Dataflow-Oriented Classification and Performance Analysis of GPU-Accelerated Homomorphic Encryption

This project is a research artifact accompanying an arXiv paper. With 0 stars and 4 forks (likely researcher mirrors), it lacks any commercial or community moat. Its primary value is as a taxonomy for how different CKKS parameters affect GPU memory and compute dataflows. While technically sophisticated, it is highly reproducible for any team working on FHE acceleration. The 'defensibility' is low because the findings are destined to be absorbed into larger, more established libraries like OpenFHE or NVIDIA's internal cuFHE efforts. Frontier labs like OpenAI or Anthropic are unlikely to build this directly, as they focus on high-level privacy-preserving ML rather than low-level FHE kernel optimization. The primary risk is 'platform domination' by NVIDIA; if NVIDIA releases a specialized 'cuPHE' library or further optimizes their Tensor Cores for modular arithmetic (NTT/INTT), academic analyses of current GPU dataflows become obsolete. The 1-2 year displacement horizon reflects the rapid pace of FHE hardware acceleration (ASICs from companies like ChainReaction or Cornami) which will likely leapfrog GPU-based implementations for production workloads.