The Ethereum Virtual Machine (EVM) have become a de-facto programming language for the blockchain industry over the past decade. While the progress in EVM optimization on the CPU has been significant, optimization of the EVM on GPUs have been limited. We report the development of GPU-EVM, the world's first GPU-parallelized implementation of the EVM. GPU-EVM harnesses recent dominance of GPGPU computing above GPCPU computing on increasingly complex tasks as we hit the limits of Moore's law. While having higher absolute latency than the most performant CPU-based EVMs, GPU-EVM exhibits magnitudes higher throughput through parallelization across just a single GPU. In particular, GPU-EVM sets a state-of-the-art on EVM throughput, being the highest throughput EVM in the world by 100x, even when accounting for the latency. GPU-EVM achieves this, while improving the unit economics of computation exe/s/\$. This leap in efficiency and speed opens new possibilities for blockchain applications, including AI/RL model training, accelerated Layer 2 solutions, and enhanced security through advanced DeFi simulation and fuzzing techniques.