The AI Economy Will Need More Power to Think: Krishan Rattan of Terra Invest

I expect the falling cost of useful intelligence to increase the ambition of the work people attempt. That is why AI’s improving efficiency makes me interested in energy. A system that can answer a question cheaply may soon be asked to investigate it thoroughly. An engineering team that can compare ten designs may choose to compare thousands and test the most promising ones.

At Terra Invest, I see that expansion as one of the most consequential links between AI and the physical economy. Total electricity demand depends on the energy used for each task and how much work is undertaken. If useful applications expand faster than energy per task falls, demand rises. My expectation of growth comes from the breadth of the work that AI could make practical.

The form of that work matters. A scientific investigation can require repeated model calls, simulation, tool use and evaluation. Higher efficiency might let a user pursue a better answer at the same budget. The desired outcome is a more effective medicine candidate or a better machine design, so the number of computational steps is secondary. We should expect the unit of demand to evolve as the technology becomes more capable.

The IEA’s 2025 Energy and AI base case put global data-centre electricity consumption at approximately 945 TWh in 2030, compared with 415 TWh in 2024. That covers the whole data-centre sector and is a projection. It is enough to make the practical connection clear: a country’s ambitions in AI need a credible way to deliver the power its infrastructure will use.

Electricity could become a more important source of geographical advantage for digital industries. Compute facilities need timely grid connections, dependable equipment and a workable cooling design. Two locations with similar headline power prices can offer very different paths to actually operating a large facility. The ability to build and connect infrastructure becomes part of the technology offer.

We should also become more precise about flexibility. A real-time service cannot simply stop when generation falls. Some offline workloads can be scheduled more freely. Large training runs bring coordination and restart costs that make their behaviour different again. The engineering opportunity is to understand that mix and design facilities around it, including where work can wait or move without destroying its value.

The IEA’s demand analysis includes greater flexibility in siting, power and operations among the conditions supporting faster data-centre growth. I find that direction interesting because it treats compute as something that can be designed in relation to the electricity system. We could build more intelligent demand at the same time as we build more supply.

Storage requires the same precision. Keeping equipment alive through a short disturbance and supplying a facility through an evening are different engineering tasks. A prolonged shortage introduces another timescale. Each requires an appropriate combination of storage, generation and transmission. Those decisions are best made together, with a clear picture of the useful work the system is intended to support.

Tesla’s 2023 Master Plan Part 3 sets out an approach to broad electrification, and it repays study regardless of one’s view of Elon Musk himself. I take from it the value of considering the production system as a whole and planning for a future with much more useful activity. AI could strengthen that work by helping engineers improve the systems that supply its own power.

The Middle East is most poised to pursue this opportunity confidently. The Gulf states already treat energy and computing as connected industrial questions, and they bring advantages that matter here: abundant generation, capital comfortable with long infrastructure horizons, and land and logistics suited to large facilities. The region has repeatedly shown it can finance, build and connect infrastructure at speed, which is exactly the capability this opportunity rewards. A founder developing robotics, scientific AI or advanced manufacturing needs a place where physical deployment is possible. Reliable electricity and competent project delivery can make the region a natural home for that work.

There are benefits beyond hosting computation. Grid equipment, storage integration and power engineering become valuable capabilities in their own right. Experience gained from demanding industrial loads can improve how other infrastructure is built. A country that develops those skills retains knowledge and productive capacity that can serve many sectors.

My bullish view is that better intelligence will invite us to attempt more. The energy system should be developed with that possibility in mind. Powering an expanding range of scientific and industrial work could become one of the defining economic opportunities of the AI age.