FHE is becoming a competition over how cheaply encrypted computation can run at scale. @primus_labs recently reported some promising AutoHoG benchmarks for encrypted execution. The numbers: > Transfer latency: ~90ms > Throughput: 210+ transfers/second > Hardware: 8× RTX 5090 GPUs > Workload: Encrypted 64-bit transfers Zama is taking a different approach with FHEVM, providing encrypted Solidity types, coprocessors and access controls for building confidential smart contracts. Primus is working further down the computation stack, where execution speed and hardware efficiency determine how expensive encrypted operations become. For stablecoins, payments and other financial applications, those costs can add up quickly. A confidential application might have excellent developer infrastructure, but if processing every encrypted transaction requires expensive hardware resources, scaling becomes an economic problem. That’s where Primus has an opportunity. If AutoHoG’s reported performance translates into lower execution costs in production, efficient encrypted computation could serve applications across multiple chains. The benchmarks aren’t directly comparable with Zama’s broader FHEVM architecture, but both are addressing different requirements of confidential finance. Zama is building the tools for encrypted applications. Primus is working on making the computation underneath them cheaper. As adoption grows, the economics of running those applications could determine how far confidential finance can scale.