AI Investment Is Expanding Into the Physical Infrastructure Economy

Artificial intelligence increasingly appears to be a physical infrastructure story as much as a technological one. Behind every new model, algorithm, and GPU cluster often sits a substantial network of land, buildings, cooling systems, power equipment, and industrial capacity. JLL’s North America Data Center Report noted 25 gigawatts of data-center absorption during the first half of 2026, roughly twice the level recorded a year earlier. That scale suggests an expanding connection between digital capability and the physical assets required to support it.

That connection is changing the character of the AI buildout. Technology companies and their infrastructure partners may increasingly find themselves dealing with industrial corridors, large parcels of land, specialized construction, cooling systems, substations, transmission equipment, and machinery with lengthy production timelines. JLL reported that 66 gigawatts of North American data-center capacity was under construction by mid-2026, with 95% of that pipeline already pre-committed. Such figures point toward an environment in which securing suitable physical capacity can become an important consideration alongside computing hardware and software development.

Power can add another layer to that equation. According to MGRID’s analysis of the PJM capacity market, the 2026/2027 capacity auction cleared at $329.17 per megawatt-day, compared with $28.92 in the 2024/2025 auction. MGrid also reported that PJM was forecasting 5 to 7 gigawatts of annual data-center load additions through 2032. These figures suggest why electricity availability can increasingly influence where new computing capacity becomes practical.

The scale of that dependency may extend beyond individual markets. Lawrence Berkeley National Laboratory projections cited by MGRID put U.S. data-center electricity consumption at between 325 and 580 terawatt-hours by 2028, compared with 176 terawatt-hours in 2023. Goldman Sachs has projected a 175% increase in global data-center power demand between 2023 and 2030. As demand expands, investors may find greater relevance in the infrastructure surrounding computation, including generation, transmission, land, water, equipment, construction, and logistics.

For Sanjay Raghavaraju, founder and CEO of 33 Holdings, this changing infrastructure landscape presents a question about how physical assets are evaluated in relation to one another. His perspective comes from a firm whose investment history began in real estate and has expanded across residential, land development, industrial, and energy-related assets. 

That experience informs a broader view of AI infrastructure, where a data center can be considered alongside the land beneath it, the industrial capacity surrounding it, and the energy systems required to operate it. “AI may be described through models, tokens, and processors, but its ability to operate depends on a chain of physical resources,” Sanjay states. “The investment lens has to follow that chain from resources to energy, from energy to infrastructure, and from infrastructure to compute.”

That chain helps explain the role 33 Holdings is seeking to play as its real-assets thesis expands. Its focus includes four areas: housing, industrial assets, compute, and energy. These are categories that can increasingly interact as economic infrastructure develops. Energy infrastructure can influence the viability of compute sites; industrial assets can support equipment and logistics; land can provide the foundation for new development; and housing can become relevant where large infrastructure projects bring employment and population growth. Sanjay notes that the investment framework considers each asset within a wider network of economic dependencies.

This perspective also informs the firm’s interest in the assets upstream of computing itself. Sanjay points to resources such as minerals, mining, drilling, magnets, transmission infrastructure, water, and specialized data-center construction as areas that may gain relevance as computing demand grows. For a real-assets investor, those components offer a way to examine the AI economy before capital reaches the finished data-center facility. The opportunity can emerge from understanding the sequence of requirements that allows a computing project to move from a concept on paper to an operating physical asset. 

Sanjay’s own background spans technology, engineering management, and real estate, giving him experience across several of the sectors now converging around infrastructure investment. The shift toward compute and energy appears connected to the firm’s broader development of its real-assets thesis, rather than simply to the current attention surrounding AI.

Execution seems to be another part of that role. Large infrastructure projects can require land assembly, entitlements, permitting, zoning knowledge, equipment coordination, infrastructure access, and relationships with local authorities and communities. Sanjay’s experience in development has led the firm to place particular importance on understanding those local conditions before committing capital. He also emphasizes identifying prospective users or customers early in the development process, a lesson drawn from the company’s experience with housing and industrial projects.

That local perspective extends across the markets where the firm operates. Its geographic interests include the U.S. Southeast, while its India-related focus includes Hyderabad and Visakhapatnam, regions Sanjay identifies in connection with compute, energy, and industrial activity. His professional networks across the United States and India can provide a framework for connecting international capital with local development knowledge. In an infrastructure cycle where projects depend upon site-specific conditions, those relationships may carry practical significance alongside financial analysis.

The broader investment thesis, then, is about recognizing the network of physical requirements that accompanies its expansion. “The interesting opportunities may sit between categories,” Sanjay remarks. “When housing, industrial capacity, energy, and compute begin depending on each other, the investor who understands the connections may see a different map of the economy.”

AI’s continued development may consequently broaden the definition of an AI investment. The next phase could involve the physical systems that provide electricity, land, equipment, logistics, construction capacity, and industrial support for computation.

For investors, that creates a landscape in which the infrastructure enabling intelligence may warrant as much attention as the technology itself. The emerging AI economy may be built across a far wider physical network than the data centers that make it visible.