In 2025, something quiet yet monumental happened.
For the first time, investment in data center construction—$45 billion—matched that of office buildings at $44 billion.
It’s the first time in thirty years.
Just take a look at this chart.
The blue line represents office building construction, which has fluctuated for three decades since 1993, weathering the Internet bubble, the crash of 2008, and the ongoing decline of remote work post-pandemic. It has stabilized around $44 billion, but the trend is downward.
The green line shows data centers. Before 2012, it was almost negligible. Then it started to climb slowly, and by 2023, it surged dramatically, reaching parity with office buildings.
According to Coatue's predictions, by 2028, the scale of data center construction will far surpass that of office buildings, pushing towards the $120 billion mark.
What does this mean?
In simple terms—construction workers in the U.S. are shifting from building offices to building facilities for AI.
This isn’t some abstract tech trend; it’s a tangible shift in the flow of concrete, steel, excavators, and construction vehicles. It’s changing the work landscape for hundreds of thousands of construction workers. It’s altering land use planning for local governments. It’s reshaping the distribution of electrical loads.
To be honest, the decline of office buildings isn’t solely due to AI. Remote work has become the norm after the pandemic, and in many U.S. cities, office vacancy rates have surpassed 20%. If no one is going to the office, who’s going to build new ones?
But the explosive growth of data centers is 100% driven by AI.
Every time you ask ChatGPT a question, or when a company runs a large model inference task, it requires a room filled with GPUs, cooling systems, and massive power supplies. These things don’t just appear out of nowhere; they need to be built.
Humans are literally constructing a physical shell for the brain of AI, using concrete and cables.
A fitting analogy is the railroad construction boom of the 19th century. Back then, not all railroad companies made a profit, but those selling tracks, ties, and building the infrastructure certainly did.
It’s the same now. While we can’t predict which AI companies will thrive, one thing is certain: those building for AI—the contractors, utility companies, and cooling equipment manufacturers—are in a strong position.
However, we should remain cautious. The sharp rise in that green dotted line is a "forecast," not a fact. Historically, all infrastructure booms eventually face issues of overcapacity. Just like in the 2000 Internet bubble, many data centers were built that later stood empty.
But at this moment, the trend is undeniable:
The most valuable real estate in the future won’t be prime office towers in the CBD.
It will be those vacant lots in the outskirts, right next to power substations.
Top comments (0)