AI Leaders Say We Need to Slow AI Down. Power and Water May Do It First.

Some of the biggest names in AI are now telling us that the technology is becoming so powerful, we may need to slow it down.

Dario Amodei, Sam Altman, Demis Hassabis and Elon Musk rarely agree on much. But this week, all four backed calls to “pace” the development of frontier AI. Whether they will actually slow down is another question.

The irony is that something far more practical may slow AI down first.

Most conversations about AI still focus on what the models can do, how quickly they are improving and who is ahead. But the next big constraint may have very little to do with the models themselves. Where will the electricity come from? How will the infrastructure be cooled? And can these facilities actually get connected, permitted and built?

Until recently, this felt like a technical issue sitting somewhere between engineering, facilities and sustainability. I don’t think it is anymore. It is quickly becoming one of the biggest strategic questions for companies building AI infrastructure, and one that will determine how and where they can scale. That also changes the kind of leaders these businesses need.

The numbers are big, although I think it is easy to lose people by throwing too many of them around.

The simplest one is this: the International Energy Agency expects global electricity consumption from data centres to more than double, from around 415 TWh in 2024 to approximately 945 TWh by 2030. AI is the main driver of that growth.

The impact is already very visible in some markets. In Ireland, data centres consumed 23% of all metered electricity in 2025. Ten years earlier, it was 5%.

Water is more difficult to measure, but no less important. Data centres use water directly for cooling and indirectly through the power plants that generate their electricity. Lawrence Berkeley National Laboratory estimated that US data centres directly consumed 17.4 billion gallons in 2023, with a much larger indirect footprint linked to electricity generation.

There is an important nuance here. Not every data centre uses the same amount of water, and not every cooling system creates the same trade-off. Climate, local grid mix, chip density, cooling design and workload all matter. A facility in Finland is not facing the same choices as one in Arizona or southern Spain.

That is exactly why broad sustainability commitments are no longer enough. These are local operating decisions, with local consequences.

For years, compute was mainly discussed as a question of chips and capital. Increasingly, the scarcer resource is permission: a grid connection, access to water, a viable site and the trust of the surrounding community.

Tucson became a very public example. In 2025, the city council rejected discussions around the Amazon-linked Project Blue development after strong local opposition over water use, electricity demand and a lack of transparency. Amazon later pulled out. The developer continued exploring a revised project, including a less water-intensive cooling design, but by then the damage was done.

This was not simply a sustainability failure. It was a failure to understand stakeholders, communicate early and design the project around the realities of the location.

And it is not an isolated case. Data Center Watch reported that local opposition blocked or delayed $152 billion worth of US data-centre projects during 2025. In the first quarter of 2026 alone, it tracked at least 75 affected projects worth around $130 billion.

Europe is heading in a similar direction, although through regulation as much as community opposition. Spain is currently debating a draft framework that would require large new data centres to cover at least 80% of their electricity consumption with additional renewable generation, matched hour by hour. The proposal may still change, and it has already received strong pushback from the industry. But the direction is quite clear: governments no longer see these facilities as just another large corporate customer.

The European Commission is also working on a common rating scheme for data centres, building on wider EU reporting requirements around energy and water performance.

In other words, building the technology is only half of the job. The company also needs to secure energy, work with utilities, navigate regulation, win local support and make trade-offs that will hold up commercially and publicly.

I was also curious about the companies working on the problem, because this is not only a story about constraint. A new infrastructure layer is emerging around AI.

ZutaCore, for example, has developed a waterless, two-phase direct-to-chip cooling system. Instead of cooling an entire room, it removes heat much closer to the processor using a dielectric fluid. Corintis, a Swiss startup, is working on microfluidic cooling channels integrated into the chip itself and has collaborated with Microsoft on bio-inspired designs. Submer is taking another route with immersion cooling, placing IT equipment in a thermally conductive liquid.

Other companies are looking at what happens to the heat after it leaves the server. UK-based Deep Green installs small data centres in places such as leisure centres and reuses the waste heat for hot water. That model will not replace hyperscale infrastructure, but it shows how differently the problem can be approached.

On the power side, companies such as Crusoe are bringing energy generation and data-centre development closer together. Others are building software that shifts flexible computing workloads towards times and locations where cleaner or cheaper electricity is available.

None of these solutions removes the underlying tension. Liquid cooling can improve thermal efficiency but creates different infrastructure requirements. Air cooling may reduce on-site water use while increasing electricity demand. On-site generation can shorten the wait for a grid connection, but introduces a completely different set of capital, regulatory and operational questions.

So this is not a problem that can simply be handed to one sustainability leader and considered solved. It cuts across product, infrastructure, finance, operations, public affairs and corporate strategy.

This is the part I find most interesting from an executive-search perspective.

The companies closest to the problem are no longer hiring only from technology. They are bringing in leaders from utilities, power generation, industrial infrastructure, renewables, real-estate development and government affairs.

Crusoe appointed John Michael Adams as SVP of Power Infrastructure in 2025. He came with more than 30 years in power generation, including as President and CEO of TexGen. Microsoft hired nuclear-development expertise from the Tennessee Valley Authority. These are not traditional tech profiles, and that is the point.

They bring experience that many fast-growing AI companies have never needed before: negotiating with utilities, understanding interconnection queues, structuring long-term power agreements, developing generation assets, navigating permits and operating infrastructure where failure has very physical consequences.

I would be careful, though, about turning this into a search for a fashionable new title. “Chief Energy Officer” may make sense for some hyperscale businesses. For others, the mandate belongs with the COO, Chief Infrastructure Officer, Chief Development Officer or a genuinely empowered sustainability leader.

The title matters far less than the decision rights.

Can this person influence site selection? Do they own the energy strategy or merely report on it? Are they involved before a location is announced? Can they challenge a design that looks attractive on cost but creates water or permitting risk? Do they have direct access to the CEO and board when commercial speed conflicts with infrastructure reality?

If the answer is no, the company has probably created an advisory role for an operating problem.

This also changes where I would look for talent.

The obvious search universe is other data-centre operators and hyperscalers. Sometimes that will be right. But it is also where every competitor is looking, and it can produce a very narrow shortlist.

The more interesting profiles may sit one or two industries away:

Of course, bringing someone from an adjacent industry creates its own risk. A brilliant utility executive will not automatically thrive inside a company moving at startup speed. Someone from a large infrastructure group may be used to much more capital, longer planning cycles and a completely different level of organisational support.

That is where the assessment needs to go beyond sector labels. Has this person built something without a mature playbook? Can they make decisions with incomplete information? Can they explain a difficult trade-off to engineers, investors and a city council without speaking three different languages? Can they move quickly without pretending that permitting and grid capacity can be “disrupted” away?

The strongest profile is not simply an energy expert joining tech. It is a translator and builder who can connect two worlds.

For boards and CEOs, I think there are a few questions worth asking before the constraint becomes a crisis:

These may sound like infrastructure questions. In reality, they determine growth, capital allocation and time to market.

I don’t think AI’s energy and water problem means the sector cannot keep growing. There are too many smart companies working on cooling, energy, heat reuse and efficiency for that to be the conclusion.

But technology alone will not decide which companies scale successfully. Leadership will.

The businesses that move well through the next phase will be the ones that treat power, water, permitting and community support as part of their core operating model, not as a sustainability appendix. They will also be willing to hire beyond the usual AI talent pool and give those leaders genuine authority.

The next bottleneck in AI may not be the model or even the chip. It may be whether the people around the table understand what it takes to build the physical system underneath it.

And for boards, founders and executive-search firms, that is becoming a very real hiring question.