Building AI models yourself vs relying on LLMs

It's launch day for my team, and I thought it'd be a good time to bring up an AI question that's been biting at us for the last few months. My startup's been in the language processing space for years now, and so with the advent of so many large language models recently, everyone has been quick to jump on the "trillions of parameters" train without much talk of the consequences. Our new platform helps users avoid reliance on GAFAs or the sometimes shady OpenAI for artificial intelligence, a concept that should be top-of-mind considering media buzz about security and AI's GDPR issues, not to mention the environment impacts of these humongous models. Have any members of the PH community tried building AI models locally? Is it worth it? What has been your experience?