AI is no longer simply another fast-growing demand segment for the data center industry. It has become the “organizing principle” around which development strategy, infrastructure design and market selection are increasingly structured.
On this episode of the Data Center Frontier Show, DCF Editor in Chief Matt Vincent speaks with Colby Cox, Managing Director for the Americas at DC Byte, about the forces determining where the next generation of AI infrastructure can actually get built.
According to Cox, the market is no longer constrained primarily by demand or access to capital. The decisive constraint is executable power: whether capacity has truly been secured, when it can be energized, and whether the grid or an onsite generation strategy can support the intended phases of development.
That gap between announced and deployable capacity is becoming increasingly visible in DC Byte’s data. Cox says the firm has recorded a roughly 20% increase in projects remaining in the committed or early-stage categories, with power availability responsible for much of the delay.
At the same time, the scale of AI development continues to expand. At the beginning of 2023, DC Byte tracked three committed projects of at least 900 MW. Today, it tracks 17, along with 49 additional projects of that size in early-stage development. Fifteen of those early-stage projects exceed 2 GW, with several approaching 10 GW.
Inside the data hall, rack-density assumptions are changing just as quickly. Designs are moving from traditional averages near 6 kW per rack toward serious planning around 100-kW racks, with some AI environments pushing beyond 300 kW.
That shift affects nearly every layer of the facility, including electrical distribution, busway, breaker coordination, floor loading, mechanical plant design, piping, commissioning and heat rejection. Liquid cooling is consequently becoming a primary design system rather than a future retrofit.
Cox also examines the emerging geography of AI infrastructure. Development is spreading beyond established markets toward locations where power, policy and community support can be aligned. Texas activity is extending beyond Dallas-Fort Worth into markets including Pecos and Abilene, while Louisiana has moved from roughly 9 MW of data center capacity several years ago to a pipeline measured in gigawatts. Indiana, West Virginia, Pennsylvania and outer portions of the Atlanta market are also attracting attention.
But power is not the only variable redrawing the map. Local resistance, permitting risk and community trust are now material elements of project underwriting. As Cox puts it, the AI campus map is being redrawn “by power first and politics second.”
The conversation concludes with a look at behind-the-meter generation. Although most data center operators would prefer not to become power companies, onsite systems—particularly those built around natural gas—are becoming necessary in some markets as either a bridge to utility service or a longer-term solution.
Listen to the full episode for a data-driven examination of power availability, gigawatt-scale development, rack density, emerging markets and the new execution realities shaping the AI data center sector.
