NVIDIA launched a free tier for AI model hosting at build.nvidia.com, giving access to over 80 language models including Deepseek and GLM through a single API key. The service works as a drop-in replacement for ChatGPT's $20/month subscription and requires no credit card or trial timer.
Never pay for ChatGPT again, cuz NVIDIA just quietly killed the $20 subscription and almost nobody noticed. See, they're hosting over 80 of the world's best AI models on their own servers now. Minimax, KIMI, Deepseek, GLM, GPT-OSS, and letting you use every single one through one free key. No credit card, no trial timer. And because it speaks the same language as OpenAI, It plugs straight into Cursor or Claude Code. So here's the setup. Go to build.nvidia.com, generate your key, paste it in, then pick any model you want. That's it. Everyone else is still burning $20 a month for one. The only people winning are the ones who actually take it.
