The agreement in brief
On July 16, 2026, Duos Edge AI announced a five-year colocation agreement with an unnamed investment-grade hyperscaler, providing 10MW of critical IT-load capacity valued at more than $111 million in contracted revenue. The capacity is at the company's Columbus, Georgia campus and is expected to be available in the fourth quarter of 2026. Duos Edge AI is a subsidiary of Duos Technologies Group, listed on Nasdaq as DUOT, and it specializes in high-density edge data centers. Doug Recker, the company's CEO, said the agreement shows how the company is investing in infrastructure in key markets so it can quickly add capacity for its customers.
The Columbus campus is scaling fast. An initial 10MW deployment is expected to begin generating revenue in August 2026, and with this new agreement the site is projected to reach 20MW of contracted capacity by the end of the fourth quarter. The company recently completed a $55 million capital raise to fund the Columbus facility acquisition and the infrastructure to support these customer deployments. Duos markets its edge model on density and speed: more than 100kW per cabinet, roughly 90-day deployment timelines, and owned infrastructure that generates long-term recurring revenue rather than one-off project fees. Those are modest numbers next to gigawatt campuses, and that is the point.
Why a hyperscaler buys 10MW in Columbus
The instinct is to see 10MW as small, because the headlines this year have been about gigawatt builds and multibillion-dollar power deals. That framing misses what is happening at the edge. Hyperscalers are not only chasing raw scale in a few mega-regions. They are also stitching together capacity in secondary markets to get compute physically closer to users and to tap power and permitting that the primary hubs can no longer supply on demand. A funded, investment-grade hyperscaler committing to a five-year contract in Columbus is a deliberate distribution play, placing AI-ready capacity where it can be delivered quickly rather than waiting years for a flagship campus.
Speed is the real currency here. A 90-day edge deployment competes on a completely different timeline than a multiyear hyperscale build that has to clear interconnection queues, environmental review, and local opposition. When AI demand is growing faster than the primary markets can absorb, capacity that can be lit up in a quarter has strategic value even at modest megawatt counts. For the hyperscaler, spreading load across many fast, smaller sites also diversifies grid and regulatory exposure. The Columbus deal is a small instance of a large pattern: the AI buildout is going wide and fast at the edge, alongside going deep and slow at the core.
The edge economics that make it work
The financial shape of Duos Edge AI's model is worth understanding because it is being replicated across the sector. The company builds and owns high-density facilities, signs multiyear colocation contracts, and books predictable recurring revenue against the fixed asset. A $111 million contract over five years against a 10MW build is a clean, financeable unit of economics, and the recent $55 million raise shows how the capital and the contracts reinforce each other. Owned infrastructure with a contracted anchor tenant is exactly the kind of asset that lenders and equity investors can underwrite, which is what lets a relatively small operator keep adding capacity.
The risk in this model is the same one that shadows the whole capacity boom: it depends on demand holding for the life of long-dated contracts, and on a single generation of high-density hardware staying relevant. High-density edge sites optimized for today's AI accelerators are purpose-built, and repurposing them if demand softens is not trivial. An investment-grade anchor tenant on a five-year term mitigates that considerably, which is why the credit quality of the counterparty matters as much as the megawatts. For operators and their backers, the durability of these builds rests on the assumption that AI inference volume keeps climbing, which so far it has.
What it means for enterprise buyers
For enterprise technology leaders, the rise of secondary-market edge capacity is a practical opportunity, especially for inference. Training runs want to be concentrated near cheap power and dense interconnect, but inference wants to be close to users to keep latency low. A network of edge sites in tier-two cities is well suited to serving production AI workloads to regional user bases without routing every request back to a distant mega-region. As this capacity comes online in places like Columbus, the set of viable locations for latency-sensitive AI expands, and with it the leverage buyers have on both performance and price.
The buy-side action is to widen the aperture in your capacity planning. When you evaluate where to run inference, do not default to a hyperscaler's flagship region simply because it is the obvious choice. Ask whether an edge provider in or near your key markets can deliver the density and latency you need, faster and possibly cheaper. The trade-offs are real: smaller operators carry more counterparty risk than a hyperscaler, and the tooling around a colocation edge site is thinner than a managed cloud region. We would weigh those against the latency and speed-to-capacity benefits deliberately, rather than assuming the mega-campus is always the answer.
The pattern behind the press release
One 10MW deal carries little weight on its own, and it fits a broader trend worth tracking. The AI infrastructure story has been dominated by the biggest numbers, the gigawatt campuses and the tens-of-billions power agreements, because those are the ones that reshape utility planning and make headlines. Underneath them, a quieter buildout is happening through smaller, faster, distributed sites that collectively add up to real capacity and get compute to more places sooner. Duos Edge AI is one of several operators executing this at the edge, and the hyperscaler appetite for their capacity confirms that the model has a paying customer at the top of the market.
For a CTO planning an AI roadmap over the next two years, the takeaway is that capacity is arriving on two clocks at once. The core is slow, huge, and constrained. The edge is fast, modular, and spreading into markets you may not have been watching. Building your strategy around only the first clock means you plan for scarcity and long lead times. Accounting for the second means you have more options, closer to your users, available sooner. The Columbus deal is a small data point, and it points at the more useful half of the buildout for most enterprise inference workloads.
