The $106 trillion opportunity isn't really about infrastructure

I watched McKinsey's recent discussion on what it calls the $106 trillion global infrastructure opportunity.

The headline number is extraordinary. McKinsey estimates that roughly $106 trillion will need to be invested in global infrastructure through 2040.

Asia alone could account for roughly two-thirds of the total.

But after listening to the discussion, I don't think the $106 trillion number is actually the most interesting part.

The bigger story is that the definition of infrastructure is changing.

Infrastructure stopped meaning roads and bridges

For most of modern history, we thought about infrastructure as roads, bridges, ports, airports, railways, power plants and water systems.

Those things still matter.

But increasingly, infrastructure also means data centers, fiber networks, satellites, battery storage, EV charging networks, distributed energy, digital connectivity, environmental infrastructure — and even some of the services required to keep these systems functioning.

More importantly, the next generation of infrastructure doesn't fit neatly into individual sectors.

A modern data center isn't simply "digital infrastructure." It is simultaneously a computing asset, a power problem, a cooling problem, a grid problem, a land problem, a water problem and a connectivity problem.

An electric vehicle isn't simply a transport story. It depends on transportation infrastructure, electricity generation, grid capacity, charging infrastructure, digital payments and communications networks.

Agricultural waste can become an energy feedstock.

Ports can become energy infrastructure.

Highways can become telecommunications corridors.

Buildings can become distributed power systems.

The opportunity increasingly exists between industries rather than inside them.

AI is a physical business

We often talk about AI as if it were purely a software revolution.

It isn't.

AI is creating an enormous physical economy underneath the digital one.

Every model ultimately runs inside a physical server. That server sits inside a physical data center. That data center needs enormous quantities of electricity.

That electricity requires generation. Transmission. Transformers. Switchgear. Storage. Cooling. Backup power. Fiber. Land. Construction. Maintenance.

The IEA expects global data-center electricity consumption to roughly double by 2030, and estimates that grid constraints could delay around 20% of planned data-center capacity.

Which means one of the biggest beneficiaries of AI may not be another AI application.

It may be the infrastructure surrounding AI.

The picks and shovels.

Tiny percentages, enormous amounts of money

But AI is simultaneously changing the other side of the equation.

Infrastructure projects are notoriously difficult to execute. They involve thousands of dependencies, contractors, engineering decisions, permits, procurement processes, schedules and billions of dollars of capital.

McKinsey describes examples where AI-assisted scheduling and design have produced substantial reductions in project cost and construction timelines for certain renewable and data-center projects.

That matters because infrastructure has an unusual economic characteristic:

Tiny percentage improvements are worth enormous amounts of money.

Improve a $2 billion project's capital efficiency by only 3% and you have created $60 million of value. Reduce the downtime of a billion-dollar asset by a few percentage points and the economics can change dramatically.

Predict maintenance before failure.

Optimize construction sequencing.

Improve procurement.

Analyze satellite imagery.

Automate inspections.

Model power demand.

Optimize grid utilization.

Identify the best location for a new asset.

Reduce contract leakage.

Those don't need to be billion-dollar infrastructure companies themselves. They can be software companies sitting on top of trillion-dollar assets.

Need is not the same as bankability

And I think this is where the entrepreneurial opportunity becomes much more interesting.

Because there is a difference between infrastructure need and investable infrastructure.

A country may desperately need a road. That does not mean investors can make money building it.

Someone still needs to answer:

• Who pays?
• Where does the cash flow come from?
• How is demand risk allocated?
• What happens if tariffs change?
• Who takes currency risk?
• Who guarantees the offtaker?
• What happens when a government changes?
• How does the project get permitted?
• How does the investor eventually exit?

This is why I don't interpret McKinsey's "$106 trillion" estimate as a $106 trillion private-market TAM.

It is a measure of capital need.

Turning that need into investable assets requires something else: bankability.

And that might be one of the largest business opportunities hidden inside the entire infrastructure boom.

The companies that make projects bankable.

The companies that make permitting faster.

The companies that make assets easier to underwrite.

The companies that reduce construction risk.

The companies that improve operations.

The companies that structure energy offtake.

The companies that solve interconnection.

The companies that provide infrastructure intelligence.

The companies that make distributed infrastructure financially viable.

The companies that allow capital to move from investors into real assets with acceptable risk.

When everyone discovers the same megatrend

Private capital clearly sees the opportunity. McKinsey reports that infrastructure fundraising reached nearly $200 billion in 2025, while 51% of surveyed institutional investors expected to increase infrastructure allocations over the following three years.

But there is another side to that enthusiasm.

When everyone discovers the same megatrend, returns don't automatically rise.

Capital floods into the obvious assets. Prices rise. Auctions become more competitive. Investors take more risk. Expected returns compress.

And McKinsey is already seeing some of this tension: infrastructure holding periods are lengthening, five-year distributions to paid-in capital have fallen considerably from previous-cycle highs, and infrastructure investors increasingly need operational value creation rather than relying on passive ownership and favorable financial conditions.

So simply saying "I'm bullish on infrastructure" isn't much of a strategy.

The more important question is: where is the bottleneck?

In AI, maybe it's power.

In power, maybe it's transmission.

In transmission, maybe it's permitting.

In construction, maybe it's skilled labor.

In emerging markets, maybe it's financing.

In financing, maybe it's currency risk.

In municipal infrastructure, maybe it's project preparation.

In water, maybe it's monitoring.

In existing assets, maybe it's maintenance.

Every bottleneck creates a value pool.

What this means for Africa

Much of Africa's infrastructure opportunity isn't simply about replacing systems built fifty years ago.

In many places, the infrastructure hasn't been built yet.

That sounds like a disadvantage. But it can also create an opportunity to leapfrog.

Distributed power rather than waiting decades for centralized grid expansion.

Fiber and satellite connectivity rather than recreating every historical telecommunications layer.

Digital payments embedded directly into infrastructure services.

Smart metering from day one.

Modular construction.

Distributed water systems.

Digitally managed logistics.

AI-assisted maintenance.

Infrastructure designed around modern technology rather than retrofitted later.

But capital will not arrive simply because the need is obvious.

The World Bank's data shows that private infrastructure investment in developing markets remains highly concentrated, which means solving financing, regulation and project risk is just as important as solving the engineering problem.

Software for the physical world

That leads me to a broader conclusion.

The next great technology companies may not all be trying to replace the physical world with software. Some will use software to make the physical world work better.

For the last decade, we built software for attention. Software for advertising. Software for communication. Software for productivity. Software for commerce.

The next decade may produce a much larger generation of companies building software for energy, construction, transportation, logistics, manufacturing, water, agriculture, cities, and physical assets.

Because underneath AI, underneath economic growth, underneath electrification and underneath digitalization sits something much less fashionable but much more fundamental: infrastructure.

And perhaps the most interesting opportunity isn't owning the $106 trillion of infrastructure the world needs.

It is building the technologies, financing systems and operating platforms that make that $106 trillion possible.