Office construction in the US is collapsing. Data center construction just hit sixty billion dollars and is headed higher. Amazon, Microsoft, Google, and Meta are collectively spending three hundred eighty billion dollars this year building the warehouse-sized buildings that run AI. That money has to go somewhere. It goes to construction companies. The ones you can actually buy stock in are reporting record backlogs, raising guidance, and trading like nothing special is happening. The chart shows the divergence. Most investors are still treating this as a tech story. It is a construction story now.
I.
Look at the chart. General office construction peaked in 2021 and has been falling ever since. Data center construction has gone vertical. The lines crossed in 2023. The gap keeps widening.
Data center spending hit sixty billion dollars in 2025, up from twelve billion in 2023. Monthly spending peaked at fourteen billion in July, which was more than the entire first half of the year. The average project now costs six hundred million dollars. That number was three hundred seventy million a year ago. These are not small buildings. Meta's Louisiana facility has a fifty billion dollar price tag attached to it.
The spending comes from four companies. Amazon is spending one hundred twenty five billion this year. Microsoft is spending eighty billion. Google is spending ninety three billion. Meta is spending seventy two billion. These are not forecasts or estimates. These are the numbers they reported on their earnings calls. They also said they are spending more in 2026.
That money does not stay in Silicon Valley. It lands in Louisiana, Texas, Virginia, Mississippi, and Pennsylvania. These states accounted for seventy four percent of all new data center projects through November. Someone has to build them. Someone has to wire them. Someone has to connect them to the power grid and the internet. Those companies are public. You can buy them. Most people are not paying attention yet.
II.
Amazon's CEO said demand is outstripping supply as fast as they can add capacity. Microsoft's CFO said their investments reflect business already booked. These companies are not building speculatively. They have customers waiting for computing power they cannot provide today. The buildout is solving a capacity problem, not betting on future growth.
AI requires vastly more computing power than anything that came before it. Training a single large language model costs tens of millions of dollars in electricity and hardware time. Running ChatGPT for hundreds of millions of users requires facilities that did not exist two years ago. The hyperscalers are racing to build them because not having capacity means losing customers to whoever builds faster.
This is not optional spending. Meta does not even have a cloud business. The company is spending seventy two billion dollars anyway because its advertising engine runs on AI models that need this infrastructure. Pulling back means falling behind competitors who will not hesitate. That is not an option when you are competing for dominance in the next generation of computing.
The construction firms building these facilities are booking work years in advance. Backlogs are at record levels. A typical data center takes eighteen to twenty four months to build. Projects breaking ground in 2025 generate revenue through 2027. The timeline creates visibility most construction companies never get.
III.
Most of the big names are private. Turner Construction, Holder Construction, and DPR Construction are among the largest data center builders in the country. None of them trade publicly. The companies you can actually buy tend to do specialized work. Electrical systems. Fiber networks. Heavy infrastructure. They are not building the data centers themselves. They are building the pieces that make data centers work.
Quanta Services trades as PWR. The company does electrical infrastructure and power systems. Every data center needs massive electrical capacity and backup systems. Quanta did twenty five billion in revenue last year. The stock is up seven hundred percent over five years. Much of that came after the company started winning large data center contracts.
Dycom Industries is DY. They build fiber optic networks. Data centers need to be connected to the internet with high capacity fiber. Hyperscalers are also building their own fiber between facilities. Dycom's backlog hit a record in 2025. Management said the growth came from AI-related network builds. They expect earnings to grow thirty five percent in fiscal 2027.
MasTec is MTZ. The company does communications infrastructure, power work, and energy projects. Their communications segment did nine hundred fifteen million in revenue last quarter, up thirty three percent from a year earlier. Backlog is sixteen point five billion. Five billion of that is communications work. Management said hyperscaler spending on connectivity is driving the growth. The stock is up sixty one percent this year.
EMCOR Group trades as EME. They do electrical and mechanical construction. Their backlog nearly doubled year over year, driven by data center work. Sterling Infrastructure is STRL. They do site development and heavy civil work. Both companies are benefiting from the same spending wave but at smaller scale.
These are not sexy names. They are not the companies building AI models or designing chips. They are the companies pouring concrete and pulling wire. That work is less visible but it is extremely profitable when demand overwhelms supply and customers are willing to pay for speed and reliability.
IV.
The spending could slow. If AI turns out to be less profitable than tech companies expect, they could dial back capital expenditures faster than construction backlogs adjust. These firms have signed contracts for near term work, but future bookings would dry up. Companies trading at high multiples based on growth would reprice quickly.
Some regions are hitting power limits. Northern Virginia is the largest data center market in the country and it is running out of electricity. If utilities cannot expand capacity fast enough, projects get delayed or moved to other states. Construction firms do not control the power grid but they get hit when projects stall.
Competition is increasing. Record profits are attracting new players. More firms are bidding on data center work. As competition rises, pricing power erodes. Margins compress. The biggest contractors have technical expertise and relationships that create some advantage, but construction is fundamentally competitive. High margins do not last forever.
Execution risk exists. Building data centers is more complex than typical commercial construction. Mistakes are expensive. Projects can run over budget or miss deadlines. Contractors without deep experience in this type of work face higher risk of problems that hurt profitability.
V.
The chart shows a structural shift in where capital is flowing. Office construction is not coming back. Data center construction is not slowing as long as tech companies keep raising their spending guidance. The construction firms on the right side of that shift are generating returns that look nothing like normal construction economics.
Most investors still think of this as a technology trade. Buy the chip companies. Buy the cloud providers. That trade has worked but it is crowded. The construction angle is less obvious. These companies do not benefit from AI directly. They benefit from the three hundred eighty billion dollars that has to get spent building the physical infrastructure AI runs on.
The firms with heavy data center exposure are trading like cyclical industrials. They should probably be trading like infrastructure plays with multi year revenue visibility and creditworthy customers. The market will figure that out eventually. It always does when the earnings keep beating and the guidance keeps rising.
The runway extends into 2027 at minimum. Hyperscalers have already committed to higher spending in 2026. Projects in preconstruction now will generate revenue in 2028. The work is sold. The question is not whether these companies will grow. The question is how long the growth lasts and whether current valuations already reflect it.
For now, the chart keeps diverging. The companies building what tech needs are printing record results while the rest of construction muddles through. That gap is worth watching.
—
This analysis is for educational purposes. It does not constitute investment advice or a recommendation to buy or sell any security. Investors should conduct their own due diligence and consult financial advisors.