Your AI Electricity Bill

You may have heard of phrases like ‘intelligence on tap’ or ‘AI is the new electricity’. There is no doubt AI is more accessible than ever and it’s providing an unprecedented level of capability to people at all levels.

However, much like your electricity bill, it’s getting more expensive. This is mainly due to the compute costs AI companies incur to attract and serve the consumer market. The free and plus tiers they offer run at a loss to grow their user-base and therefore investment.

As a result, power users are getting bitten receiving bills in the $100,000’s in a single day or hitting their monthly limit overnight (I’m looking at you Cursor.

In this article, I want to show you an example of the costs of using the current frontier models at an API level. I will use the example of an email assistant for the average office worker (approx. 100 emails a day, 200 words per email).

OpenAI – GPT-o1 ($15.00/M input, $7.50/M cached, $60.00/M output)

Cost to read emails: $0.39

Cost to summarise and sort: $1.50

Total cost: $1.89 USD/person/day

Anthropic – Claude Opus 4 ($15/M input, $1.50/M cached, $75/M output)

Cost to read emails: $0.38

Cost to summarise and sort: $1.88

Total cost: $2.26 USD/person/day

This puts the price tag for an email assistant at an eye-watering $500,000 USD/year for a 1,000 employee company. Only two months ago, I did the same calculation coming up with $200,000 USD/year, this is how much the prices have risen. But have the frontier models become 150% smarter? No, the power they consume has. With the release of Grok 4 heavy behind a subscription of $300 USD/month Elon himself has said that the heavy model uses multiple Grok 4’s to find the best answer. In other words, an insane amount of compute is required.

What many business leaders are now realising is that it’s a lot cheaper to hire AI Engineers who can optimise cheap “legacy” or open-source models that achieve the same result, but at a hundredth of the price. Not only does it keep costs down, but it also removes platform risk and future proofs from the inevitable price hikes going forward.