The AI Search Power Problem

Everyone in SEO is talking about AIO, GEO, conversational search, the death of the blue link, and the pivot to generative answers. The consensus is that AI search is the future and the transition is already underway. We agree. At Six Agency Group, our position is simple: we go where the user goes. Always have. When Google launched local packs, we followed. When featured snippets changed how answers are served, we adapted. When AI Overviews started eating the top of the SERP, we started optimising for them. That will not change.

But here is the bit that is not making the conference keynotes: the future of search runs on electricity. A lot of it. And right now, we are facing the AI search power problem.

The Numbers Are Hard to Ignore

In 2024, global data centres consumed around 415 terawatt hours of electricity. By the end of 2025, that figure had climbed to 485 TWh. A 17% jump in a single year. AI-focused data centres specifically saw their electricity consumption surge 50% in 2025 alone.

The International Energy Agency projects data centre consumption could hit 945 TWh by 2030 and 1,200 TWh by 2035. That is roughly double the current global usage. If data centres were a country, they would already rank among the top five energy consumers on earth, sitting between Japan and Russia.

Virginia uses 26% of its entire state electricity supply to power data centres. Ireland is at 21% nationally, potentially 32% by 2026. These are not marginal numbers. They are infrastructure-defining ones.

“I think the limiting factor for AI deployment is fundamentally electrical power.” Elon Musk, World Economic Forum, Davos, January 2026

Musk said it plainly at Davos. Not chips. Not talent. Not regulation. Electricity. He also flagged China’s advantage directly: its energy grid is being built at a scale the US cannot currently match, and that gap translates into an AI capability gap. The country that wins the energy race, his argument goes, wins the AI race. Nobody in the room disagreed.

Then Came March 2026

The Iran conflict changed the conversation fast. Iranian drone strikes on AWS facilities in the UAE and Bahrain in March 2026 marked a genuine first: private data centres came under deliberate military attack. Cloud services went down. Banking,
payments, enterprise systems. Not a theoretical risk. An actual one.

THE INTERCEPT, MARCH 2026
“For the first time in military history, private sector data centres came under deliberate attack. The physical infrastructure of the cloud is now considered a viable military target, forcing a complete reassessment of site selection, security protocols, and investment strategies.”

But the energy story goes deeper than missiles. QatarEnergy was forced to shut its Ras Laffan facility. If you do not know that name, you should. It is the world’s largest LNG plant. Its shutdown cascaded immediately. European natural gas prices surged
approximately 70%. And then came the helium problem, which almost nobody outside chip manufacturing circles saw coming.

The Helium Nobody Talks About

Qatar and Iran, between them, account for around 30% of the world’s commercial helium supply. Helium is not a luxury in semiconductor manufacturing. It is essential. Chipmakers use it to cool silicon wafers during the etching process. There is no viable industrial substitute. None.

“The AI economy runs on helium. The Iran war just created a $650 billion problem.” Fortune, April 2026

US gas distributor Airgas declared force majeure on helium shipments on 17 March, cutting customer deliveries by up to 50%. HP, Dell, Lenovo, Acer and ASUS have already warned enterprise clients of 15-20% price hikes. Experts are now estimating a minimum two to three month shutdown of helium production and four to six months before supply chains return to anything close to normal.

The chips needed to run the next generation of AI search infrastructure are going to cost more and take
longer to arrive. That is not spin. It is the supply chain.

You Felt It If You Use These Tools

This is not abstract. Here is what has happened to the AI platforms that were supposed to replace traditional search results.

These are not blips. They are signals. The tools meant to replace traditional search results cannot stay online consistently right now. That matters when agencies are telling clients to pivot their entire content strategy toward AI-driven discovery.

There is an interesting footnote in the Grok story. In February 2026, SpaceX acquired xAI in an all-stock deal valuing the combined entity at $1.25 trillion. Grok’s infrastructure is now being integrated into the Starlink satellite constellation. The stated aim is “compute-in-orbit”: move heavy AI workloads into space, run them on near-constant orbital solar power, and use the vacuum of space for cooling instead of the enormous air-conditioning systems that consume a significant share of every data centre’s energy budget. xAI also announced a 30 megawatt solar farm next to its Colossus supercomputer in Memphis. The message is clear: Musk does not think the Earth’s energy grid can keep up. His answer is to leave it behind. Whether orbital AI infrastructure becomes viable in the timeframe AI search needs it is another question entirely.

What This Means for SEO and AI Search Adoption

The SEO community has been treating AI search adoption as an inevitability with a timeline measured in months. The technology is real. The user intent is shifting. AI Overviews are already changing how organic results are displayed and clicked.

But adoption speed is not just a function of how good the technology is. It is a function of whether the infrastructure can support it at scale, reliably, affordably, and without geopolitical risk, cutting the supply chain in half.

Right now, none of those conditions is fully met.

Data centres are military targets. Helium supply chains are disrupted. Gas prices are up 70%. Energy grids in the US, UK and Europe are struggling to keep pace with demand, which is already straining supply before real scaling has happened.

WORLD ECONOMIC FORUM, APRIL 2026
“The resilience of large-scale compute infrastructure becomes inseparable from broader questions of economic continuity and national security. It is time to start treating AI infrastructure as critical infrastructure.”

When the World Economic Forum is publishing reports calling for AI infrastructure to be treated on the same level as power grids and water systems, that is not a sign of a smooth transition. That is a sign of how fragile the foundations currently are.

Our Take

We are not betting against AI search. We are betting on it. We are already building strategies for clients that account for AI Overviews, GEO, and conversational answer optimisation. But we are also grounded in reality. The transition will take longer than the hype cycle suggests. The energy constraints are real, documented, and getting harder to ignore. Any agency telling clients to abandon traditional SEO completely because AI search is here right now is running ahead of the infrastructure. Go where the user goes. But watch where the power grid is pointed.