fully compatible with domestic IT innovations, passing compatibility tests with mainstream Chinese operating systems, databases, and middleware. In the enterprise market, AI office tools lacking private deployment and domestic innovation compatibility aren’t even in the running. While Claude’s model might excel in certain areas, it faces structural barriers at the infrastructure level in the Chinese enterprise market.

Pricing Strategies: The Limits and Potential of User Growth

Pricing is a strategic lever for enterprise SaaS products. Claude for Word is priced at $20 per month for Pro users (~145 RMB) and $100 per month for Max users (~725 RMB), requiring a prior Microsoft 365 subscription. This means a Chinese user must spend at least 200 RMB per month (including Microsoft 365 fees) to fully experience Claude’s Office collaboration features, with higher costs for enterprise users.

WPS AI, offered as a value-added service for WPS members, has a lower price barrier. Behind this pricing difference lie two distinct commercial approaches. Claude pursues a premium route, using price to filter users, achieving higher average revenue per user (ARPU) but facing a clear limit on user growth. WPS follows an inclusive route, lowering usage barriers with affordable pricing to expand its user base, then leveraging scale effects to build data flywheels and ecosystem barriers.

From an investment perspective, both approaches carry risks. The ARPU advantage of the premium route may erode as Claude’s own model iterations commoditize capabilities, reducing user willingness to pay. The inclusive route risks commercialization efficiency, as only a fraction of 80 million monthly active users convert to paying customers. However, the inclusive route has a long-term asset the premium route can’t match: the scale effect of scenario data. More users mean broader scenario coverage, higher-quality training data, and stronger product defenses. This flywheel effect grows stronger over time, not weaker.

The Final Verdict: Scenario Barriers Trump Model Capabilities

So, what truly defines the core competitiveness of AI office tools?

Model capabilities are essential but not enough. An AI skilled at reviewing English contracts might struggle with local legal risks in Chinese contracts; an AI adept at navigating Office apps might not grasp the formatting norms of government documents; an enterprise architecture secure on international cloud platforms might fail China’s domestic innovation compliance reviews.

These gaps can’t be quantified like model parameters or compared through benchmark rankings, but they directly determine a product’s usability and market penetration. With over three decades of local expertise, WPS has built three layers of defense in this sector: language and knowledge barriers—accumulated through massive Chinese contract corpora, legal knowledge bases, and official document norms; compliance and security barriers—private deployment capabilities and domestic innovation compatibility certifications; and scale and data barriers—the data flywheel powered by 80 million monthly active users.

The common thread among these three barriers is that they can’t be overcome through model upgrades alone—they require time and scenario-specific accumulation. Claude’s integration with Word is a symbolic milestone in the globalization of AI office tools. But in the Chinese market, the competition isn’t about “whose model is smarter”—it’s about “who better understands Chinese users’ office scenarios.” In this light, the ultimate outcome of AI office competition hinges not on model parameters but on scenario barriers.