Why Data Infrastructure Is the Future

Every AI headline is about a model: a new release, a new benchmark, a new demo that goes viral overnight. But underneath every one of those models sits something far less glamorous and far more decisive, the physical infrastructure that makes intelligence possible at scale. Racks of GPUs. Substations. Cooling systems. Fiber. Land. Power purchase agreements negotiated years in advance.

Models come and go. Infrastructure is the layer that decides who actually gets to compete.

The bottleneck has moved

For the first two years of the generative AI boom, the constraint was chips. Now it's power, land, and permitting, and that shift changes who wins.

Goldman Sachs Research projects US data center power demand will climb from 31 gigawatts in 2025 to 66 gigawatts by 2027, more than doubling in two years. Data centers' share of total US peak summer power demand is expected to jump from roughly 4% to 8.5% over that same window. That is not incremental growth. That is a structural shift in how the country allocates electricity, and it is happening faster than the grid, permitting processes, or local political systems are built to absorb.

Which is exactly why New York's legislature just passed a moratorium on new large AI data centers, with Gov. Hochul following with an executive order pausing construction for up to a year. It's tempting to read that as AI infrastructure hitting a wall. The more accurate read: demand for this infrastructure has outpaced the physical and political capacity to build it. The moratorium is a pause, not a ban, and it doesn't change the underlying demand curve, it just confirms how scarce and contested this infrastructure has already become. States don't move to pause supply unless demand is outstripping what's easy to build.

The capital numbers are hard to ignore

McKinsey & Company estimates that AI-ready data center infrastructure will require between $3.7 trillion and $7.9 trillion in capital investment globally by 2030, depending on how fast adoption scales, with $5.2 trillion as their base-case estimate. Even at the low end of that range, it's a generational capital allocation shift, on the scale of the interstate highway system or the original electrification of the country.

The hyperscalers are already voting with their balance sheets. The four largest, Amazon, Microsoft, Alphabet, and Meta, are projected to spend roughly $725 billion combined on capital expenditure in 2026 alone, up around 77% from 2025, with the overwhelming majority earmarked for AI data centers, custom silicon, and power. Analysts expect that figure to cross $1 trillion in 2027. Add in Stargate, the $500 billion joint venture between OpenAI, SoftBank, and Oracle to build out AI data center capacity across the US, and the picture is clear: the companies with the best visibility into AI demand are betting more on physical infrastructure than on any single model or product.

Why infrastructure, not applications, is the more durable bet

AI applications are a land-grab, fast-moving, winner-take-most-for-now, and vulnerable to the next model release making today's product obsolete. Infrastructure behaves differently:

• It's scarce by physical constraint, not by competitive moat. You can't out-code your way into more grid capacity or more buildable land near fiber and water.
• It's monetized regardless of which model or application wins. A data center, a power agreement, or a cooling system gets paid whether the workload running on top of it is OpenAI, Anthropic, Google, or whoever comes next.
• It sits at the center of a demand curve that isn't slowing. Every scenario McKinsey modeled, from constrained to accelerated adoption, still points to trillions in required capex. The range is how much, not whether.

That's the distinction worth sitting with: the AI revolution won't just be defined by the companies building the models. It will be defined by whoever controls the infrastructure underneath them.

Why this matters for how a portfolio is built

For investors, this creates a genuinely uncommon setup, a multi-trillion-dollar demand curve colliding with real, physical, near-term supply constraints on power, permitting, land, and water. That combination is what tends to reward investors who are positioned early, rather than chasing the loudest application-layer story of the moment.

That exposure requires discipline. King Fund III allocates 20% to AI and technology ventures, including AI infrastructure, compute pipelines, enterprise automation and applied deep-tech, alongside 80% senior-secured real estate credit. Early-stage positions carry technology, execution and liquidity risks, including loss of the entire investment. The credit allocation does not guarantee recovery of venture losses.

The infrastructure underneath AI isn't the exciting part of the story. It's the part that decides how the rest of the story gets written.

Happy to share more with accredited investors exploring this space.