Premium

The Real AI Trade: These Stocks Are Powering A $220 Billion Boom

The artificial intelligence boom is widely viewed as a story about chips, models, and software. But the real constraint - and increasingly, the real opportunity - is far more physical.

AI does not run on code alone; it requires megawatts of continuous, ultra-reliable power. And right now, the global economy is in the midst of the largest datacenter build-out in history to deliver it. To that end, RBC has compiled a 67-page deep dive on companies driving this boom - which pro subs can read here, or keep reading for the short and sweet version as a ZeroHedge Premium subscriber.

Le Situation

Hyperscalers are ramping capital expenditures at unprecedented rates, pouring hundreds of billions into datacenters designed to support a new generation of compute-intensive workloads. But as rack densities surge toward megawatt scale and power demand outpaces grid capacity, the limiting factor is no longer semiconductors - it is power delivery, thermal management, and the physical systems that keep these machines running. Nothing you haven't heard from us before, a few times.

RBC forecasts 15%+ per annum growth in the relevant electricals/HVAC addressable market through 2030 (with recent actuals at 30–50% and Vertiv guiding ~30% for 2026), driven by hyperscaler capex surging ~60% y/y and rack densities exploding toward 1MW+.

The total opportunity has tripled since 2024 estimates: a theoretical 400MW datacenter costs ~$11bn, with 29% (~$3.2bn) tied to electrical/HVAC equipment — implying a $220bn annual addressable market for 2026–2030E (vs. $113bn in 2025). Key components include UPS (16%), CRAH (12%), switchgear (11%), and generators (19%). Cooling (especially liquid) and power infrastructure are the biggest growth vectors.

800VDC architecture is the next evolution (phased over 3–5+ years; only ~20% penetration by 2030). It delivers ~3× power per cable, ~95% efficiency (vs. 80–85% today), and better grid stability via BESS. Disruption risk is concentrated in gray space (especially UPS), but leaders are adapting via organic/bolt-on moves and partnerships. Overall, 800VDC is net upside, not a threat.

But what about ORBITAL DATA CENTERS you might ask? According to RBC, not a current concern to the thesis until costs of doing so fall at least 50%.

Here's what powers a data center

Transformer: Steps down power from the medium-voltage distribution grid to the low-voltages required for usage within the facility. Smaller datacenters will tend to have a single medium-voltage transformer as electricity enters the center, while hyperscale facilities (typically 10MW+) may have a multi-transformer set-up, in which the voltage is only stepped down locally – this has efficiency and infrastructure-simplicity benefits.
Main Switchgear: Facility ‘circuit breaker’, protecting the rest of the components downstream from overheating/damage in event of power surges. A facility may also feature a number of other switchgears further down the circuit/closer to IT systems, depending on design.
Automatic Transfer Switch (ATS): A device that transfers electrical loads between power sources (i.e., from the mains to a back-up generator).
Critical Power Busduct: Sheet metal ducts that facilitate power distribution inside a facility. May also be achieved with busbars. Both are typically superior to electrical cabling/wiring given cost, electrical properties and installation ease.
Uninterruptible Power Supply (UPS): Provides instantaneous emergency backup power in the event of disruption to the electrical mains, ensuring IT equipment uptime. Typically provides power for 5–20 minutes, until a back-up generator can be activated. May operate via a battery pack, capacitor or other designs.
Power Distribution Unit (PDU): Distributes electrical power within the IT room, with additional responsibilities for systems monitoring and control. Depending on design, PDUs may have a final voltage/current step-down integrated into it – simpler PDUs will otherwise largely resemble a power strip.
IT Rack: The end-user computing set-up, with the processing units typically housed in either a metal rack or cabinet. The design of the rack/cabinet plays a role in facilitating datacenter cooling infrastructure.
Cooling Systems: Cooling systems help to protect IT equipment from excessively high temperatures, given almost 100% of IT equipment energy usage eventually becomes heat. In lower wattage rack set-ups (i.e., 10kW/rack or below) cooling systems are predominantly air-powered; higher-powered racks (i.e., those with more excess heat generation) will typically include an element of liquid-cooling (i.e., immersion, heat exchangers, direct-to-chip).
Back-up generator: A stand-by power source, providing critical facility functions with emergency back-up power in the event of a mains power outage. Typically, back-up generator uptime is within 2 minutes of facility outage. Most commonly a diesel-powered generator.
Non-critical power: Typically non-IT equipment related electrical loads. This may include lighting and power to personnel facilities. These are not necessarily re-activated by the back-up generator.

And here are RBC's core investment ideas in the space

Vertiv: The Closest Thing to a Pure-Play AI Infrastructure Company

Vertiv has emerged as the most direct way to gain exposure to the physical buildout of AI datacenters. Unlike diversified industrial peers, the company is almost entirely focused on digital infrastructure—spanning power management, thermal systems, rack integration, and lifecycle services. Its product stack includes everything from UPS systems and switchgear to liquid cooling and modular datacenter solutions, allowing it to act as a full-system integrator rather than a component supplier.

This positioning has translated directly into operating momentum. Orders have surged dramatically (as noted in your RBC data), and the company has reinforced its leadership through deep partnerships with Nvidia and Intel, particularly around next-generation liquid cooling systems required for high-density AI clusters. These collaborations are not incidental—they are central to the industry’s shift toward direct-to-chip and immersion cooling, which is rapidly becoming mandatory as AI racks exceed traditional thermal limits.

What distinguishes Vertiv structurally is its role as a reference architecture partner. Industry analysis shows it is helping define how modern AI datacenters are physically designed, particularly for Nvidia-based systems. This places Vertiv not just in the supply chain, but in the design layer of the AI infrastructure stack—a far more valuable position.

The tradeoff, however, is concentration. Compared to Schneider or Eaton, Vertiv is more exposed to datacenter cycles, making it both a high-beta beneficiary of AI capex and potentially more volatile over time.

Eaton: The Power Backbone Expanding into Cooling

Eaton’s evolution reflects one of the most important strategic shifts in the sector: the convergence of power delivery and thermal management. Historically a leader in electrical systems—switchgear, circuit protection, and power distribution—Eaton has aggressively expanded into cooling, most notably through its acquisition of Boyd Thermal.

That move is widely viewed as transformative. Boyd brings advanced liquid cooling technologies—cold plates, CDUs, and heat exchangers—that are essential for AI workloads, where heat density is now one of the primary constraints on compute scaling. By integrating these capabilities with its existing electrical portfolio, Eaton is positioning itself as a full-stack infrastructure provider, similar to Vertiv but with a broader industrial base.

Strategically, Eaton benefits from diversification. Its legacy businesses provide stability, while datacenter exposure—now roughly a quarter of revenue—offers high-growth upside. This balance allows it to scale aggressively into AI infrastructure without the same cyclicality risks as more concentrated peers.

Importantly, Eaton is also at the forefront of the 800VDC transition, working directly with Nvidia on new power architectures. This suggests it will play a central role in redefining how electricity is delivered inside next-generation datacenters, where efficiency gains and power density are becoming critical constraints.

Schneider Electric: The Global Systems Integrator

Schneider Electric occupies a unique position as arguably the most comprehensive and globally entrenched provider of datacenter infrastructure. Its offerings span the full stack: electrical systems, cooling, racks, software (EcoStruxure), and lifecycle services, allowing it to act as a long-term partner to hyperscalers rather than a transactional vendor.

This breadth has translated into large-scale contracts and sustained growth. The company has secured multi-billion-dollar datacenter deals in the U.S., supplying both power modules and cooling systems as AI deployments accelerate. At the same time, Schneider is investing heavily in liquid cooling and co-developing solutions with Nvidia to support increasingly power-dense chips, where traditional air cooling is no longer viable.

Where Schneider differs from peers is its emphasis on software and services integration. Its platforms provide real-time monitoring, predictive analytics, and energy optimization—capabilities that are becoming increasingly important as datacenters evolve into complex, energy-intensive “AI factories.”

While execution has been somewhat uneven relative to faster-growing peers, Schneider’s scale, global reach, and installed base make it one of the most structurally advantaged companies in the space—particularly as infrastructure becomes more standardized and lifecycle-driven.

nVent: The Quiet Winner in High-Density Infrastructure

nVent has rapidly emerged as one of the most underappreciated beneficiaries of the AI datacenter boom, particularly in the white-space and liquid cooling layers. Its core business—rack power distribution, enclosures, and thermal management—places it directly inside the datacenter, where rising rack density is driving exponential increases in component complexity.

Recent performance reflects this positioning. The company has seen strong order growth, expanding backlog, and rising investor interest as AI-driven demand accelerates. More importantly, nVent has spent years developing liquid cooling capabilities, including rear-door heat exchangers and immersion systems, which are now moving from niche to necessity.

Strategically, nVent sits at the intersection of two major trends:

Electrification of infrastructure
Thermal constraints at the rack level

This combination gives it leverage to both power and cooling spend, particularly as datacenters move toward higher-density configurations and wider racks.

Unlike Vertiv or Schneider, nVent is less of a full-system integrator and more of a critical subsystem provider. But in an environment where content per rack is rising sharply, that positioning may prove equally valuable.

Cooling Specialists: Carrier, Trane, Johnson Controls, Modine

The cooling layer is rapidly becoming the bottleneck - and therefore the profit pool - of AI infrastructure. Cooling already accounts for a significant share of datacenter energy consumption (often 30–40%), and that percentage is rising as AI workloads scale.

Traditional HVAC companies - Carrier, Trane, and Johnson Controls - are aggressively repositioning themselves to capture this opportunity. All three are expanding beyond legacy air cooling into liquid and hybrid cooling systems, often through acquisitions and partnerships. They are also leveraging their global service networks, which can generate multiples of initial equipment revenue over time.

Modine stands out as a more focused growth story. Following its portfolio restructuring, the company is effectively becoming a pure-play thermal management provider, targeting extremely high growth rates tied to AI datacenter demand. Its emphasis on immersion and liquid cooling aligns directly with where the industry is heading.

Collectively, these companies represent the thermal backbone of AI infrastructure, with growth increasingly tied to how quickly hyperscalers adopt liquid cooling at scale.

Power Generation & Grid Enablers: Caterpillar and GE Vernova

While much of the focus is inside the datacenter, a critical constraint lies outside it: power availability.

Caterpillar and GE Vernova play essential roles here by supplying backup and primary power systems, including gas turbines and large-scale generators. As AI datacenters push into regions with constrained grids—or require faster time-to-power—these solutions are becoming indispensable.

GE Vernova, in particular, has seen a surge in demand for gas turbines, with datacenters accounting for a significant portion of future capacity reservations. This reflects a broader shift toward on-site or dedicated power generation, especially as grid interconnection timelines stretch into years.

Both companies are also investing in future-facing technologies such as solid-state transformers and fuel cells, which could become increasingly relevant in a world transitioning toward DC-based architectures.

Emerging Layer: Distributed Power and Storage (Bloom Energy, Fluence)

A newer layer of the ecosystem is forming around distributed energy and storage, driven by the growing mismatch between datacenter demand and grid capacity.

Bloom Energy is a key player here, providing fuel-cell-based on-site power generation that allows datacenters to bypass grid constraints. Its systems are already being deployed at scale and are designed to support future DC architectures, making them highly relevant to the 800VDC transition.

Fluence, meanwhile, operates in battery storage, offering systems that improve grid stability, load balancing, and backup power. While its datacenter exposure is still emerging, its technology is particularly well-suited to DC environments, where eliminating conversion losses becomes increasingly valuable.

Stock Symbols

Preferred / Content Leader Names

Vertiv → VRT (NYSE)
Eaton → ETN (NYSE)
nVent → NVT (NYSE)
Schneider Electric → SU (NYSE ADR / Euronext: SU.PA)

Full List of All Companies Profiled in the Report

(Alphabetical, covering every company in the 32-page “Company profiles” section)