Certain shifts in the world don’t announce themselves loudly.
They show up as a line in a report — a number that makes you read it twice.

This one did that for me.

In a piece for TechCrunch on the latest International Energy Agency (IEA) analysis, Tim De Chant points out that the world will spend $580 billion on data-centres this year — about $40 billion more than it will spend on developing new oil supplies.

When data-centres outpace oil exploration, the centre of gravity in the global economy has quietly moved.

Not towards software.
Not towards models.
Towards the physical infrastructure that keeps all of it alive.

This is one of those moments where the economic map redraws itself, even if most people aren’t looking.

The News

According to TechCrunch, citing a new report from the International Energy Agency (IEA), the world in 2025 will:

• spend around $580 billion on data-centres

• spend about $40 billion less on developing new oil supplies

TechCrunch’s report highlights several key IEA findings:

• Electricity consumption from AI data-centres is expected to grow fivefold by the end of the decade, doubling today’s total data-centre usage.

• Conventional data-centres will also use more electricity, though not as dramatically as AI-heavy ones.

• Roughly half of this demand growth will occur in the United States, with much of the rest in Europe and China.

• Most new data-centres are being built near large cities with populations of over 1 million people.

• About half of the projects in the pipeline are at least 200 megawatts each.

• Many of these are being built near existing data-centres, forming clusters.

The IEA warns that:

• grid congestion and connection queues are increasing

• connection queues for new data-centres are already long in many regions

• in northern Virginia, waits for grid connection can be as long as a decade

• in Dublin, new interconnection requests have been paused entirely until 2028

The report also notes that:

• the grid supply chain is another pinch point: cables, critical minerals, gas turbines and transformers are delaying upgrades

• companies like Amperesand and Heron Power are working on solid-state transformers, which can better integrate renewables, react more quickly to grid instabilities and handle a broader range of conversions — but deployments are still at least 1–2 years away

On the supply side, the IEA expects that by 2035:

• renewables will provide the majority of new data-centre power, regardless of how aggressively countries push to lower emissions

• solar will be a particular favourite thanks to falling costs

Over the next decade, the IEA projects that data-centres will draw:

• about 400 terawatt-hours from renewables

• around 220 terawatt-hours from natural gas

• about 190 terawatt-hours from small modular nuclear power plants, if they deliver as expected

That’s the picture TechCrunch paints from the IEA report.
The question for us is what it actually means.

The Surface Reaction

You’ll see headlines about “AI’s insane energy use” or “The cloud is the new coal.”

But the more interesting story here is simpler:

Data-centres have quietly become one of the core industrial assets of the global economy.

They’ve moved from “back-end facilities” to “frontline infrastructure”.

And that shift is being driven by AI workloads that:

• train on enormous datasets

• require dense compute clusters

• demand high reliability

• expect low latency

• run at scales old data-centre designs weren’t built for

This is why the spending graph now shows data-centres overtaking new oil-supply investment.

The resource that matters most is changing.

What Is Being Built or Changed

If you zoom into the IEA–TechCrunch details, you start to see what’s actually being built.

1. Urban-proximate compute hubs

Most new data-centres are near cities with populations over 1 million.

That’s not accidental.

Cities offer:

• proximity to enterprise demand

• better network infrastructure

• labour pools with data-centre and electrical skills

• existing energy infrastructure

But they also come with:

• tighter grid constraints

• more public scrutiny

• stricter regulation

We’re relocating industrial-scale energy use closer to where people live and work.

2. Mega-scale facilities and clusters

Half of new projects are 200 MW or larger.
That’s a very different ballgame from the racks we used to think about.

At this scale:

• cooling is a major design problem

• internal network topology matters

• grid contracts become multi-decade commitments

• local politics and regulation are unavoidable

The clustering effect (building new data-centres next to existing ones) is logical from a compute standpoint — but hard on grids.

3. The grid as a hard constraint

Ten-year waits in northern Virginia and a full interconnection pause in Dublin are not small details.

They’re early warnings.

They tell us:

• the internet’s physical foundation is under real pressure

• AI growth is colliding with grid capacity and upgrade timelines

• planning cycles are out of sync: software moves in weeks, infra moves in years

That mismatch is where outages and cost spikes tend to hide.

4. Grid tech playing catch-up

Companies like Amperesand and Heron Power are working on solid-state transformers that can:

• respond faster to demand fluctuations

• better handle renewables

• offer more flexibility in power conversion

But innovation at the hardware and grid layer takes time.
1–2 years to first deployment.
More years to scale.

Meanwhile, AI demand curves are much steeper.

5. Renewables as default, not optional

The IEA’s outlook — summarised by TechCrunch — suggests that by 2035, most new data-centre power will come from renewables, with solar playing a central role.

The projected mix (400 TWh renewables, 220 TWh gas, 190 TWh SMRs) is as much about economics as it is about sustainability.

Renewables are no longer just a “green story”.
They are a cost and availability story.

The BitByBharat View

From a builder’s perspective, this is the kind of macro shift that quietly defines what’s possible for the next decade.

I’ve spent enough time around infra teams to know that when the foundation starts to strain, everything built on top eventually feels it:

• latency becomes unpredictable

• capacity fluctuations appear at the edges

• cost structures become more volatile

• deployment regions become strategic, not just convenient

We are watching the AI stack transition from being “cloud-backed” to “grid-constrained”.

That’s a very different mental model.

In the cloud era, you think about:

• regions

• availability zones

• scale-out patterns

In the emerging era, you also have to think about:

• which regions can actually get new power

• how long interconnection will take

• which geographies have paused new data-centre connections

• what mix of energy sources underlies your compute

If you’re building serious AI products or infra, you are now — whether you like it or not — in the energy business, at least indirectly.

The Dual Edge (Correction vs Opportunity)

Correction

The “infinite cloud” assumption is over.

If your roadmap quietly assumes:

• unlimited GPUs

• easy region expansion

• stable energy pricing

• fast infra onboarding

…this IEA data is a healthy correction.

You don’t get infinite capacity on a finite grid.
Not at the pace AI is trying to grow.

Opportunity

At the same time, the transition opens up an entirely new problem space for builders:

• tools that help teams plan AI deployments around grid constraints

• schedulers that optimise training around renewable availability

• cost-modelling tools that consider energy sources, not just instance prices

• smarter replication strategies across clusters and regions

• observability for energy use and carbon impact per workload

• optimisation tools for running AI workloads in data-centres with better energy mixes

These are not abstract ideas.
They’re emerging needs, driven by very real physical limits.

This is where infra-aware startups can create leverage.

Implications (Founders, Engineers, Investors)

For Founders

If your product is compute-heavy, treat infrastructure as a first-order concern.

Questions to ask:

• Which regions can reliably support your growth?

• How sensitive is your product to latency if you need to move regions?

• What does your cost curve look like if energy prices spike?

• How does your value proposition change if you can’t get capacity where you want it?

The founders who think about this early will be less surprised later.

For Engineers

This is a moment to deepen your understanding of infra:

• how data-centres are designed

• how power contracts work

• how cooling and density trade-offs play out

• how geographic distribution affects reliability

Low-level awareness will translate into better system design.

For Investors

The IEA–TechCrunch comparison is a macro signal.

When spending on data-centres exceeds spending on new oil supply, you’re not just looking at a tech trend — you’re looking at a structural reallocation of capital.

Pay attention to:

• companies that sit at the intersection of AI and energy

• tools that help manage the infra bottlenecks

• specialised hardware and grid tech (like solid-state transformers)

• infra-layer platforms that optimise AI workloads, not just run them

This is a long runway, not a quarterly story.

Closing Reflection

It’s easy to talk about AI in terms of parameters and benchmarks.
But under every model is a very physical reality — land, steel, cables, transformers, electrons.

When data-centre investment surpasses new oil exploration, it’s a reminder that the real story of AI is not just happening in labs or on product roadmaps.

It’s happening in how we build, power and govern the infrastructure that lets these systems exist.

If you’re building in this era, it’s worth stepping back and asking:

What assumptions am I making about compute, energy and geography — and are they still true?

Because the future of AI won’t just be defined by smarter models.
It’ll be defined by the infrastructure that can actually keep up with them.

(TechCrunch - Nov 2025)