Shouldn't the cost of Computes be billed much the same as Electricity?

I've been thinking about AI compute in the same way we think about electricity, and I'm curious whether this is ultimately where the industry is heading.

Today, AI companies generally charge in one of two ways: subscriptions or usage-based pricing. But under the hood, AI compute is ultimately constrained by physical resources, especially electricity.

That made me wonder:

Will AI compute eventually be priced more like an energy market?

Electricity prices fluctuate throughout the day depending on demand and available generation. Data centres already have incentives to move workloads between locations or schedule low priority jobs when power is cheaper or when renewable energy is abundant.

Imagine if AI worked similarly.

For example:

- Interactive requests (like chatting with an AI) would always be processed immediately and carry a premium because they require instant availability.

- Background tasks such as model training, fine-tuning, large-scale analysis, or batch image generation could be scheduled when electricity is cheaper, reducing costs for both providers and users.

- Businesses could choose between "real-time compute" and "economy compute" depending on urgency. A lot of wastage occurs because tasks are often assigned before the relevant decisions have even been taken.

In other words, AI compute could become something closer to cloud electricity than simply a software subscription.

I suspect AI companies already optimize some of this internally, but I'm wondering whether it will eventually become visible to customers. Instead of buying a fixed monthly plan, perhaps we'd buy compute that fluctuates in price depending on demand, available GPU capacity, and even regional energy markets.

Do you think AI will eventually be sold as a true utility, with pricing that reflects underlying compute and energy costs? Or do you think subscriptions will remain the dominant model because they're simpler for customers?