A thorough understanding of how best to use AI is about to become really important

Monday, 15 June 2026

The advent of cloud computing taught businesses an important lesson that for some of them was painful. “AI is the same dynamic, but faster and less forgiving,” Ben Kepes warns.

Ben Kepes is a Canterbury-based entrepreneur and professional board member. He is a regular opinion contributor.

OPINION: I've spent 20 years working in a fairly arcane area, only to watch it become mainstream.

When I started out as an industry analyst covering cloud computing, the whole idea of cloud was a mystery to most people. I still remember being introduced to a notable New Zealand parliamentarian as a global cloud computing analyst. He looked at me with complete sincerity and asked whether I studied meteorology.

Two decades later pretty much every organisation under the sun uses the cloud. All those early years spent explaining it are now unnecessary. The arcane became obvious.

After spending a few days recently in San Diego talking about IT spend management and visibility, I am again reminded that what I do is going to become very important, very soon. And this time, the stakes are considerably higher.

There is a subtext worth naming. Three organisations, OpenAI, Anthropic and SpaceX, are in the process of, or already have, becoming publicly listed, each carrying valuations above a trillion dollars. That is a staggering number, and they will need to do something substantial to justify it. More AI products. More AI services. More AI consumption, at scale, by businesses just like yours. Which is exactly where the two threads of this story come together.

Here is the thing about AI most business leaders haven’'t fully reckoned with yet. It is not a software licence. It is not a flat monthly fee. It is a consumption model, and the meter is always running.

Every time an employee uses an AI tool to draft a proposal, summarise a meeting, generate a report or answer a customer query, that interaction costs money. Every time an automated AI process runs in the background of your business, checking inventory, flagging anomalies, responding to inbound requests, that costs money too. Individually, these costs might be trivial. Aggregated across an organisation, across hundreds of employees and dozens of automated workflows running around the clock, they are anything but.

The trillion-dollar initial initial public offering valuations of OpenAI, Anthropic and others are, in part, a bet “that AI spend becomes one of the largest lines on the enterprise technology budget within the next few years”, Ben Kepes writes.

The cloud taught us this lesson, and many organisations learned it the hard way. In the early years of cloud computing, the ease and speed of spinning up new infrastructure felt like pure liberation. Need a new server? Done in minutes. Need 10? Same answer. The bill arrived later, and for many organisations it was a shock. Unmanaged cloud sprawl became a genuine financial problem, with companies routinely discovering they were spending two or three times what they thought on infrastructure nobody had properly tracked.

AI is the same dynamic, but faster and less forgiving. The difference is that cloud costs were at least somewhat predictable once you understood the model. AI costs are tied directly to how your people and systems use the technology, and that behaviour is highly variable and very difficult to forecast without deliberate instrumentation. A poorly designed AI workflow, one that calls a model repeatedly when it doesn't need to, or processes far more data than necessary, doesn’t cost a little more than a well-designed one. It can cost orders of magnitude more, and you won’t know until the invoice lands.

Goldman Sachs projects global AI consumption will multiply roughly 24 times between now and 2030. The hyperscalers and AI providers are building their businesses on the assumption that AI spend becomes one of the largest lines on the enterprise technology budget within the next few years. The trillion-dollar IPO valuations of OpenAI, Anthropic and others are, in part, a bet on exactly that outcome. Someone has to pay for those valuations. That someone is you.

None of this is an argument against using AI. The productivity case is real, and the competitive pressure to adopt it is already significant. The argument is simply that adoption without visibility is a trap, and it is a trap that businesses have walked into before.

The organisations that are going to come out of the next five years in good shape are not necessarily the ones that use the most AI. They are the ones who know what they are spending, understand what they are getting for it, and can make rational decisions about where AI investment is generating genuine returns versus where it is simply running up a tab.

That requires treating AI the way mature organisations treat any significant operational cost. With ownership, with accountability, and with someone asking the uncomfortable question of whether the bill is justified by the outcome. It is, funnily enough, just the sort of thing industry analysts spend their time talking to customers about.

I've seen this movie before. In the early cloud years, the organisations that took cost visibility seriously were the ones that scaled efficiently, while others quietly haemorrhaged budget on infrastructure nobody was watching. The AI reckoning is coming on a faster timeline and at a larger scale.

This time, nobody is going to mistake the question for meteorology.