Choosing AI inference hardware involves complex tradeoffs between cost, latency, and throughput that are hard to predict.
Build a database of benchmark results across CPU, GPU, and specialized AI accelerators. Let users simulate their expected workloads.
Monetize through referral fees from cloud providers and enterprise subscriptions for custom benchmarking.
Start with a simple comparison table of common models on major hardware, then add simulation features.
Risk is keeping up with rapidly evolving hardware - requires constant updates to stay relevant.