We are building increasingly powerful AI systems. But I believe the next global AI competition may be decided by something much more fundamental: access to reliable and affordable energy. AI energy consumption can be the next big bottleneck.
For the past few years, the world has been obsessed with making artificial intelligence more intelligent.
Bigger models.
Better reasoning.
More powerful AI agents.
Faster processors.
Larger data centres.
Every few weeks, we hear about another breakthrough that promises to change how we work, conduct research, or run businesses.
And as someone who has spent years studying technology, I find these developments fascinating.
But recently, I have become interested in a different question.
What happens when our ability to develop artificial intelligence grows faster than our ability to generate and deliver the electricity required to operate it?
When The Global AI Bubble is Beginning to Crack, I think we are approaching a point where this question could become one of the defining challenges of the AI industry.
On November 3, 2026, the Electric Power Research Institute announced the launch of its AI-Power Exchange, a global initiative bringing together researchers, energy specialists, regulators, and other stakeholders to examine the relationship between artificial intelligence and electricity infrastructure.
On the same day, a UK-backed research programme announced that it would evaluate optical computing technology designed to reduce the energy requirements of certain AI workloads.
These developments caught my attention.
Because they suggest something important.
The future of artificial intelligence may depend as much on progress in energy engineering as on progress in AI models.
AI Is Not Just Software Anymore
When people use ChatGPT, Gemini, Claude, or another AI system, the experience appears almost effortless.
You type a question.
A response appears.
Perhaps the entire interaction takes only a few seconds.
But behind that response is an enormous physical infrastructure.
Data centres.
Semiconductors.
Cooling equipment.
Electrical substations.
Transmission networks.
Backup power systems.
And thousands of engineers responsible for keeping everything operating.
This is something I think we often forget.
We talk about artificial intelligence as though it exists entirely in the digital world.
But AI depends on the physical world.
And the physical world has limitations that cannot always be solved through better software.
A data centre cannot operate without electricity.
A processor cannot perform complex computations indefinitely without generating heat.
And a new facility cannot simply connect to an electricity grid that lacks the required capacity.
AI may be virtual in how we experience it, but it is physical in how we produce it.
That distinction is becoming increasingly important.
The Numbers Are Beginning to Look Serious
According to the International Energy Agency’s 2026 report, Key Questions on Energy and AI, global data-centre electricity consumption could increase from approximately 485 terawatt-hours in 2025 to 950 terawatt-hours by 2030.
That would represent roughly 3% of global electricity demand.
AI-focused data centres are expected to grow even faster, with electricity consumption potentially tripling during this period.
Now, 3% may not sound particularly alarming.
And I don’t think we should exaggerate the figure.
Data centres are not responsible for most of the world’s electricity consumption.
Industries, buildings, transportation, and other sectors account for much larger shares.
But global averages can hide local problems.
Imagine a city attracting several large data-centre projects.
These facilities may require substantial amounts of electricity throughout the day and night.
The local power grid must be capable of delivering that electricity reliably.
If transmission infrastructure is inadequate, additional investment becomes necessary.
If new generating capacity is not available, some projects may experience delays.
And if the cost of expanding the electrical network is not allocated carefully, other electricity consumers could face financial consequences.
A relatively modest share of global electricity consumption can still create enormous pressure in a particular location.
I think this is where the next major AI infrastructure challenge will emerge.
The Problem Is Not Just How Much Electricity We Have
My academic background is in electronics and communication engineering.
One of my early engineering projects involved controlling a robot through mobile and desktop systems.
That experience helped me appreciate something that remains relevant today.
Technology is not merely about making a device perform a task.
You also need to understand the conditions under which it must operate.
Power availability.
Communication reliability.
Hardware limitations.
Response times.
System failures.
The same thinking applies to AI infrastructure.
A country may generate substantial amounts of electricity.
But that does not automatically mean it can deliver enough power to a particular data centre at the required time.
Electricity generation and electricity delivery are different challenges.
You need transmission lines, substations, transformers, local distribution infrastructure, and grid-management capabilities.
The IEA’s Electricity 2026 report highlights the growing number of electricity-generation, storage, and large-demand projects waiting for grid connections worldwide.
It also estimates that annual grid investment would need to increase substantially by 2030 to meet growing electricity requirements.
This is why I believe we are entering a period where electricity infrastructure could influence where the world’s next major AI investments are located.
Access to computing hardware will remain important.
But reliable electricity may become an equally important competitive advantage.
The AI Industry Has Another Uncomfortable Problem
There is something else we need to understand.
AI computing is becoming more energy-efficient.
The IEA reports that the electricity required for many individual AI tasks has declined dramatically as hardware and software have improved.
That is encouraging.
But efficiency creates an interesting possibility.
When something becomes cheaper to use, people often use more of it.
Imagine that an AI task becomes ten times more energy-efficient.
That sounds like a significant improvement.
But what if the number of tasks performed increases fifty times?
Total electricity consumption can still increase.
This is particularly relevant as we move from simple AI conversations toward continuous AI agents, video generation, autonomous workflows, and increasingly complex reasoning systems.
A text-based question and an AI system performing extensive computational work do not necessarily have comparable energy requirements.
So I would be cautious about claims suggesting that every AI interaction carries an enormous environmental cost.
That oversimplifies the problem.
But I would be equally cautious about assuming that improving efficiency per task will automatically reduce the total energy consumption of the AI industry.
AI can become cheaper per task while becoming more expensive to operate at scale.
And I think that is an important distinction for businesses and policymakers.
India Cannot Ignore This Challenge
I believe India has an extraordinary opportunity to become an important centre for artificial intelligence.
We have a large technology workforce, growing digital infrastructure, expanding research capabilities, and substantial potential to develop applications for domestic and international markets.
But we also face infrastructure challenges.
A 2026 report from the Council on Energy, Environment and Water examines the electricity and water requirements of India’s expanding data-centre ecosystem.
It estimates that data centres accounted for approximately 0.5% of India’s national electricity consumption in 2024, with electricity and water requirements projected to grow substantially by 2030.
India’s Ministry of Electronics and Information Technology has also acknowledged the growing electricity requirements associated with AI and high-performance computing.
For India, I think the objective should not simply be to attract the maximum number of data centres.
We should ask whether those investments are supported by reliable electricity infrastructure, appropriate cooling arrangements, and long-term resource planning.
Water deserves attention too.
Some data-centre cooling technologies consume considerable water, while other designs reduce direct water consumption but may involve different costs or energy trade-offs.
These choices become particularly important in areas already experiencing water stress.
I would like India to develop AI infrastructure that is globally competitive without creating unnecessary pressure on local communities.
That requires cooperation between technology companies, energy planners, researchers, and governments.
Not simply announcing larger investments.
Could the Next AI Breakthrough Come From a Different Kind of Computer?
One of the developments I find particularly interesting came from the United Kingdom on November 3.
Lumai, a company developing optical computing technology, announced that its AI inference system would be evaluated through a research programme supported by the UK’s Advanced Research and Invention Agency.
Instead of relying exclusively on conventional electronic computation, the technology uses light to perform certain mathematical operations.
The company believes its approach could significantly reduce the electricity required for some AI workloads.
It has identified potential reductions of up to 90% in certain power requirements.
However, those are technology targets and claims that still require independent evaluation in realistic commercial conditions.
We should not treat them as proven industry-wide savings.
Nevertheless, the research direction is fascinating.
Because it reminds me that the future of AI computing may not be limited to making existing chips more powerful.
New computing architectures could become increasingly important.
Optical computing.
Specialized accelerators.
More efficient memory systems.
Smaller models.
On-device AI.
Better algorithms.
Each could contribute to reducing unnecessary computational work.
I believe the next generation of AI innovation will require a deeper collaboration between computer science, electronics, materials science, and energy engineering.
This is an area I would love to see more Indian researchers and startups explore.
Who Should Pay for the AI Infrastructure Boom?
There is another question that deserves attention.
Who should bear the cost of expanding the electricity infrastructure required by AI data centres?
The companies building the facilities?
Electricity utilities?
Governments?
Or ordinary electricity consumers?
I don’t think there is one simple answer.
The appropriate arrangement will depend on local regulations, investment structures, grid conditions, and the wider economic benefits of each project.
But the question matters.
A large data centre can create jobs, generate investment, and contribute to the digital economy.
It can also require substantial infrastructure that may not have been necessary otherwise.
If the financial arrangements are poorly designed, the costs and benefits may be distributed unequally.
This is why I think conversations about AI infrastructure should extend beyond technology companies and investors.
They should include energy regulators, local communities, electricity providers, and environmental specialists.
The objective should be to support technological development without ignoring its wider consequences.
My Suggestion: Start Measuring AI Value Differently
Through my work in Lean Six Sigma and process excellence, I have learned that improving performance requires choosing the right measurements.
I think AI companies and their customers should begin looking beyond model capability alone.
A powerful AI system is impressive.
But what is the cost of producing a useful outcome?
How much computing capacity does the system require?
Could the same task be completed using a smaller model?
Does the business need advanced AI reasoning, or would a conventional software system solve the problem?
Can processing happen locally when appropriate?
How much infrastructure is required to support the application at scale?
These are questions I would encourage founders to ask before making major investments.
I would also encourage businesses to measure AI through actual improvements in cost, quality, productivity, reliability, and customer outcomes.
If a simpler technology can solve the problem effectively, using a more expensive AI model may not create additional value.
The goal should not be to consume more intelligence. It should be to generate more value from the intelligence we consume.
That is the kind of thinking I believe will become increasingly important.
Final Thoughts: The Future of AI Will Be Decided Beyond the Data Centre
I remain extremely optimistic about artificial intelligence.
I believe it will transform scientific research, engineering, healthcare, manufacturing, education, and countless other industries.
But I also believe we are approaching a period when technological ambition will collide with physical limitations.
Electricity cannot be generated merely by writing a better prompt.
Transmission networks cannot be expanded as quickly as software can be updated.
And building a sophisticated AI model does not automatically solve the infrastructure problems required to operate it.
The next phase of the AI revolution may therefore depend on how effectively we connect several different fields.
Artificial intelligence.
Electrical engineering.
Energy generation.
Grid infrastructure.
Semiconductor design.
Environmental sustainability.
And business economics.
As an entrepreneur and researcher, I find this challenge exciting because it creates opportunities for innovation far beyond the traditional AI software industry.
Perhaps the next major breakthrough will not be another model with extraordinary reasoning capabilities.
Perhaps it will be a technology that makes powerful AI available using a fraction of today’s computing resources.
Or an energy system capable of supporting advanced computing without compromising reliability and affordability.
Whatever form it takes, I believe one thing is becoming increasingly clear.
The world is racing to create more powerful artificial intelligence. But the countries and companies that can power that intelligence efficiently, reliably, and sustainably may ultimately shape the next chapter of the AI revolution.
— Jaideep Parashar
Founder & Director, ReThynk AI Innovation and Research Pvt. Ltd.
Research References:
1. International Energy Agency: AI and Energy Nexus
2. EPRI: AI Power Exchange
3. LUMAI: Press Release
4. CEEW THE COUNCIL: Scaling India’s Data Centre Ecosystem