Google searches for boss to get grip on climate, energy costs of this AI hype cycle

AI and its effect on datacenter energy consumption are a concern at Google’s parent Alphabet, with the search giant seeking a manager to lead on AI/ML Carbon Reduction and Net Zero efforts. The Mountain View-based biz is on the hunt for a program manager for its Google Climate Operations team to develop a long-term roadmap and …

[Charlie Clark]

Possibly good news

If energy costs, even is somewhere like America where energy is very cheap, become a significant cost factor, then they will intensify efforts like this to reduce energy use. Though there will also no doubt be attempts to use arbitrage to go where energy is cheapest.

[Anonymous Coward]

Re: Possibly good news

Energy costs are already a huge chunk of costs, and operators are very aware of that. The impetus here is the concern that the US energy networks may not be able to support the assumed increase in consumption from AI. Personally I think that as we're in the learning stage with "AI", that its not clear whether the vast future growth will occur - where's the value? Even if AI finds the cures for cancer, what then? Better sifting of cat videos, or using their spare cycles to support SETI@home?

Don't forget that they will consider net energy costs, so not just where energy is cheapest. There's hints in the quotes in the article that they hope to dip their snouts into the trough of "ancillary services" to support local or national electricity grids, and done well that can generate significant income streams. I used to work for a company with a sizeable district heating operation and at some sites they made more money from providing local grid support services from their CHP plants in export mode than they did from selling heat to businesses and consumers.

[Charlie Clark]

Re: Possibly good news

I don't think "AI" in general is the worry for networks. But for service providers like Google, they may be keenly aware of the marginal costs or "surge pricing" if they have to buy electricity to support workload spikes. Client-side inference uses a lot less power.

Energy utilities have bigger headaches like boosting generation and network capacity to support the switch to electrification for transport, heating, etc. They have, of course, traditionally profited from selling more power, but that's only once the capital costs of increasing capacity have been amortised.