If you've been paying attention to AI news over the past year and a half, you've probably heard of DeepSeek, and more recently Qwen or Kimi. Nonprofit and school staff bring them up in almost every class I teach now, usually with some version of the same question: these are free, and people say they're just as good as ChatGPT or Claude, so why doesn't Cochise AI recommend them? It's a fair question, and it deserves a real answer instead of a shrug.
Why They're Getting Attention
The short version: they're free, or close to it, and they're genuinely good. Qwen, built by Alibaba, is the most downloaded open-weight model line in the world. In 2024 it accounted for more than 30 percent of all model downloads on Hugging Face, the platform where AI developers share and build on each other's work, and by August 2025 Qwen-based models made up over 40 percent of all new model variants released there. Moonshot AI's Kimi K3, released July 16, 2026, is a 2.8-trillion-parameter open-weight model that benchmarks close to Anthropic's Claude Fable 5 and OpenAI's GPT-5.6, two of the most capable models available anywhere, and Moonshot released it for anyone to download and run at no cost. For a cash-strapped nonprofit weighing a $20-a-month subscription against a free tool that performs almost as well, the appeal is obvious. It's not hype. These are legitimately strong models.
The Legal Reality Behind the Terms of Service
Here's what that free price tag doesn't advertise. DeepSeek's own privacy policy states plainly that user data, including every question asked, every file uploaded, and every conversation held, is collected, processed, and stored on servers in mainland China. That alone isn't unusual; every AI company stores your data somewhere. What's different is what Chinese law says a company must do with it once it's there.
Article 7 of China's 2017 National Intelligence Law requires that "all organizations and citizens shall support, assist, and cooperate with national intelligence efforts." Combined with China's National Security Law and Personal Information Protection Law, companies operating in China (a category that includes DeepSeek, Alibaba, and Moonshot AI) can be legally compelled to hand data over to the government on request, without a warrant process an independent court can review, and without a right to refuse. That obligation exists whether the person whose data it is lives in Shanghai or Sierra Vista.
Governments Are Already Acting, Not Just Talking
This isn't a hypothetical concern dreamed up by AI consultants. By early 2026, Italy, Australia, Taiwan, and South Korea had banned or sharply restricted DeepSeek across their government sectors, citing exactly this risk. In the United States, the Pentagon, NASA, the U.S. Navy, and the Department of Commerce have all blocked DeepSeek on government devices, and seventeen states, including Texas, New York, Georgia, Florida, Iowa, Tennessee, and Virginia, have done the same on state government systems. A bipartisan bill, the No DeepSeek on Government Devices Act, is currently moving through Congress. None of this amounts to a ban on ordinary personal use in the U.S.; you can still download DeepSeek's app today. But when defense, intelligence, and state government IT departments decide a tool is too risky for their own internal use, that's a strong signal worth taking seriously before handing it your donor list or your students' records.
What This Actually Means for Your Organization
None of this is really about whether DeepSeek, Qwen, or Kimi are good AI models. They are. It's about what happens to your data after you hit send. Nonprofits and schools routinely handle information that carries real legal and ethical weight: donor names and giving histories, grant budgets, board deliberations, student names and records. When you use Claude, ChatGPT, or Copilot, you're covered by a U.S. company's terms of service, and if your organization is on a paid plan, often a formal data processing agreement that limits what the company can do with your data and gives you legal recourse if it doesn't honor that agreement. When you use a Chinese-hosted model, that kind of protection doesn't exist in any comparable form. The company isn't necessarily acting in bad faith; it may simply not have the legal standing to say no if asked.
A Fair Caveat About Open-Weight Models
One nuance worth naming honestly: Qwen and Kimi are released as open-weight models, meaning a sufficiently technical organization could download the model itself and run it entirely on its own computers, never sending a single byte back to China. Moonshot has said as much about Kimi K3. If your organization has the in-house technical capacity to self-host a model that way, the calculus changes, because the risk described above is really a risk about where your data travels, not about the model's country of origin in the abstract. But that kind of setup is well beyond what the vast majority of the nonprofits and schools we work with have the staff or budget to run. For everyone using the hosted app or website version, which is nearly everyone, the risk above applies in full.
Where That Leaves You
If cost is what's drawing you toward DeepSeek, Qwen, or Kimi, that's a real and understandable pressure, and it's worth solving directly rather than routing around it. Claude, ChatGPT, Copilot, and Gemini all offer capable free tiers that cover the large majority of everyday nonprofit and school tasks, and our AI Tools guide walks through what each one is actually good at and what it costs. If your organization is weighing a paid plan and budget is the obstacle, that's exactly the kind of question worth a real conversation rather than a workaround. Use the contact form and let's talk through it.
