TL;DR
- AI infrastructure strategy is shifting from centralized training environments toward distributed inference deployments that require lower latency, regional proximity and edge connectivity.
- Power availability remains a major obstacle, but policy, permitting, and community engagement are emerging as equally important factors influencing where and how projects move forward.
- Workforce shortages and supply chain limitations involving transformers, turbines, fiber and construction labor are creating new execution risks, even as capital investment in digital infrastructure continues to accelerate.
- Connectivity and interconnection density may become as critical to the future of AI as power itself, particularly as AI-generated data volumes increase and inference workloads expand across distributed environments.
# # #
At the recent IMN Data Centers Private Equity East 2026 and Investor Forum held at the Union League Club in NYC, one conversation stood apart from the recurring themes of capital deployment, land acquisition and power procurement. Instead of simply asking how fast the industry can build, or how do we overcome barriers like power and community, the panel, moderated by Ilissa Miller, CEO of iMiller Public Relations, challenged the executive and investor attendees to consider a more fundamental question: Are we building the right infrastructure for the future of AI?
The panel titled “The Data Center Industry Future: Technology, Power, Government & The Kind of Opportunities That Will Be Created,” brought together leaders from operations, investment advisory, market intelligence and infrastructure development to discuss the rapidly changing future of digital infrastructure. The esteemed panelists included: Bob DeSantis, Co-Founder and Board Member at 365 Data Centers; Daniel Watts, CEO at US Signal; Benton Erwin, Principal at Arup; Colby Cox, Managing Director – Americas at DC Byte, and Goncalo Bernardo, Investment Partner at Palistar Capital.
The discussion quickly evolved beyond traditional conversations around hyperscale growth and into deeper questions about LLMs, AI inference, distributed infrastructure, Agentic AI, policy friction, workforce shortages, and whether the market is overbuilding for the wrong model of compute.
The Industry’s Biggest Miscalculation: Assuming Bigger is Better
Opening the panel, Miller framed the conversation around a thesis that resonated throughout the session,“the question isn’t just can we build fast enough. It’s whether we’re building the right things for the future.”
That framing became especially relevant as panelists discussed the shift from AI training infrastructure toward AI inference infrastructure, a distinction many outside the industry still misunderstand.
Today’s massive gigawatt-scale AI campuses are largely designed around training large language models (LLMs). And, the future economic opportunity may lie in inference, the real-time deployment and operational use of AI applications closer to users, enterprises and machines.
Watts explained the difference succinctly, “Inference is just where you leverage the model your computer creates… and that’s why that market is so massive.”
The implication is significant. Training can happen in concentrated hyperscale environments. Inference, however, requires low latency, geographic distribution and proximity to users and enterprise systems. That changes everything.
The Rise of Distributed Infrastructure
One of the strongest themes throughout the discussion was that the next generation of infrastructure may not be dominated solely by mega campuses, but by distributed networks of smaller, strategically placed facilities.
Cox argued that inference “brings edge to reality,” noting that the market may evolve toward “a distributed network” of smaller deployments rather than only centralized hyperscale campuses.
Similarly, Cox observed that enterprise customers are already demanding higher density deployments in edge facilities closer to population centers, “There are demands now for doubling commits during the term of a five-year contract, and they’re doing it… primarily [in] edge data centers, meaning data centers that are close to population centers, close to users and subscribers.”
This shift could reshape everything from investment models and site selection to permitting strategies and community engagement. It also reframes the future role of existing regional and edge operators.
Power is Still King and Policy May Become the Bigger Constraint
The panel confirmed what many in the industry already know: power availability remains the primary bottleneck. And, several panelists emphasized that policy, permitting and community resistance are rapidly becoming equally critical constraints.
According to Cox, “The conversations that I’m having with private equity… [are] not just ‘where can I find power,’ but ‘where can I find power under today’s policy?’” He further warned that communities are mobilizing faster against projects, often before developers establish a local narrative or educational framework.
That observation aligns closely with the growing industry need for proactive community engagement and digital infrastructure education, particularly as public scrutiny around power, water, land use and AI accelerates nationwide. And is near and dear to Miller’s practice at her firm, where she offers a Groundswell™ community engagement program. This sentiment was reinforced by Watts, who emphasized that “part of our job as an industry… [is] educating people.”
The panel also acknowledged that misconceptions continue to dominate public discourse, especially around water use, environmental impact and economic value creation.
Workforce and Supply Chain Pressures Are Intensifying
While power dominated much of the conversation, panelists repeatedly returned to workforce shortages and supply chain constraints as looming challenges.
Colby described the workforce challenge as a “massive problem,” explaining that customers are now evaluating engineering firms, contractors and development partners based not only on expertise, but on whether they actually have labor availability to execute projects.
Meanwhile, Bernardo warned that the industry is simultaneously scrambling for available transformers, turbines, fiber infrastructure and broader supply chain ecosystems. The discussion highlighted an uncomfortable reality, even if capital and demand remain abundant, physical execution capacity may become the limiting factor.
Are We Assuming the Future will be Mission-Critical, as such, are we Overbuilding for “Five Nines”?
Another provocative topic centered around whether the industry is designing infrastructure for a world that no longer needs the five nines guarantees. Historically, data center design has prioritized “five nines” reliability, near-perfect uptime standards built for centralized enterprise architectures.
But AI inference and distributed compute may require a different model.
Panelists questioned whether future infrastructure needs to be centralized or whether resiliency itself may become distributed. The conversation suggested that future architectures could prioritize flexibility, modularity and geographic distribution over singular perfection.
That shift could fundamentally alter capital allocation strategies and infrastructure design philosophies over the next decade.
Connectivity May Become the Most Underestimated Constraint
As the panel concluded, Miller asked each speaker what the industry is underestimating today that will matter most in five years.
DeSantis pointed toward connectivity,“bandwidth for network capabilities are going to be way below the needs for AI.” As AI-generated datasets become exponentially larger, the conversation around infrastructure may increasingly shift from simple power availability to network transport capability and interconnection density.
In other words: the future AI economy may depend just as much on fiber and connectivity as it does on megawatts.
Final Thought: The Industry is Entering Its Next Phase
The panel made one thing abundantly clear: the digital infrastructure industry is no longer simply expanding, it is evolving, and it must evolve to meet local demands, manage resources and commit to industry promises like sustainability commitments. .
The next wave of opportunity may not belong exclusively to the largest campuses or the biggest power procurements. It may belong to the operators, investors, communities and policymakers who understand how AI inference, distributed infrastructure, connectivity and public trust intersect.
As Miller concluded during the session, “hopefully you learned something a little bit different… Are we building for the right or wrong future?”
To learn more about IMN, visit www.imn.org.
To find out how iMiller Public Relations can help you navigate communications around digital infrastructure investments, visit www.imillerpr.com.
