The AI data center construction trends shaping the next decade

As AI adoption accelerates, many contractors are asking, “When does the music stop for data center investment?” The question within the question is, how much of the AI demand passes through construction budgets going forward?

The near-term outlook for AI data center construction is strong, with Goldman’s leading indicators near the top of their range and capex forecast to keep rising through 2028. But the long-range forecasts are built around equipment refresh cycles more than construction starts, and the construction share of total spend narrows every year of the model.

To better understand where the market is headed, here’s an overview of the factors shaping the long-term outlook for mission-critical construction:

Is data center construction slowing down?

The long-term outlook on AI spending and data center demand looks largely positive:

• Goldman Sachs Research puts global AI investment above $1 trillion in 2026, $581 billion of it in the U.S., and sees U.S. AI capex rising from 1.8% of GDP to 2.8% by 2028. Prior technology build-outs peaked at 2% to 5%.
• Barron’s found that across 250 years of U.S. capex booms, transformational spending has reached roughly 25% of GDP before breaking. On that math, the AI cycle runs into the early 2030s, and the historical pattern is that these booms end on an outside shock rather than on investors questioning them.
• PwC’s central case models $31.6 trillion of global data center capex through 2050, $15.1 trillion of it in the U.S., with annual spend climbing from about $800 billion this year to $1.8 trillion by 2050.

The catch? PwC estimates every dollar of construction capex commits the market to roughly $12 of future equipment spend. Servers, GPUs and networking gear are 70% of data center capex today and reach 93% by 2050, because the hardware turns over every four to six years. The asset is a chip-replacement subscription with a building wrapped around it.

What that means for the trades is that the building is the entry ticket to a much longer stream of spending that occurs inside it. The cycle keeps growing for decades because the hardware keeps aging out, and an increasing share of the growth accrues to whoever is installing, powering and cooling that hardware rather than pouring the foundation and building the envelope for new buildings

What obstacles are there to new data center construction?

There are two main obstacles to new data center projects:

• Power: Transformer lead times, substation availability and interconnection queues set the order in which projects break ground, regardless of how well capitalized the developer is.
• Local politics: The Wall Street Journal reported that the fight over data centers has widened from electricity bills, water and noise to jobs. Politicians on both sides have paused approvals ahead of the midterms, and the building trades are now telling them that political support is conditional on backing the projects. One steamfitters local outside Washington called it an existential moment.

What are the risks of data center construction projects?

Strong demand does not eliminate risk. Contractors should look beyond current demand and consider how the industry is changing, including potential risks such as:

A concentration in one-time work.

The contractors carrying the most risk are those that have let their backlog fill with new construction shells, the part of the cycle that happens once per building, while scaling the safety infrastructure that mission-critical work demands to support current volumes. Safety programs, training, and site controls at that standard are a real investment, and they’re being built for a project mix that PwC’s model says is the least recurring piece of the spend.

Mission-critical contractors have also enjoyed pricing power the rest of the industry hasn’t, passing cost volatility through on change orders. That leverage belongs to the scarcity of the moment, and it should be treated as temporary.

A fragmenting buyer base.

The cloud era had one buyer type. There could now be six, including neoclouds and model developers whose demand depends on chip access and on AI revenue that hasn’t fully arrived.

Barron’s adds that every prior capex boom ended with the market consolidating into a handful of winners, which means some of today’s counterparties won’t be around for the refresh.

A lagging view of the cycle.

Goldman tracks a dashboard of leading indicators for AI capex, including semiconductor equipment imports into Taiwan and South Korea, GPU rental prices and PMI components. All of them currently sit near the top of their range since 2022.

A backlog report will tell you the cycle has turned well after those indicators do.

What are the opportunities for data center projects?

For contractors, the greatest opportunity may be in viewing data centers as sources of recurring and reoccurring revenue, rather than one-time projects. Some considerations for forward-thinking contractors include:

The building gets rebuilt from the inside.

PwC assumes a 20-year asset life with an equipment refresh every four to six years, which works out to three to five rounds of ICT investment per facility. Each refresh raises rack density, and they note that higher density frequently forces power and cooling upgrades with it.

That’s mechanical and electrical scope recurring on a schedule for as long as the building stands, and it’s the side of the model that grows.

The work is moving closer to the user.

Roughly 30% of AI workloads already carry requirements to be processed locally, for latency, privacy or sovereignty reasons, and that share is growing quickly. Training clusters chase cheap power wherever it sits. Inference follows population, which points to more facilities across more markets, including secondary ones where regional contractors already have relationships.

Sponsors are bringing their own power.

Operators are increasingly adding on-site generation to get around interconnection delays. That expands scope on the utility side of the fence and moves the electrical trades further into the critical path of every project.

The first contract is the start of the relationship.

A contractor who builds the shell and then earns the refresh, retrofit and power infrastructure work on the same building for the next two decades has a very different revenue profile than one who builds the shell and moves on. That requires the same discipline leading firms apply everywhere else: core markets, trusted delivery models and the customers where the relationship is strongest.

Bidding the initial work with that in view changes what you’re willing to invest in capability.

What the data center building outlook means for you

The music doesn’t appear to be stopping. Goldman, Barron’s and PwC each land on AI investment that keeps growing into the next decade. For mission-critical contractors, that means a pipeline of new shells, campuses and power infrastructure that runs at least through 2028 and plausibly into the early 2030s.

What changes over that horizon is where the dollars go:

• The equipment inside gets replaced every four to six years.
• Sites get chosen by power availability and local approval before anything else.
• The trades that install, power and cool the equipment come back to the same building long after the shell is done.

The contractors who do well through the full cycle will be the ones who manage their mission-critical customer relationships the way they manage backlog: as strategic assets, managed aggressively and with pace.

The question to take into your next planning meeting isn’t how much data center backlog you have. It’s this: how much downsell risk is there within your existing customer base, and how do you retain market share of the recurring demand that will arise from those same customers in the future?

Read more:

• 5 ways contractors can navigate interest rate uncertainty
• Why construction firms need an outsourced CFO to grow
• How data lakehouses can power AI in construction