WHAT IF THE MOST IMPORTANT MACHINES IN THE AI REVOLUTION WERE BUILT BY OUR GRANDFATHERS?

 

We have become accustomed to looking at the future through the machines that are new: the enormous AI models, the GPUs, the humanoid robots, the data centres glowing behind anonymous fences, the chips getting smaller while their ambitions get larger. Billions of dollars are pouring into machines designed to make machines think, see, speak, reason and eventually act. Every day brings another announcement, another breakthrough, another staggering number. The story of artificial intelligence is usually told as a story of acceleration: faster chips, bigger models, more compute, more data, more intelligence. It is a story about what is coming next.

But underneath all of this sits another machine, quietly and almost invisibly, belonging to a much older world.

It is older, heavier and almost aggressively unglamorous. It does not write poetry, drive a car or answer a question. It may have been designed decades before ChatGPT existed, before smartphones existed, perhaps before the people now building the AI revolution were born. While investors are racing to build the newest intelligence on Earth, some of the machinery carrying that intelligence may have been designed for a world that no longer exists.

And without it, the future stops.

The machine is the transformer.

That sounds almost absurdly mundane. A transformer does not create electricity. It does not create intelligence. It simply changes electrical voltage so that power can move through the system and eventually reach the machines that need it. Yet that apparently modest function is becoming a strategic constraint as artificial intelligence drives an extraordinary expansion in electricity demand. The International Energy Agency says electricity consumption from data centres is projected to roughly double from 485 terawatt-hours in 2025 to 950 terawatt-hours by 2030. AI-focused data-centre consumption is expected to grow even faster. The world's most ambitious technology may therefore be running into a machine whose basic job has barely changed in a century.

And suddenly the machine nobody talks about becomes one of the most important machines in the room.

The IEA says it can now take up to four years to secure large power transformers, with procurement times for transformers and cables having roughly doubled since 2021. Transformer prices have also risen sharply. The problem extends beyond transformers to cables, switchgear, power electronics and other equipment required to connect new electricity supply and demand.

That creates a dangerous contradiction at the heart of the AI revolution. A new data centre can be constructed in roughly one to three years. A new grid can take five to fifteen years to plan and build. The computer industry thinks in product cycles. The electricity industry thinks in infrastructure generations. The machines that consume the future can arrive years before the machines required to feed them.

AI is moving at software speed. The machinery beneath it is moving at industrial speed.

And industrial speed may not be fast enough.

The problem becomes clearer when you stop looking at the transformer as an isolated object. Behind every powerful AI data centre is an invisible chain of machines: generators, substations, transformers, switchgear, cables, cooling systems, backup power, storage and transmission infrastructure. Behind those machines are factories. Behind the factories are supply chains for copper, steel, electrical steel, aluminium and other materials. Behind those supply chains are countries, ports, engineers, skilled workers and industrial capacity.

The cloud suddenly looks rather less like a cloud.

It looks like a continent-sized machine—and someone has to keep every part of it alive.

The IEA has warned that AI data-centre expansion is placing additional pressure on power-equipment supply chains, while critical minerals used in data-centre construction overlap with those required for broader energy infrastructure. It estimates that grid constraints alone could delay around 20% of global data-centre capacity planned for construction by 2030.

That is the uncomfortable part: the bottleneck may not be artificial intelligence at all.

It is a problem with everything underneath artificial intelligence.

A recent study from Johns Hopkins researchers makes the point even more starkly. Their analysis suggests that transformers, inverters and other grid-supporting equipment could become some of the biggest constraints on the expansion of AI infrastructure, with demand for critical grid equipment potentially exceeding available manufacturing capacity by roughly 30% by 2030.

Think about what that means. The world could have the chips, the algorithms, the money and even the land—and still be waiting for a transformer.

This is where the title of this story becomes more than a rhetorical trick. Much of the infrastructure carrying the modern world was designed for a different age. The U.S. grid, for example, contains large amounts of equipment based on engineering and infrastructure dating back decades. Reuters reported this month that a new California factory is being planned specifically to produce advanced transformers for a grid struggling to keep pace with data-centre growth, with industry leaders citing supply challenges and waits of up to four years for standard transformers.

The irony is extraordinary. We are building machines capable of generating answers in seconds while depending on machines whose replacement can take years. One side of the system is accelerating; the other is waiting in a factory queue.

The AI revolution wants to move at the speed of thought. The electrical system underneath it still has to wait for steel to be manufactured, wound, assembled, tested, transported and installed.

For decades, the digital economy taught us to think of physical infrastructure as yesterday's problem. Software scaled globally at almost no marginal physical cost. Platforms could reach hundreds of millions of people without building a factory in every country. The internet made distance feel obsolete.

Artificial intelligence is now reversing that illusion.

The more powerful AI becomes, the more intensely physical it becomes. The closer we get to the supposedly frictionless future, the more we run into steel, copper, concrete and wires.

A model may live in software. But training it requires electricity. Electricity requires generation. Generation requires equipment. Equipment requires factories. Factories require materials. Data centres require land and cooling. Transmission requires cables and transformers. And all of it ultimately depends on a physical industrial system that took generations to build.

The future is digital. Its bottlenecks are not.

This is also why the coming competition may be larger than the familiar semiconductor race. The world has become accustomed to thinking about technological power through chips. Washington wants semiconductor factories. Beijing wants semiconductor independence. Europe wants strategic autonomy. India wants to build a domestic semiconductor ecosystem.

All of that matters. But it may be only half the race. Underneath the chip race, another industrial race is beginning to emerge.

Who can build the machines that deliver the electricity to the machines that run the AI?

That question puts transformers, electrical steel, copper, switchgear and grid equipment into a much larger geopolitical story. The IEA warns that concentrated supply chains for some critical minerals and power technologies create vulnerabilities to trade disruptions, industrial accidents, extreme weather and geopolitical shocks.

Suddenly, a transformer factory doesn't look boring. It looks like a chokepoint. It looks strategic. So do the companies that manufacture them, the countries that possess the industrial capacity to produce them and the engineers who know how to design, install and maintain them.

This creates an extraordinary second life for some of the industrial giants of the previous century. Companies whose histories are rooted in electricity, machinery and heavy engineering are finding themselves closer to the centre of the AI story than their public image might suggest. The new economy may be producing enormous valuations for companies building artificial intelligence, but the physical economy is quietly rewarding companies that can provide the electricity infrastructure required to keep it alive.

The old industrial world has not disappeared.

It has been waiting underneath the new one.

This matters far beyond America. Europe is wrestling with grid constraints and long connection queues. China is simultaneously expanding AI infrastructure and enormous power systems. The Middle East is investing in data centres while possessing major energy resources. India wants to become a major AI and digital-infrastructure power while simultaneously expanding manufacturing, electricity generation, transmission and semiconductor capacity.

Everywhere, the same physical question eventually appears: Where will the electricity come from—and what will deliver it?

The International Energy Agency says more than 2,500 gigawatts of renewable, large-load and storage projects are currently stalled in grid queues worldwide. It estimates that annual grid investment would need to rise by around 50% by 2030 from today's roughly $400 billion to keep pace with emerging electricity demand. At the same time, building new transmission infrastructure can take four to eight years in advanced economies.

The implication is more uncomfortable than a simple equipment shortage because it changes the meaning of “AI capacity.”

We normally imagine capacity as something inside a computer.

Perhaps capacity is increasingly something outside it.

A country may have world-class AI researchers but inadequate grid infrastructure. A company may have billions of dollars and thousands of GPUs but no available connection. A developer may have land but no transformer. A government may announce an enormous AI strategy but discover that the physical system required to execute it cannot be manufactured quickly enough.

The bottleneck keeps moving backward: from the model to the chip, from the chip to the data centre, from the data centre to electricity, from electricity to the grid, from the grid to the transformer, and from the transformer to the factory that has to build it.

That is the part of the AI story we rarely see: the revolution is being constrained by things that do not look revolutionary at all.

The people who built the electrical systems of the twentieth century did not know they were preparing infrastructure for artificial intelligence. They were building grids for factories, homes, offices, trains and cities. They were solving the problems of their own time.

Yet some of their machines are now being asked to carry ours.

And that creates perhaps the strangest irony of the entire AI revolution. We talk constantly about technological obsolescence. We assume the old is swept away by the new. But in the physical world, the opposite can happen. The new can become completely dependent on the old because replacing physical infrastructure is slower, harder and more expensive than replacing software.

A software company can release a new version next month. A transformer cannot. A model can be retrained overnight. A transmission corridor cannot be moved overnight. A semiconductor design can change in a product cycle, while a power network may still be carrying equipment designed for an entirely different era.

The AI revolution may therefore be creating an unexpected hierarchy of machines.

At the top are the machines everybody photographs; at the bottom are the machines everybody forgets. And the machines everybody forgets may determine how far the machines everybody photographs can go.

Perhaps that is the real lesson hiding beneath the transformer shortage. The future is rarely built entirely from new things. It is built on old foundations that suddenly acquire new strategic value.

A transformer manufactured decades ago does not know what artificial intelligence is. It does not know Nvidia. It does not know ChatGPT. It does not know the difference between a chatbot and an AI agent. It has no idea that somewhere beyond the horizon, billions of dollars are being wagered on machines that can reason. It has only one job.

Carry the power.

And perhaps that is the most unsettling thing about the future. We may be racing toward a world whose most advanced machines depend on machinery that cannot be rushed.

The AI revolution may belong to our children. The machines that make it possible may belong to our grandparents.

And the world may discover their value only when it is already too late to build enough of them.

YOU MAY ALSO LIKE

JAPAN WANTS TO BUILD THE MACHINES BEHIND THE AI MACHINE

WHAT DO BILLIONAIRES KNOW ABOUT LAND THAT THE REST OF US DON'T?

By: Manish Kumar
Founder & Editor, Explain It Clearly