OpenAI and Anthropic are throwing money at this market.

Whether you're a parent or a teacher, in the past two years, you've undoubtedly faced this scenario: students take photos of their homework problems with their phones and send them directly to ChatGPT or Doubao, where the solution and answer appear instantly—easy and enjoyable. AI companies have clearly recognized the potential of the education market long ago, and thus, amidst fierce competition, they've also dedicated resources to entering this market. However, unlike Chinese AI companies that directly target students and parents, American AI companies prioritize capturing the teacher population. On July 14th, Anthropic released Claude for Teachers, offering the Pro-level Claude to K-12 teachers across the US for free. The news itself isn't exactly groundbreaking—after all, OpenAI launched ChatGPT for Teachers last November, and Google had already integrated Gemini into its education-focused Workspace early on. Over the past year, almost all major AI companies have been adding products to the education sector. OpenAI is offering free services to teachers and university students; Google is directly covering school districts through its Workspace ecosystem; Khan Academy's Khanmigo is funded by Microsoft, allowing teachers across the US to use it at zero cost. Why are top AI companies using the same "free" strategy as the internet in its early days to conquer schools? What are their calculations behind this land grab? Perhaps we can find some answers in Anthropic's approach. Anthropic's "Infrastructure" Approach What's most noteworthy about Claude for Teachers isn't its "free" nature, but rather its underlying technological architecture. Anthropic integrates with a curriculum knowledge graph called Learning Commons. This system, developed by the Chan Zuckerberg Foundation, covers academic standards from all 50 US states, and it's not just about the standards themselves—it also includes the sub-skills under each standard, as well as the typical learning sequence for students.

Claude for teachers registration page｜Image source: Anthropic

This means that when a fifth-grade math teacher in Texas asks Claude for help with lesson preparation, Claude is not "guessing" what to teach using general knowledge, but rather following the Texas curriculum standards, specifying exactly how to teach this grade, this semester, and this knowledge point.

Google wanted to use Chromebooks to seize the student market to pave the way for the future | Image source: GoGuardian

Around 2012, Google pushed Chromebooks to schools at extremely low prices, along with free Google Classroom and Google Docs.

By 2020, over 60% of devices in US K-12 schools were Chromebooks, and an entire generation of students grew up using the Google suite. After they graduated and joined companies, Google Workspace required almost no promotion. While Chromebooks have consistently underperformed in the mainstream consumer market, they have indeed provided a pool of potential users for Google's services and software. Today's AI companies are doing exactly the same thing. Students use Claude to write papers, lab reports, and prepare for interviews; the first thing they do after graduating and joining a company is renew their Claude subscription. Teachers use ChatGPT for three years to prepare lessons, grade assignments, and write comments; the cost of switching platforms becomes increasingly high. Free services are not a cost; they are customer acquisition. Drew Bent, head of education at Anthropic, put it restrainedly, but his meaning was clear: Anthropic's goal is to "collaborate with universities to explore AI-enabled teaching models." In other words, it's about making AI a part of the teaching process, not a tool for students to secretly use as a cheat. This also explains why several companies are emphasizing "not using teacher data to train models" and "compliance with FERPA student privacy regulations"—they need to allay the concerns of school administrators to actually get on the procurement list. Anthropic has prepared a K-12 data processing protocol specifically for Claude for Teachers and launched a formal pilot evaluation in Detroit public schools, collaborating with the Gates Foundation to track teacher experience and teaching effectiveness. Turning our attention back to China, an interesting comparison emerges. China is not lacking in actions related to AI education. ByteDance promoted "Doubao Aixue" to the core entry point of the Doubao App in January 2026, Alibaba's Qianwen launched "One-Click Exam Paper Search," and TAL Education Group's Jiuzhang Big Model showcased "Jiuzhang Lobster" for teachers and "Xiaojinglong" for students at the World Artificial Intelligence Conference in July. QuestMobile data shows that as of the third quarter of 2025, the monthly active users of AI education applications in China had exceeded 120 million. However, a closer look reveals that AI education in China and the US is following two almost completely different paths. The American companies' approach is to "empower teachers." Anthropic creates a curriculum knowledge graph, OpenAI runs a teacher training academy, and Google integrates teaching workflows—their common characteristic is that they don't touch the content itself, but rather provide teachers with tools, allowing them to decide how to use AI. This is a platform logic, with the core being to become the infrastructure of the education system. The Chinese companies' approach is to "directly reach students." Xueersi relies on its 20-year-accumulated question bank to build a large-scale vertical model, Doubao relies on free traffic as a general entry point, and Zuoyebang and Yuanfudao each have their own AI-powered problem-solving products. Their common characteristic is that they revolve around "doing practice questions," with paid conversion as the core. TAL Education Group has gone further—developing its own learning machines, large-scale models, and content, creating a closed ecosystem. Both paths have their merits. The US education system is decentralized, with different curriculum standards in each state, giving teachers significant autonomy in instruction; therefore, "empowering teachers" is a reasonable entry point. China's education system revolves around standardized testing, and the core need of students and parents is to improve scores; therefore, "helping students do practice questions" is a more direct business logic. However, a reality that cannot be ignored is that while all "AI education products" are focused on problem-solving tools, what students actually use on a large scale are general-purpose models like Doubao. This isn't because Doubao has particularly good educational features, but because it's free, sufficient, and can ask any question. Specialized AI education products, on the other hand, find themselves in an awkward position—they're not as fast at problem-solving as general-purpose models, and their teaching isn't as in-depth as that of human teachers. American companies choose to focus on the teacher side, building open infrastructure, while Chinese companies choose to focus on the student side, creating closed, paid products. Which path will succeed remains to be seen until this generation of "AI natives" grows up. But at least one thing is clear: the entry of AI into education is not a technological issue, but a power issue concerning "who defines teaching and learning." Different answers will lead to completely different educational outcomes.