AI’s expansion runs on smaller companies

Key takeaways

• Artificial intelligence’s (AI’s) expansion is not solely a mega-cap story—it is a multi-year infrastructure cycle supported by smaller companies building, powering and equipping the data center ecosystem.
• From mechanical construction and distributed power generation to semiconductor and electrical componentry, smaller-cap specialists are supplying the essential inputs enabling hyperscaler AI deployment.
• Investing beyond the headline-grabbing model developers may provide differentiated exposure to sustained AI-driven capital expenditure across construction, energy and enabling technologies without having to bet on a singular technology, developer or architecture.

AI buildout is not solely the purview of mega-caps

AI has become closely associated with a small group of mega-cap technology companies and large language model (LLM) developers. However, while headlines and capital markets have thus far been driven by hyperscalers and chip designers, the reality is that the scaling of AI capabilities depends on a far broader ecosystem.

As a generational investment in graphics processing units (GPUs) and data centers is occurring to fuel LLM developments, we are witnessing a compressed infrastructure cycle to support this buildout. The expansion of compute capacity requires sustained investment across a number of different areas including physical plant construction, power generation, heating  and cooling solutions, and increasingly complex electrical and semiconductor systems. As record-breaking capital expenditure announcements continue to surpass expectations, those investments are being spent by and with companies operating well outside the mega-cap universe.

Building the physical backbone of AI

The most visible evidence of the AI investment cycle is the rapid construction and retrofitting of advanced data center facilities, a critical and technically demanding task due to higher-performance servers requiring effective cooling of their significant thermal loads not to mention the networking capabilities to support high data bandwidth needs and the ample electricity to run power-hungry GPUs. As a result, mechanical and electrical contractors with specialized and scarce expertise are central to the data center buildout.

As data centers grow in number, scale and complexity, smaller, more specialized companies such as Comfort Systems, which provides HVAC and mechanical systems installation and maintenance, will likely find themselves able to exercise selectivity, maintain pricing discipline and expand margins.

Power generation represents a critical bottleneck

While construction services, particularly in the mechanical, HVAC and electrical realms are in short supply, securing adequate power for these new or revamped AI data centers is proving to be equally challenging for hyperscalers and LLM developers. AI training and inference workloads are materially increasing in power demand and intensity. As hyperscalers expand the number of data centers along with their regional footprints, electricity availability is becoming a gating factor in deployment timelines, with grid interconnection delays and transmission bottlenecks already presenting increasingly visible constraints (Exhibit 1).

The AI cycle is not only increasing demand for computing capacity—it is reshaping the energy landscape that supports it. Given the challenges in securing traditional grid interconnection, smaller-cap companies developing new and innovative alternative energy sources (fuel cells, gas turbines, nuclear) are increasingly being looked at as key partners to provide structural solutions to these challenges. For example, Bloom Energy’s on-site solid oxide fuel cell systems provide reliable and clean energy directly at the point of consumption, reducing reliance on traditional grid expansion. In high-demand environments, this ability to provide power locally with fast start-up times can accelerate data center deployment.

At the same time, broader investment in nuclear and advanced generation technologies reflects rising recognition that a long-duration baseload power system will be required to support structural electricity growth. Hyperscalers like Microsoft, Amazon and Meta Platforms have all signed long-term agreements with nuclear energy producers to secure carbon-free power for their new data centers and are also investing in small modular reactors development. This positions companies such as BWX Technologies, with its long track record of producing nuclear components and services (since the 1950s) and emerging business lines in managing complex reactor operations and security, at the forefront of this longer-term shift toward a more modernized and scalable power mix.

Unprecedented capex and complexity opens opportunity for specialized solutions

Beyond construction and energy generation, the rapid scaling of AI is driving system complexity, which has opened up opportunities for providers of specialized semiconductor chips, connectivity solutions and electrical systems.

These high-performance computing environments require increasingly precise power management, signal integrity and connectivity solutions—a dynamic benefiting a range of analog and control-oriented semiconductor companies not initially viewed as direct AI participants. This includes companies like Lattice Semiconductor and Allegro MicroSystems, both of which provide semiconductor chips that address rising power density and control requirements in data center and high-performance applications. Through longstanding investment in innovative technology solutions, these companies have begun to win share in next-generation compute architectures.

Similarly, companies such as Regal Rexnord, which supplies engineered power transmission and motion control technologies, are participating in broader electrification and industrial automation trends that intersect with AI-driven capital deployment.

Smaller companies can provide differentiated and model/architecture agnostic AI exposure

Some of the most compelling AI-related opportunities reside in companies supplying enabling technologies rather than those developing models. As compute architectures grow more sophisticated, the demand for components and systems from “picks and shovels” providers, oftentimes smaller-cap companies, is increasing concurrently.

While the AI narrative remains dominated by mega-cap hyperscalers and increasingly, private LLMs, the underlying capital investment supporting this growth is far more distributed. Data center construction, power generation expansion and semiconductor complexity all represent multi-year structural themes where we believe smaller companies are exceptionally well positioned as key partners and contributors to the AI ecosystem’s success.

For investors, applying this broader infrastructure lens may offer differentiated exposure to the AI investment cycle. Rather than relying solely on application-layer adoption or model leadership, the opportunity extends to businesses with durable competitive advantages positioned within sustained capital expenditure cycles. Moreover, these smaller, picks and shovels companies can offer a more diversified approach to gaining exposure to the equity market’s dominant investment theme.

The information provided is not a recommendation to purchase, sell, or hold any particular security and should not be construed as an endorsement of or affiliation with Franklin Templeton.