If you've tried to scale any AI workload recently you already know this: getting reliable GPU access outside of big enterprise contracts is still a nightmare. Spot markets get preempted, hyperscaler pricing is opaque, and smaller teams are basically last in line. Came across a project called Inferra that's taking a genuinely different angle on this. Rather than building another GPU marketplace, they're creating a derivatives exchange — perpetual futures for specific chips (H100, H200, A100, ...