As healthcare systems move toward real-time data-driven operations, traditional diagnosis coding manual, retrospective, and delayed has become a bottleneck. Unlocking timely clinical insights requires new solutions that are both scalable and secure. To address this, we explored the use of generative AI to automate diagnosis classification directly from clinical notes, without relying on pre-coded data or compromising patient privacy. Using synthetic clinical notes generated with ChatGPT, we benchmarked multiple open-source large language models (LLMs) on a cloud platform, assessing both their accuracy in generating ICD-10-CA codes and their infrastructure performance. This approach offers a privacy-preserving, scalable path toward real-time diagnosis extraction, positioning LLMs as a strategic enabler for next-generation healthcare analytics.