The growth that AI comes with as its becoming part of our daily lives comes with a price. The World Economic Forum reports that AI driven data centres now use more electricity than entire nations such as South Africa and Indonesia. Projections show energy use increasing from 260 terawatt hours in 2024 to 500 terawatt hours in 2027.
One query in OpenAI’s ChatGPT consumes 2.9Wh of electricity, about 10 times that of a Google search, according to the World Economic Forum. Multiply that by millions of users each day and the scale becomes clear.
At the 2024 World Economic Forum in Davos, Sam Altman said: “An energy breakthrough is necessary for future artificial intelligence, which will consume vastly more power than people have expected.” His words capture the scale of the task. AI can grow, but electricity demand cannot increase without limit.
Is On Device AI The Answer?
One answer lies in moving AI away from the big data centres and into the devices people already use. The World Economic Forum reports that on device AI can deliver a 100 to 1,000 fold reduction in energy consumption per AI task compared to cloud based AI.
Cloud systems send data back and forth between devices and remote servers. That constant transmission uses electricity. On device AI processes data locally. It cuts the need for heavy data traffic and lowers power use per task.
Startups such as Groq, DeepSeek and DeepX are building chips designed for this kind of computing. Their hardware favours energy efficiency over raw computing muscle. The World Economic Forum describes this direction as the most promising solution to AI’s power problem.
Governments are reacting too. Singapore stopped approving new data centres between 2019 and 2022 due to energy shortages. The country runs more than 70 data centres and accounts for 60% of Southeast Asia’s total data centre capacity, according to the World Economic Forum. That pause shows how electricity limits can shape tech growth.
The same report floats the idea of a global energy credit trading system. Companies that adopt low power AI could trade credits and gain financially. DeepX presented this idea at the 2024 World Economic Forum Annual Meeting in Dalian, China.
More from News
- Silicon Valley’s Hottest Coding Startup Got Caught Building On Chinese AI – And Founders Should Be Asking Why It Matters
- Will The US Lifting Sanctions And Easing Tariffs Actually Stabilise Oil Prices?
- UK Unemployment Holds At 5.2%, But Can We Trust ONS’ Jobs Data?
- What Will The EU Inc. Proposal Mean For The Fintech Industry? Experts Discuss
- New Data Shows That Only 31% Of UK Firms Report Positive ROI From AI
- In A Galaxy Not So Far Away, The “Empire” Strikes Back – Microsoft And OpenAI Heat Up The Cloud AI Wars
- Is the Global Race for AI Talent Creating New Hiring Risks?
- Bank Of England Holds Interest Rate At 3.75% – What Does This Mean For The UK Economy Today?
What Is The UK Doing Differently?
The UK is betting on decentralised AI. Nottingham Trent University leads a new TinyML UK Network, funded by UK Research and Innovation through the Engineering and Physical Sciences Research Council. University of Southampton and Imperial College London act as co leads.
TinyML allows machine learning to run on small, low power devices rather than in vast server farms. AI runs directly on sensors, wearables and embedded systems. It responds in real time and keeps working even when internet access drops.
Professor Eiman Kanjo of Nottingham Trent University said: “AI adoption is accelerating, alongside concerns over energy consumption, infrastructure cost, resilience, privacy and sustainability.”
She said: “This is our opportunity to bring together our engineering, electronics and AI communities to build decentralised, low-energy, privacy-preserving and affordable systems.
“The new TinyML UK Network is our chance to grow UK capability and help lead in this space.
“The network will connect AI, hardware, embedded systems and engineering researchers across the UK. It will build strong links with international industry and global TinyML leaders, run training, competitions and events for students, researchers and SMEs, support real-world impact in health, sustainability and security, and help shape a UK roadmap for future TinyML research and skills.”
TinyML already works in livestock monitors that learn animal behaviour and flag health problems. Personal safety devices can detect unusual motion or sound and trigger alerts without storing recordings. Data stays close to where it is created, which improves privacy and lowers energy use.
Will We See More Efficient AI In Our Lifetime?
The signs so far are saying yes. Energy use in data centres is going up more and more by the day, and governments are placing limits. At the same time, chip designers and universities are building systems that use far less power per task.
On device AI can cut energy use per task by up to 1,000 times, according to the World Economic Forum. TinyML shows that small models can handle real world jobs without vast infrastructure. Funding from UK Research and Innovation gives that work momentum.
We’re most likely going to see AI grow in two ways at once. Large data centres will continue to train advanced models. Smaller, specialist systems will run on everyday devices. The both of those together could deliver smarter services without electricity bills increasing out of control.
