The Economics of Neoclouds

The economics of "neoclouds"—specialized GPU-as-a-service providers—hinge on balancing high capital expenditures with operational utilization. While an H100 graphics card can yield a 28% compounded annual growth rate at full utilization, profitability drops significantly below 55% utilization, making broad market stocks a safer alternative. Because the market is heavily commoditized, success depends on efficient scheduling and technical optimization to manage communication bottlenecks inherent in model parallelism. Scaling beyond a few cards introduces non-linear efficiency losses, requiring sophisticated resource management to handle diverse hardware demands. Furthermore, because code is often tied to specific architectures like CUDA or ROCM, providers face risks of underutilization if they diversify hardware. Ultimately, these businesses rely on high-quality supply chain relationships and asset-backed financing, where the residual value of GPUs serves as critical collateral against market volatility.