By Saf Malik
Senior Content and Insights Manager
Executive Overview
The defining equation of digital infrastructure development has fundamentally shifted. For decades, tech giants, enterprise colocation providers, and real estate developers looking to build massive data storage and computing facilities chased regional economic sweeteners. State tax breaks, cheap acreage, and fast-tracked municipal approvals were the primary currencies of site selection, dictating where capital flowed.
Today, that playbook is obsolete. According to FiberLight Chief Executive Bill Major, power availability has decisively overtaken tax incentives and land costs as the singular deciding factor in artificial intelligence (AI) data centre site selection.
Driven by the explosive, energy-hungry computational demands of large-scale machine learning models, generative AI, and hyperscale cloud workloads, developers are no longer asking states what financial carrots they can offer. Instead, they are asking a much more fundamental question: Can you provide the megawatts, and can you do it now?
This paradigm shift is redrawing the data centre map of the United States. Traditional legacy hubs—long accustomed to capturing the lion’s share of digital infrastructure capital—are facing fierce competition from emergent regions equipped with robust legacy power grids, such as Oklahoma and Western Pennsylvania. Meanwhile, regulatory pressures are mounting in fast-growing markets like Texas, forcing a maturation of the sector where power, fiber connectivity, and compute can no longer be treated as isolated engineering problems.
Detailed Chronology & Market Shift
To understand how the industry reached this inflection point, one must look at the meteoric rise of AI computational density. Over the past three years, the deployment of advanced graphics processing units (GPUs) and tensor processing units (TPUs) has transformed data centres from passive storage warehouses into hyper-dense processing engines. Facilities that once operated on densities of 5 to 10 kilowatts per rack are now routinely designing for 40, 60, and even upwards of 100 kilowatts per rack.
This exponential jump in power consumption caught both regional utility grids and municipal planners flat-footed.
- The Historical Baseline: Prior to the generative AI boom, data centre developers operated on predictable, linear growth trajectories. Site selection teams prioritized proximity to dense population centers (to reduce latency), low-cost land, and multi-million-dollar tax abatements negotiated with state economic development boards.
- The Inflection Point (2023–2024): As commercial AI adoption surged, hyperscalers like Microsoft, Google, Meta, and Amazon Web Services (AWS) began aggressively securing every available megawatt of electrical capacity. Traditional data centre alleyways—such as Northern Virginia’s Data Center Alley—began experiencing severe grid constraints and transmission bottlenecks.
- The Present Reality (2025–2026): Developers have abandoned traditional site-selection sequencing. As Bill Major notes, economic incentives are completely irrelevant if a project cannot be energized. If a state offers a zero-percent corporate tax rate but cannot supply 300 megawatts of power within an 18-to-24-month window, the site is immediately disqualified.
This new reality has birthed an unexpected renaissance in regions boasting abundant energy resources and accessible transmission infrastructure. Oklahoma has emerged as a premier destination, successfully capturing massive infrastructure commitments from tech titans like Meta and Google. Simultaneously, Western Pennsylvania has become a magnet for heavy-duty AI investments—exemplified by Amazon’s monumental $20 billion commitment to regional AI infrastructure—primarily because the area offers immediate access to robust energy generation and transmission vectors.
Supporting Context & Metrics: The Anatomy of a Modern Site Selection
The evolution of site selection is underpinned by a complex matrix of operational requirements. While power is now the gatekeeper, developers must balance a volatile triad of resources: power, water, and fiber connectivity.
1. The Megawatt Imperative
Modern AI clusters require unprecedented scales of electrical power. A single hyperscale AI data centre campus can easily demand anywhere from 100 megawatts to over 1 gigawatt of continuous power—equivalent to the electrical consumption of a mid-sized American city. Because traditional transmission line construction and grid interconnection queues can take anywhere from three to seven years, developers are ruthlessly prioritizing brownfield sites with existing substation capacity or direct access to power generation assets.
2. The Rise of Behind-the-Meter Solutions
With traditional public utility interconnection queues clogged across nearly every major Independent System Operator (ISO) in the United States, developers are increasingly exploring behind-the-meter (BTM) generation. By partnering directly with natural gas pipelines, nuclear plants, or renewable energy microgrids, data centre operators attempt to bypass bureaucratic grid delays.
However, industry leaders urge caution regarding BTM approaches. As Major points out, off-grid generation is a pragmatic workaround rather than a silver bullet. While it accelerates deployment timelines when the public grid stalls, it shifts massive financial and operational burdens onto the developer. Operators must absorb higher capital expenditures, navigate complex fuel supply chains, assume direct emissions liabilities, and manage intricate local environmental permitting processes.
3. The Fiber-Power Convergence
A recurring pitfall in modern site selection is the siloed approach to infrastructure. Securing an isolated tract of land with abundant power generation is meaningless if the location lacks the high-capacity telecommunications infrastructure required to transport petabytes of training data in and out of the facility in real-time.
Major emphasizes that the next generation of data centre hubs must treat power, compute, and fiber connectivity as an integrated, unified ecosystem. Facilities built away from dense metropolitan cores must be anchored by robust, high-fiber-count dark fiber routes to ensure ultra-low latency connection back to primary Internet exchanges and enterprise networks.
Official Statements and Industry Insights
The transformation of the site selection ethos has brought forward-looking industry executives to the forefront of national infrastructure planning. Speaking on the shifting priorities of developers, FiberLight’s leadership underscores that the rules of engagement have permanently changed.
"The conversation starts to change from, ‘What incentives can you give me?’ to, ‘Can you actually give me the power I need, and when can I get it?’ If you cannot deliver the power within the developer’s timeframe, the economics don’t matter."
— Bill Major, CEO, FiberLight
Major’s perspective is particularly salient given FiberLight’s heavy operational footprint in Texas, a state currently at the epicenter of both the AI boom and intense regulatory scrutiny. Following Texas Governor Greg Abbott’s directive to initiate a comprehensive grid audit covering roughly 300 proposed data centre projects, industry watchers questioned whether capital would flee the Lone Star State.
Major views the Texas audit not as a roadblock, but as a necessary and measured course correction. Given historical vulnerabilities in the Electric Reliability Council of Texas (ERCOT) grid—most notably highlighted by winter storm crises—scrutinizing the impact of massive industrial power loads on local ratepayers, water tables, and transmission infrastructure is entirely warranted.
"Taking a more deliberate approach to new development doesn’t mean the underlying demand and construction is going away. For FiberLight, our projects are already approved, funded, and under construction on a five-year planning horizon. Demand keeps growing regardless of near-term regulatory adjustments."
— Bill Major
Reflecting this long-term conviction, FiberLight earlier this year committed a $500 million capital investment specifically aimed at expanding its West Texas fiber network to accommodate hyperscaler and AI workload demands. This aggressive capital deployment signals that while regulators may slow down speculative applications, committed infrastructure providers are doubling down on regions capable of scaling sustainably.
Future Outlook: Parallels, Pacing, and Policy
As the United States races to secure its geopolitical and economic dominance in the global artificial intelligence landscape, the friction between rapid technological expansion and physical infrastructure limitations will continue to test developers, utilities, and local communities.
Drawing Parallels to Cellular Infrastructure
Industry veterans often look to past infrastructure revolutions for guidance. Major draws a direct historical parallel between today’s local pushback against data centres and the public resistance encountered during the mass rollout of cellular communication towers in the late 1990s and early 2000s.
Initially plagued by zoning battles, aesthetic complaints, and unfounded health concerns, cell towers eventually became universally accepted as indispensable civic utilities. What makes the current AI buildout uniquely challenging, however, is its sheer physical scale and velocity. Technologies that historically took decades to mature have integrated into the global economy within a matter of years, compressing decades of engineering challenges into tight, multi-month deployment windows.
What Markets Will Win the Next Cycle?
The map of America’s digital economy is being redrawn in real-time. The established primary markets—Northern Virginia, Silicon Valley, Chicago, and Dallas—will undoubtedly remain vital, but their growth will be tempered by physical grid constraints and rising real estate costs.
The true winners of the upcoming infrastructure cycle will not necessarily be the regions that dominated the cloud computing era of the 2010s. Instead, market leadership will belong to jurisdictions capable of executing a synchronized strategy:
- Pragmatic Energy Planning: Regions that can streamline grid interconnections or safely integrate BTM energy generation without penalizing residential ratepayers.
- Resource Management: Forward-thinking municipalities that balance industrial water usage with local environmental stewardship.
- Seamless Connectivity: Areas fully laced with high-capacity fiber backbones that link massive remote compute clusters directly to global networks.
Ultimately, the AI revolution is forcing a brutal reckoning with the physical laws of energy and geography. In this new era, power is no longer just a utility input—it is the ultimate strategic asset, dictating the winners and losers of the next great technological epoch.
