Duos Edge AI Hits 20MW Contracted Deployment in 2026 with Columbus, Georgia Colocation Expansion

Hyperscalers are signing colocation deals in secondary markets like Columbus, Georgia for four structural reasons: grid constraints and power availability in Tier 1 hubs, the latency and decentralization requirements of AI inference workloads, faster land, permitting, and construction costs outside saturated markets, and a deliberate regionalization strategy that spreads capacity risk across a wider geographical distribution of sites. None of these reasons are specific to Duos Edge AI; they explain the increasing demand that leads a hyperscaler to choose a 10 MW block in Columbus over waiting for capacity in a market it might otherwise prefer.

The clearest driver is power. Site selection for new data center capacity is increasingly decided by power availability rather than location or cost, since the average wait time for a grid connection in primary US data center markets now exceeds four years (Source: JLL, 2026). Northern Virginia alone holds 5.6 GW of data center capacity, more than triple the 1.5 GW in Dallas-Fort Worth, the second-largest US market, and cloud providers and technology companies already account for 65 percent of all data center leasing activity nationally. Vacancy across major North American markets fell to 2.3 percent in the first half of 2025, with 73 percent of the capacity currently under construction already preleased (Source: JLL, 2025). When primary hubs are this constrained on grid capacity, hyperscalers with near-term AI capacity needs increasingly look at markets where power and land are available now rather than years out, and some are turning to renewable energy sources, natural gas, and nuclear energy as longer-term answers to rising energy demands and power consumption at scale, though JLL notes widespread nuclear deployment for data centers may still be a decade away (Source: JLL, 2026).

AI workload decentralization compounds the pressure. AI represented roughly a quarter of all data center workloads in 2025, with training driving most of that demand, but JLL projects inference workloads will overtake training as the dominant AI requirement by 2027, and inference benefits far more from proximity to users and lower network latency than training does (Source: JLL, 2026). For technical teams: a 10 MW block dedicated to inference or GPU-based high performance computing does not require the same interconnection density as a hyperscale training campus, which is part of why a site like Columbus, with Tier III-equivalent, N+1 infrastructure and its own cooling systems and networking equipment, can support a meaningful AI deployment such as the Nistar MSA's planned 1,024 NVIDIA B200 GPUs without needing primary-hub-level fiber density (Source: Data Center Dynamics, 2026). AI is also pushing rack densities toward 100 kW with liquid cooling becoming a standard requirement for new builds, a power density profile close to the 100 kW+ per cabinet Duos Edge AI already advertises for its edge facilities (Source: JLL, 2026).

Land, construction pipeline, and market economics round out the picture. Phoenix, Chicago, and Atlanta together lead the US data center construction pipeline outside Northern Virginia, at 1.3 GW, 1.18 GW, and 1.11 GW respectively, evidence that developers are actively building capacity in secondary and adjacent markets rather than only in the traditional primary hubs (Source: JLL, 2025). Secondary markets that can offer available power, buildable land, and local government support are increasingly viable alternatives, not fallback options, for hyperscalers that need capacity sooner than the multiyear waits now common in saturated primary hubs, and building data centers in these markets is often the only realistic way to meet growing demand on schedule.