Your AI strategy is someone else's problem

Your AI strategy is someone else's problem

In eight weeks studying AI transformation with senior leaders around the world, I've yet to hear a single conversation about what it will cost someone else.

That gap is more dangerous than most boardrooms realise.

Training the model behind ChatGPT evaporated 700,000 litres of freshwater. By 2027, global AI water demand is projected to be 4.2 to 6.6 billion cubic metres, more than Norway's entire annual freshwater consumption. In Q2 2025 alone, community opposition blocked or delayed $98 billion worth of US data centre projects.

One project makes the point. SunCable plans to build one of the world's largest solar farms in Australia's Northern Territory to power AI data centres in Singapore. The land agreement with Traditional Owners took six years and more than 70 consultations to seal. A Charles Darwin University study found only half of Australians consider it fair to generate energy on their land and export it abroad. Was that a choice they were ever clearly offered?

AI learns from the data it trains on. If that data reflects wealthy, digitally-connected economies, its judgment will too, everywhere it operates. In Kenya, a 10-algorithm audit found female entrepreneurs face a 37% underfunding penalty despite repayment records equal to or stronger than male borrowers. AI inherits bias, processes it at speed, and delivers it at scale.

The IMF estimates 60% of jobs in advanced economies face AI disruption, against 40% in emerging markets. The real danger is that gains flow to economies already equipped to use them, while those most exposed to displacement are least equipped to respond. They are not in offices in London, Tokyo or New York.

We are automating the work that builds professional judgment. The contract reviews, the first drafts, the analysis that felt like grunt work but was the apprenticeship. The junior professionals we automate away today are the senior leaders we will not have in a decade. A human reviewer who can't tell whether the output is right isn't governance. It's ass covering.

Paul Polman spent a decade arguing purpose and profit were the same thing. Many came around, then went back when margins tightened. In the 1960s, motorways were driven through communities across Britain because the economic case was overwhelming. The value accrued broadly. The cost fell on specific streets, specific families, specific places. We now see that as a failure of judgment. Is there a danger AI, at its current pace, risks the same?

Ask of your AI strategy: who captures the benefit, who carries the cost, and do those people have any genuine say? Not because regulation demands it, though it will. But because the organisations that think this through now will be far harder to compete with in future.

Your AI strategy is a social contract. Have you thought it through from all sides?