For most of modern economic history, electricity sat quietly in the background.

A business chose a location.

It built the facility.

It ordered the machinery.

And somewhere near the end of the project, somebody arranged the electrical connection.

That sequence worked for so long that almost nobody treated it as a strategic assumption.

Now, in a growing number of regions, that assumption is no longer safe.

A company can have the land.

The capital.

The equipment.

The customers.

And still be unable to operate because there is no available path to connect the building to the grid for several years.

This is the first physical layer we examine in Series II.

And by the framework established in Series I, electricity produces the most severe dependency reading we have seen so far.

One grid.
Total criticality.
Substitution measured in years.

The headline number is extraordinary.

Somewhere between roughly 2,300 and 2,600 GW of generation and storage projects are sitting in US interconnection queues.

That is more than the entire installed generating capacity of the country.

But this number is easy to misread.

It does not mean thousands of gigawatts are already under construction and about to arrive.

Historically, only around 19% of projects entering these queues actually reached operation.

Most never get built.

So the queue is not simply a backlog of physical construction.

It is a list of options.

Developers enter early because waiting is dangerous.

They may enter several queues at once because the cost of applying has historically been low.

That is individually rational.

But it creates a feedback loop.

Long queues make early applications more valuable.

More applications make queues longer.

And the process reinforces itself.

This means the queue measures something slightly different from what the headline suggests:

how long it takes to discover whether a project is viable.

Much of the policy debate focuses on the administrative layer.

That makes sense.

The study process is slow.

Rules were often designed for a system receiving far fewer applications.

Regional operators are now moving toward batch processing, revised study procedures and new rules for large users.

Those reforms are real.

They will help.

But they only address the first layer.

A workflow can be redesigned.

This may take one or two years.

But process reform makes the answer arrive faster.

It does not create the physical capacity required for the answer to be yes.

Underneath the study process sit transformers, substations and transmission equipment.

And their lead times have been moving in the wrong direction.

Large power transformers now commonly carry delivery times measured in years.

Substation transformers have moved beyond three years in some reported cases.

This is a very different kind of constraint.

You cannot reform a transformer with a new procedure.

It has to be manufactured.

And manufacturing itself depends on another supply chain:

special electrical steel;

copper;

specialised factories;

qualified assembly capacity.

The bottleneck simply moves one layer downward.

Then comes the deepest layer.

Engineers.

Modern large electrical loads increasingly sit at the intersection of power systems, data centres, liquid cooling and highly specialised infrastructure.

Utilities, developers and manufacturers all recruit from overlapping pools of expertise.

That capacity cannot be procured on the same timetable as equipment.

A country can fund factories.

It can redesign the queue.

It cannot create ten years of engineering experience in two.

This gives us three different clocks:

Process: 1–2 years

Equipment: 3–5 years

Engineers: around 10 years

The longest one ultimately governs.

This is one of the central lessons carried over from Series I.

Fixing a bottleneck often does not remove scarcity.

It relocates it.

Reform the study process and equipment becomes more visible as the constraint.

Expand equipment production and the skills shortage becomes more visible.

There is no point in the sequence where the problem simply disappears.

It moves downward until it reaches the slowest thing in the stack.

When demand exceeds what a system can absorb, the system begins to economise.

That is exactly what overloaded institutions did in Series I.

The grid is now doing the physical version.

It delays applications.

It creates special categories for large users.

It raises the financial cost of entering through collateral and minimum billing commitments.

Some jurisdictions restrict new large-load connections.

And then comes a fifth response that only physical infrastructure can make:

users leave.

Large companies increasingly consider dedicated generation and microgrids.

Instead of waiting for the shared grid, they build their own supply.

For the company doing it, this can be entirely rational.

It reduces dependence on the queue.

But it changes the shared system.

The largest and best-capitalised users can increasingly solve the constraint privately.

Smaller users cannot.

That points toward the same two-tier structure identified in Series I:

those able to purchase their own resilience move outside the common bottleneck; those unable to do so remain inside it.

Our base case is Rationing by Price and Category — 50%.

Queues remain long, but access becomes increasingly determined by collateral, minimum payments and dedicated rate classes rather than simple order of arrival.

We assign 25% to Exit to Private Generation.

Large users increasingly bypass the common grid.

We assign 15% to Demand Slowing to Meet the Grid.

Some planned compute and industrial expansion disappears because electricity cannot be secured.

Only 10% goes to the grid catching up sufficiently to remove the constraint.

The planning sequence is changing.

Historically:

Site → Building → Equipment → Power

Increasingly:

Power → Site → Building

For every meaningful expansion, businesses should establish three things early:

current available capacity;

headroom before an upgrade is needed;

and the actual observed wait for additional capacity in that region.

Not the target connection time.

The real one.

If a transformer takes three years and the building takes eighteen months, the transformer decision has to happen before decisions that used to come first.

That is not a minor engineering change.

It changes project sequencing.

The compute investment story is often framed around demand, chips and capital expenditure.

Those remain important.

But there is now another question:

Can the announced capacity actually connect to power?

Planned capacity and connected capacity are not equivalent.

A data-centre project can have financing and customers and still fail to open on schedule because the electrical connection is not ready.

That creates both risk and scarcity.

The bottleneck increasingly sits around:

transformers;

high-voltage equipment;

engineering services;

and existing sites with secured grid access.

These are not necessarily the most glamorous parts of the AI infrastructure story.

They may be among the most important.

Electrical capacity is also becoming a location advantage.

A region with spare grid capacity can attract investment for reasons that have nothing to do with tax rates, labour costs or subsidies.

A region whose connection queue runs several years has a structural growth constraint.

And because the deepest bottleneck takes close to a decade to address, that advantage or disadvantage can persist much longer than a normal business cycle.

Series II applies the same three questions to every physical substrate:

How many genuinely independent alternatives exist?

For a grid connection, usually one.

What stops if it fails?

Everything on the site.

Immediately.

How long until another option actually functions?

Years.

This gives electricity the most severe dependency profile we have measured so far.

And that is the central finding.

The modern economy has spent years discussing software, AI, chips and capital.

But underneath all of them sits an older system moving at a completely different speed.

The digital economy operates in milliseconds.

The grid expands in years.

The collision between those two clocks is becoming one of the defining constraints of the next decade.

Read the full analysis:

THE CONNECTION QUEUE
Why electricity became the limit on everything built on top of it

THRIVE IN CHAOS
Decision Intelligence for an Uncertain World

Analysis → Forecast → Recommendations
Signal → Meaning → Action → Stability