The next war isn't about oil, it's about land for data centres

Why I started asking where my AI’s power actually comes from

A few months ago I read Amy Webb’s 2026 Convergence Report, and something has been sitting with me ever since. The next major conflict of this century probably won’t be fought over oil. It’ll be fought over land that has two things AI desperately needs: electricity and water. Why? Because that’s what a data centre runs on – and data centres are needed for AI.

I went down a rabbit hole after that. I’d heard China was building data centres underwater. I’d heard Elon Musk wanted to put them in space. Both sounded like science fiction. Neither turned out to be quite what I expected. But what I found instead has changed how I think about my AI usage at work and home each day.

Data centres under the sea and outer space are real

China switched one on near Shanghai this year, a couple of thousand servers sealed inside a structure sitting on the seafloor. It is cooled by the surrounding ocean instead of industrial air conditioning, and powered almost entirely by an offshore wind farm. It cuts land use by more than 90% compared with an equivalent facility on dry ground.

Scientists are watching closely for the catch: heat discharged into the water could, at scale, disturb marine ecosystems in ways nobody’s tracked over a full 25-year lifespan yet. Still, it’s not a thought experiment. It’s switched on, processing real workloads. Only time will tell what the repercussions will be.

Meanwhile, Elon Musk is planning to send data centres into Earth’s orbit. The idea is to soak up endless sunlight, overriding the need for electricity. Crazy, but smart I suppose. SpaceX has filed to launch up to a million of them. The trouble is that the cost of these still outweigh land-based AI data centres and sober estimates suggest it won’t happen till 2040. Meanwhile, actual demand for land-based capacity is on track to roughly quadruple by 2030.

So the real fight today is for land in Memphis and Utah

Once I’d gotten past the insaneness of server racks on the seafloor and satellites in the sky, I couldn’t help but feel a sense of helplessness for today’s problem it’s yet to solve.

Right now, in towns most of us will never visit, conflict is already happening between local communities and those who want to control AI power.

Similar to climate change, the demand for land to build AI data centres is drawing bold lines between the haves and the have-nots. Just as low-lying countries are slowly disappearing under warm seas, low-income regions are becoming victim to the wants and needs of the wealthy.

• Hundreds of residents in Utah are protesting against a data centre proposed next to a shrinking lake, in a region already short on water.
• A community in Memphis already carrying a heavy asthma burden is now living next to dozens of gas turbines installed to power an AI supercomputer.
• Locals are experiencing unexpected illnesses such as nausea, sleep disruption, and tinnitus from the low-frequency humming of cooling systems and generators.
• Some US residents in data-centre-dense parts of the country are saying their electricity bills have climbed by hundreds of per cent over five years because providers require the funding for rising infrastructure costs.

Responsible AI used to mean privacy – now it means power

For the last several years, when people in my industry talked about “responsible AI,” we meant something fairly narrow: don’t paste a client’s confidential brief into a public chatbot, be honest that a human reviewed the final copy, don’t let a tool write in a voice it doesn’t understand.

That conversation still matters. But it’s not the whole conversation anymore.

The newer, harder question is: do you have any idea what’s actually powering the thing you use 40x a day… Whose water it’s drawing on, whose electricity bill it’s quietly nudging upward, whose backyard it sits in?

What happened when I tried to figure out where my AI power was coming from for my own agency

I decided to ask generative-AI about my own AI use. (At Avion, we use the paid versions of ChatGPT and Claude.) Similar to how you can track your energy usage, I thought there’d be a straightforward answer somewhere. But there isn’t one.

Today, nobody can hand you a figure for how much power your specific conversation used, or exactly which grid it came from.

That gap isn’t a failure on my part for not looking hard enough – it’s an honest reflection of how new this topic is. Accountability simply hasn’t caught up to how quickly we’ve all started relying on these tools.

What we actually changed

Instead of producing a number I couldn’t back up, I thought I could at least tidy up some things within my control. We’d had an AI usage policy sitting untouched since 2024. I reviewed and revised it so that:

• we now name the actual tools we use and explain why we chose them
• we made explicit that if anything confidential must go in, it needs to go through a temporary, un-saved chat session, and
• we built in a proper review date, so that the policy is visited more than just ‘regularly’.

If you work with an agency, or run one, the six questions below outline where I’d start in terms of being more mindful about AI.

What I don’t want to lose in all of this

There’s something exciting about data centres being built on the ocean floor and in orbit in the future. But until this happens at scale, we’re experiencing AI technology at the expense of others who are maybe not as lucky.

Today, being a responsible user of AI is more than just having guardrails and an appreciation for how LLMs are created. It’s about being mindful about its impact on a global scale.

The Chinese AI data centre on the seafloor is real proof that we don’t have to fight over land to build these things forever. But the solution doesn’t quite yet match the size of the problem given how quickly we’re needing more and more AI infrastructure.

Responsible AI, in 2026, isn’t just about guardrails on the model or curiosity about how it was trained. It’s about asking the same question of AI that we’d ask of any other industry reshaping the towns around it: “Who’s paying for this, and did they get a say?”

– – –

*I’m talking with digital, content, and marketing leaders at large enterprises about trends, considerations, and the impacts of AI adoption. My goal is to help everyone feel more informed about what’s working, what’s not, and what the ramifications might be for the workforce. If you’d like to share insights, let’s chat. Book a call with me online.