Time slips away slowly while you’re in it, but not when you look back. Go through old photos and you realize just how much time has passed. Organize a closet, and find things that used to matter, but with time, no longer do. You suddenly feel…old. We are all aging.

Here on earth, our US infrastructure is old too. It’s been put in an old shoebox waiting for an update. The average American bridge is 44 years old and designed for a 50-year life. The electrical grid is about 30 years old on average, and showing its age even before the AI data center boom started demanding power. 

In fact, the American Society of Civil Engineers (ASCE) graded the US infrastructure with an overall C in their 2025 Report Card. The best grade it's given in decades, but still a C.

The funding needed to improve it has only grown. In 2021, ASCE estimated the shortfall at $2.6 trillion. In their 2025 report, that number increased to $3.7 trillion. This is despite attempts to fund it through infrastructure law. 

The reason almost everything is aging at once isn't random. When Eisenhower signed the Federal Aid Highway Act in 1956, America embarked on the most ambitious infrastructure build in its history—47,000 miles of interstate highway, alongside a wave of bridges, water systems, and power infrastructure. The whole country was built at once, across roughly a 30-year window from the late 1950s through the mid-1980s.

Since then, infrastructure spending as a share of GDP has fallen in nearly every decade. The political will that sustained the build—a combination of Cold War urgency, postwar economic confidence, and bipartisan consensus—faded when the system was complete.

Now it's all coming due at once. 

Data centers currently consume about 4% of US electricity. The Lawrence Berkeley National Laboratory projects the demand will increase from 6.7% up to 12% by 2030, tripling at the high end. A single hyperscale campus uses approximately 100 MW, equivalent to 100,000 homes. There are hundreds planned. The grid being asked to power them averaged a D+ from ASCE and just got downgraded further in the 2025 report.

The North American Electric Reliability Corporation (NERC), who is the grid reliability authority, has already warned of an elevated risk of summer electricity shortfalls starting this year across all 3 US grid regions. Separately, the International Energy Agency (IEA) estimates up to 20% of planned data center projects could face delays without significant transmission investment. Goldman Sachs projects $720 billion in grid investment will be required over this decade to support demand growth. 

The physical world is progressing on a different timeline than the agentic world. This will cause fits and starts in the AI trade along the way. 

The infrastructure buildout is a precondition for the AI payoff and I don’t believe the market is pricing the enabling layer properly.

So where does this show up for investors? Getting exposure to the enabling layer is important to the growth story we are witnessing given the interaction between the physical and agentic world:

• Raw materials: Steel and copper. Grain oriented electrical steel is the raw material layer of the entire grid. S&P Global projects copper demand to reach 42 million metric tons by 2040, a 50% increase from current levels.

• Building physical infrastructure: Crushed stone, sand, gravel, and ready-mix concrete, which are the literal material that roads and foundations are made of. This also includes construction equipment makers.

• Transporting the infrastructure and materials: Rail is often the only practical inland transport for oversized, heavy cargo. For example, a large power transformer weighs 100–400 tons.

• Getting the power and holding the power: This includes contractors that build out the electrical grid and equipment makers for it, as well as alternative energy suppliers and battery components.

• Managing the power once it’s built: This includes power distribution, uninterruptible power supplies, thermal management, and liquid cooling for high-density AI racks.

The agentic AI trade is the biggest theme of this decade. But it has a prior chapter that's aging and is $3.7 trillion underfunded and constrained by an interconnection queue. The fits and starts we’ll experience from here aren't a bug, they are part of the physical buildout required for a reconstructed agentic-forward future.