Moving ‘per-user’ computation to the cloud feels like a net-negative for energy consumption. Is the reduction in on-device resource usage greater than the increase in data centre usage?

When coupled with the transfer of wealth not to the service, but the infrastructure, I wonder if we’ve got this all the wrong way around. Not only does local compute give power to the user — what happens on your local device, stays on your local device — but it also distributes the power consumption more broadly.

Are SaaS vendors moving calculations to the client today, and only storing the results in cloud storage, off-loading work to the clients? Does this end up trading computation for network egress costs?

Maybe I just don’t understand the aggregate power usage patterns?