Every major AI company is now shipping a free product aimed at schools. OpenAI put ChatGPT for Teachers out last November, free for verified US teachers through June 2027 — a date that has since quietly moved to June 2028. Anthropic launched Claude for Teachers this month, free for verified US K-12 educators through June 30, 2027. Google keeps folding Gemini deeper into its education stack and has been handing out AI Pro access to students. Microsoft runs Copilot through M365 Education and funds training through Elevate.
Free access for teachers stopped being a differentiator months ago. It’s the entry fee.
The expiry dates are the part worth staring at. A genuine philanthropic commitment doesn’t come with a renewal date attached. What these are is a paid trial with the invoice deferred — long enough for a teacher to build a year or two of lesson planning around one tool, long enough for a district to write procurement policy around it, and then a conversation happens. Nobody involved is being dishonest about this. It’s just customer acquisition with a very long payback assumption, and the assumption is that switching costs will have grown teeth by 2027.
The Target Moved From Students to Teachers
The first wave, back in early 2025, went straight at students. Anthropic shipped Claude for Education, OpenAI made ChatGPT Plus free for US and Canadian college students, Google offered AI Pro across several countries. Direct-to-student, timed around exams, aimed at habit formation before graduation.
That approach has an obvious ceiling. Students are transient, they don’t hold budgets, and a tool students adopt on their own gets treated by institutions as a cheating vector rather than an asset. Teachers are the opposite on all three counts. They stay, they influence procurement, and their adoption confers legitimacy that student adoption actively undermines.
So the pitch changed shape. Anthropic’s version leans on alignment to academic standards across all fifty states, lesson planning, differentiated materials. It also comes with commitments schools have been demanding for three years: conversations in verified teacher accounts stay out of model training, plus a K-12 data processing addendum. OpenAI has made comparable educator privacy commitments. Those clauses are not marketing garnish — they’re the specific thing that unblocks a district legal review, and they exist because someone worked out that the binding constraint on school adoption was never product quality.
The Edtech Partners Are Distribution, and Distribution Is a Precarious Job
Claude for Teachers launched alongside ASSISTments, Brisk Teaching, Canva Education, Coteach, Diffit, Eedi, MagicSchool, Snorkl and TeachFX. Anthropic also says it’s open-sourcing the underlying teaching skills and publishing how they were evaluated.
Read that list again with a cold eye. Several of those companies built their businesses on doing exactly what a frontier model now does natively — generating lesson plans, differentiating reading passages, producing classroom materials from a prompt. They are being brought in as channel partners to reach teachers who already trust them. That trust is the asset being borrowed.
What happens after the borrowing is the question nobody in the announcements addresses. An edtech startup whose product is a well-designed wrapper around text generation, partnering with the company that supplies the generation, is in a structurally awkward position. It gains reach now and hands over the customer relationship in the process. Some will end up as durable interface layers — grading workflows, student data integration, the unglamorous plumbing models don’t do. Others are being used as a distribution rail and will find out what that cost when renewal season arrives.
Open-sourcing the teaching skills cuts both ways too. It’s a real credibility move, and it also commoditises the exact layer the partners sell.
What the Money Actually Looks Like
There’s a cooperative layer running underneath the competition. The American Federation of Teachers’ National Academy for AI Instruction was launched with $23 million from Microsoft, OpenAI and Anthropic together. Rivals co-funding the training that shapes how a generation of teachers thinks about these tools is not a contradiction — it’s a shared interest in the category existing at all, and in the terms of the conversation being set by vendors rather than by regulators or unions acting alone.
K-12 has never been a good direct revenue market. Budgets are thin, cycles are slow, procurement is political, and the willingness to pay per seat is a fraction of enterprise. Nobody is doing this for education revenue in 2027. They’re doing it because a teacher who plans with one model is a household that subscribes to it, a student who learns on it defaults to it at university, and a graduate who has used it for six years brings it into whatever company hires them. The value is realised somewhere else entirely, which is also why the free tier can be so generous.
The thing I’d watch is whether any of those expiry dates gets extended a second time. OpenAI’s already slipped a year. If they all keep sliding, it means nobody has found a price schools will actually pay, and free becomes permanent by default — which would be genuinely good for schools and a fairly grim signal about the education market as a business. If the dates hold and pricing lands in 2027, the interesting number won’t be the per-seat cost. It’ll be how many districts discover they’ve built a year of curriculum on a tool they can’t leave.