ZK Proving's Compute Costs Skyrocket Amid AI-GPU Competition

ZK proving is facing an unexpected challenge as its compute costs rise due to intense competition with AI data centers for Nvidia GPUs. This has led companies like Cysic to rethink their approach and explore alternative hardware paths, such as Field-Programmable Gate Arrays (FPGAs) and Zero-Knowledge Specific Application-Specific Integrated Circuits (ZK ASICs).

Cysic CEO Leo Fan explained that the problem is not a lack of computing capacity but rather an architectural mismatch between zkVM software and the accelerators used to generate proofs. This inefficiency leaves part of the available GPU capacity unused, resulting in higher costs for each proof.

Cysic's Venus proving engine has exposed this issue by reducing the time spent coordinating work between CPUs and GPUs, leading to a 9% performance gain without replacing the underlying hardware. The company is now working on multiple backends, including a GPU-optimized version and an FPGA acceleration backend that targets AMD UltraScale+ devices.