SemiAnalysis ClusterMAX 2.0: CoreWeave leads, FluidStack debuts gold, AWS networking is a debugging nightmare
Nov 10, 2025 with Jordan Nanos
Key Points
- FluidStack debuts at Gold tier in SemiAnalysis ClusterMAX 2.0 by aggregating lower-grade GPUs and repackaging them for frontier AI labs.
- AWS networking infrastructure represents less than 1% of cluster cost but consumes over half of user debugging time, exposing a structural friction problem.
- The disconnect between AWS's minimal cost contribution and outsized operational drag suggests pricing mechanisms are not addressing the core infrastructure problem.
Summary
FluidStack enters the SemiAnalysis ClusterMAX 2.0 rankings at Gold tier, a notable debut for a provider that aggregates lower-grade GPU capacity and repackages it into usable infrastructure for frontier AI labs. The model is essentially arbitrage on hardware quality, taking what ClusterMAX classifies as Bronze or worse-tier GPUs and making them viable for demanding workloads.
AWS lands in a different conversation entirely, and not a flattering one. Its networking infrastructure accounts for less than 1% of total cluster cost of ownership by SemiAnalysis calculations, yet users report spending more than half of their debugging time wrestling with AWS's custom networking stack. That ratio, minimal cost contribution against disproportionate operational drag, points to a structural friction problem that pricing alone does not explain.