Would the AI “GW-scale” clusters be able to run the Top500 benchmarks meaningfully? And what might be the outcome?