Waiting for AI: on the cost of a decision not taken

A convenient excuse

In conversations about investment in energy efficiency (and in other industries too), I increasingly hear: "let's wait — we will soon do it ourselves with AI and it will be cheaper". It is a convenient argument. It sounds modern, it gives a sense of independence, and it has one more significant advantage: it lets you do nothing today.

To be clear — I am not dismissing AI. We use it ourselves and we see places where it delivers very quickly. In many areas it really will change the way we work, perhaps more than we assume today. It is just that this appealing story usually leaves out three less spectacular details.

Cost one: compute

Systems that deliver AI consume computing power and energy on a scale that is hard to compare with anything in history, and sooner or later those outlays will reach the customer who wants or needs to use the tools. OpenAI alone increased its inference costs from USD 3.76 billion in 2024 to over USD 5 billion in the first half of 2025. Assuming that everything will get cheaper, and that the free tiers will be enough for serious applications, requires a fairly romantic approach to economics.

Cost two: capital

The industry's inflationary growth model has attracted an enormous wave of money. Bain estimates that by 2030 this industry needs roughly USD 2 trillion in annual revenue to balance, while realistic forecasts point to a gap of some USD 800 billion. Nobody puts up sums like that without expecting a return. So first we will hear that AI solves every problem, and then — once we have started using it — we will be burning through notional tokens. Plenty of tokens. Preferably on subscription.

Cost three: procrastination

The third cost is the most interesting one, because it is unlikely to appear on any invoice. It is the cost of procrastination. According to a 2025 MIT study, around 95 percent of corporate generative AI deployments produced no measurable effect on the P&L — so simply adding AI guarantees nothing, and neither does waiting for it.

Postponing a decision is comfortable. It relieves you of thinking, of responsibility for the choice and of the risk of getting it wrong. It requires only one thing: that nobody calculates what the decision not taken costs.

In energy efficiency, every month counts

Here it matters particularly. Every month of delay means bills paid on the old model, no optimisation, no data, no control and none of the experience that a competitor may be acquiring right now.

Some of our customers reached payback long ago — not necessarily using AI methods, but they won. Waiting usually has a price too; it is just that nobody (fortunately) invoices for it.