It's not clear what the metric to use for benchmarking would be.Possible…however if as Tech has highlighted, if Simon Thorp has been pushing for this benchmarking for years I doubt the recent engagement with Carnegie Mellon is the reason.
Even if you say the best metric is inferences per watt, the other systems have taken many many watts just to be trained. Do you include these watts? Possibly not if you can use the model in many chips.
i.e. the akida based chip will be doing a different job in a very different way: Akida will pick up anomalies that others can't because they haven't been trained on those particular cases.
Without a metric there can be no benchmarking.