The AI boom has a familiar look from the outside: gleaming data, flashy demos, and the relentless race for faster chips. But if you zoom out from the headlines and follow the money to where it physically lands, another story emerges—one measured in concrete pours, electrical substations, steel deliveries, and the unglamorous logistics of getting power and cooling to sites that didn’t exist a few years ago.
That’s why “picks and shovels” companies—businesses that supply the infrastructure rather than the software—have started to draw sustained attention from investors. The latest market moves underline the point. Industrial and construction-linked names that were once treated as slow, cyclical, or merely defensive are increasingly being reframed as beneficiaries of a structural buildout cycle tied to artificial intelligence. In this new narrative, the AI spotlight doesn’t just illuminate compute; it also illuminates the supply chain that makes compute possible.
At the center of the shift is the data centre scramble. AI workloads are not simply “more computing.” They are different in scale, intensity, and timing. Training runs demand large bursts of power and high-density cooling. Inference at scale then turns those same facilities into near-constant engines. The result is a wave of new builds and expansions that is pushing demand beyond what traditional capacity planning would have predicted.
And when demand accelerates like that, the winners are often the firms that can deliver the physical reality on schedule.
From chips to sites: why the infrastructure story is different this time
For years, data centres were built with a relatively predictable rhythm: incremental capacity additions, steady improvements in efficiency, and long-term leases that smoothed demand. Even during earlier technology cycles, the industry’s bottleneck was often framed as software, networking, or semiconductor supply.
AI changes the bottleneck map. The limiting factors increasingly include:
Power availability and grid connection timelines
Cooling capacity and heat-rejection design
Construction throughput—how quickly contractors can mobilize labor, equipment, and materials
Specialized engineering for high-voltage distribution, redundancy, and resilience
Permitting and site readiness, including land acquisition and environmental approvals
These constraints don’t disappear because a model becomes more efficient. In many cases, efficiency improvements reduce power per unit of compute, but total compute demand rises even faster. That means the industry still needs more sites, more electrical infrastructure, and more mechanical systems.
So while the market may obsess over the next generation of accelerators, the real-world question becomes: how fast can the world build enough rooms to house them?
This is where industrial “enablers” come in. They aren’t selling the AI product. They’re selling the ability to construct and equip the environment in which AI products run.
Caterpillar and Hochtief: the logic behind the re-rating
In the current market conversation, companies such as Caterpillar and Hochtief have been cited as examples of once-staid industrial stocks lifted by the broader AI boom. The appeal is straightforward: data centre buildouts require excavation, earthworks, heavy construction, and ongoing site development support. When the number of projects rises, so does the demand for the machinery and contracting capacity that can execute them.
Caterpillar, for instance, sits in a category of businesses that investors often treat as cyclical—sensitive to construction activity, mining cycles, and general industrial capex. But the data centre wave adds a new layer to that cyclicality. It’s not just about whether construction is happening; it’s about whether a specific kind of construction is accelerating in a way that requires heavy equipment and sustained operational tempo.
Data centres are not simple warehouses. They involve complex foundations, large-scale earthworks, and extensive infrastructure work around the building footprint. They also require careful sequencing: electrical installations, mechanical systems, and commissioning all need to align. That sequencing tends to extend project timelines and increases the amount of on-site activity relative to simpler commercial builds.
Hochtief, meanwhile, represents the contracting and construction side of the equation. Contractors are often judged on margins, backlog quality, and execution risk. But in a buildout cycle driven by urgent capacity needs, the market begins to reward firms that can secure and deliver projects reliably. Data centres also tend to be capital-intensive and technically demanding, which can favor contractors with strong engineering capabilities and experience managing complex stakeholders—utilities, regulators, and specialized subcontractors.
The “picks and shovels” framing isn’t claiming these companies are immune to downturns. It’s arguing that the AI-driven buildout cycle may provide a different kind of demand floor—one tied to long-duration capacity commitments rather than short-lived consumer trends.
Why investors are shifting from “AI winners” to “AI enablers”
There’s a psychological element to the shift. When markets chase a theme, they often start with the most visible beneficiaries. Chips, cloud platforms, and AI software are easy to understand and easy to price. But as the theme matures, investors begin to ask a harder question: what happens when the bottleneck moves?
In the early phase of AI adoption, the bottleneck was often assumed to be compute supply. Now, the bottleneck increasingly includes the physical infrastructure required to deploy that compute at scale. That’s a different investment thesis—less about product differentiation and more about execution capacity across the supply chain.
This is why “enablers” can attract attention even if their business models don’t change dramatically. The underlying demand driver changes. A contractor’s order book can improve not because the economy suddenly booms, but because a specific sector—data centres—ramps up with urgency.
There’s also a portfolio-management angle. Many investors want exposure to AI without concentrating entirely in a handful of high-multiple tech names. Industrial and construction-linked stocks can offer a different risk profile: more tangible revenue drivers, different valuation dynamics, and sometimes a clearer relationship between backlog and future earnings.
However, the shift also comes with risks that investors should not ignore.
The hidden costs of building fast: supply chains, labor, and grid realities
A data centre buildout is not just a construction story. It’s a systems story. The most expensive and time-consuming parts are often the ones that don’t show up in glossy renderings.
Power is the headline constraint. Even when a site is ready, connecting to the grid can take months or years depending on local capacity, permitting, and utility planning. Some developers respond by pursuing on-site generation, battery storage, or alternative power procurement strategies. But those solutions add cost and complexity—and they can become a bottleneck of their own.
Cooling is another constraint. High-density racks generate heat that must be managed efficiently. Liquid cooling, advanced air handling, and heat-reuse strategies are increasingly discussed, but implementation varies widely by region and by developer. Cooling systems also require specialized components and engineering expertise.
Then there’s the labor and equipment question. Heavy construction depends on skilled operators, electricians, mechanical engineers, and commissioning teams. If multiple data centre projects compete for the same talent pool, costs rise and schedules slip. Machinery availability can also become a factor, especially when global construction demand is elevated.
Materials and logistics matter too. Steel, transformers, switchgear, and other electrical components can face lead-time pressures. Even if a contractor has the right equipment, delays in critical components can stall progress.
All of this means that “AI infrastructure demand” is not a straight line. It’s a dynamic process shaped by constraints. The companies that benefit most are often those that can manage these constraints—through procurement discipline, supplier relationships, and execution capability.
That’s why the market’s interest in industrial and construction firms is not merely about the existence of data centres. It’s about the ability to deliver them under pressure.
The new competitive landscape: who captures value in the buildout
Data centre value capture is increasingly distributed across the ecosystem. Developers and operators build the facilities, but they rely on a web of suppliers and contractors. The “picks and shovels” thesis suggests that some of the value accrues to firms that provide:
Equipment and machinery for site preparation and construction
Engineering and construction services for complex builds
Electrical and mechanical components and installation capacity
Logistics and project management expertise
Long-term maintenance and upgrades as facilities evolve
But not all “infrastructure” exposure is equal. Some companies are exposed to the buildout directly through construction contracts. Others benefit indirectly through equipment sales or service work. The timing of revenue recognition can differ significantly.
Construction-linked businesses may see revenue tied to project milestones, which can create volatility if schedules shift. Equipment suppliers may experience demand spikes but also face normalization if the buildout cycle slows. Service providers may benefit from longer-duration recurring work, but their growth can depend on how quickly facilities move from commissioning to full operations.
Investors looking for “AI enabler” exposure should therefore pay attention to business model specifics: backlog composition, contract terms, geographic exposure, and the degree to which demand is concentrated in data centres versus broader industrial activity.
A unique take: AI is turning infrastructure into a strategic asset class
One of the more interesting shifts in the market is how data centres are being treated less like real estate and more like strategic infrastructure. In earlier eras, data centres were often viewed as property plays—important, but not necessarily central to national competitiveness.
Now, the facilities are increasingly tied to economic and technological sovereignty. Governments and large enterprises care about latency, resilience, and energy security. That changes how projects are funded and how quickly they are prioritized.
When infrastructure becomes strategic, it tends to attract different forms of capital and different urgency. That can accelerate buildouts and increase the importance of reliable contractors and equipment providers.
This is also why the “picks and shovels” narrative resonates. It’s not just that AI needs more data centres. It’s that data centres are becoming a core component of the AI supply chain—analogous to how ports, rail, and power generation became strategic assets in earlier industrial transitions.
In that context, industrial firms that can support rapid deployment gain a new kind of relevance.
What could go wrong: the risks behind the