The Four Pillars of AI Infrastructure: Chips, Electricity, the Grid and Water
AI feels like software, but it stands on four very physical pillars. Remove any one of them and ChatGPT goes silent. Here is what each pillar is, in plain words.
First, the building: the data centre
The four pillars all live inside data centres, warehouses packed with racks of computers running day and night. There are roughly 11,000 worldwide. The United States hosts about 5,400 of them, around 45 percent of the global total, followed by Germany with 529, the UK with 523 and China with about 450, according to the Cloudscene tracker. India has between roughly 155 and 260 depending on the tracker, clustered in Mumbai, Hyderabad, Delhi NCR, Bengaluru and Chennai. Now to the four things every one of these buildings depends on.
Pillar 1: The chips
Ordinary computers run on CPUs, chips that work like one brilliant chef doing tasks one after another. AI needs GPUs, chips that work like a thousand cooks chopping in parallel, because AI maths is millions of small calculations done at once.
One company dominates. NVIDIA controls roughly 80 to 85 percent of the AI data centre chip market, about 81 percent according to research firm IDC, which made it the first company in history valued above 4 trillion US dollars.
The rest is split between AMD, with an estimated 5 to 8 percent, and the tech giants building their own chips: Google’s TPU, Amazon’s Trainium and Microsoft’s Maia, together heading towards 10 to 15 percent of the market. In the AI gold rush, NVIDIA is the shop selling shovels, and there is a queue outside.
Pillar 2: Electricity, the fuel
AI chips are hungry. Your ceiling fan uses about 75 watts; a single top AI chip can draw around 700, and a large data centre bundles hundreds of thousands of them, consuming as much power as a small city.
The International Energy Agency projects that data centres worldwide will double their electricity use from 485 terawatt hours in 2025 to about 950 terawatt hours by 2030, more than Japan, a country of 124 million people, uses today. AI is the runaway driver: power used by AI focused data centres is set to triple in that period, from an estimated 10 to 50 terawatt hours slice in 2023 to a plausible 200 to 400 terawatt hours by 2030.
Pillar 3: The grid, the delivery pipes
Electricity and the grid sound like one thing, but they are two different problems. Electricity is how much power AI consumes. The grid is how that power physically reaches the building, through the national network of power stations, pylons, cables and substations. Think of electricity as the water in a reservoir and the grid as the pipes to your tap: you can have a full reservoir and still a dry tap.
The grid is proving to be the weakest link. In the UK, around 140 proposed data centre projects have asked for 50 gigawatts of power, more than the entire country’s peak demand of 45 gigawatts, and some projects have been told to wait up to 15 years for a connection. A data centre is like parking a small city on a quiet street: the roads cannot cope, and building bigger ones takes years.
Pillar 4: Water, the coolant
All that electricity becomes heat, and hot chips fail. Most data centres cool themselves the way your body does, by evaporating water, and that water is gone once it evaporates. A large data centre can consume up to 19 million litres a day, equal to a town of 10,000 to 50,000 people, according to the Environmental and Energy Study Institute.
Newer designs are switching to closed loop cooling that recycles the same water, and experimental chips, brain-inspired neuromorphic designs and light-based photonic processors, promise big energy savings, but none of them can yet run the giant models behind today’s chatbots.
The bottom line
AI’s future will be decided as much by engineers laying cables and pipes as by people writing software. As the IEA’s director puts it: there is no AI without energy.
Suggested readings:
AI Summit London 2026: Key Takeaways, Trends & Big Announcements
AI, Machine Learning, Deep Learning and Generative AI: A Plain English Guide