A shift is happening in how companies are building AI infrastructure, and it's moving away from massive centralized data centers toward smaller, distributed facilities positioned closer to where AI actually gets used. Duos Technologies, a Jacksonville-based infrastructure provider, just reported record 2025 results that illustrate this emerging trend: the company achieved 270% year-over-year revenue growth by deploying modular edge data centers and launching a GPU-as-a-Service (GPUaaS) offering that's already generating significant revenue .

This represents a fundamentally different approach to solving AI's power and infrastructure challenges. Rather than betting everything on mega-scale data centers powered by nuclear reactors or other centralized energy sources, companies are now experimenting with distributed networks of smaller, more flexible compute facilities. The implications are significant for how AI services get delivered, where they get deployed, and who can afford to participate in the AI economy.

Why Are Companies Moving Away From Mega Data Centers?

The traditional data center model concentrates enormous amounts of computing power in single locations. This approach works well for training massive AI models, but it creates bottlenecks for AI inference, the process where trained models actually answer questions or perform tasks for end users. Inference happens constantly, everywhere, and moving that data back and forth to distant data centers creates latency, increases costs, and wastes energy on data transmission.

Duos Technologies deployed 15 edge data center pods in 2025, with additional units in production designed to support AI inference, training, and high-performance computing workloads . The company also expanded into high-density configurations, including a 4.8-megawatt facility designed to support hyperscaler and AI-driven workloads. These aren't massive installations; they're modular, deployable in underserved markets, and designed to be closer to where demand actually exists.

What Does GPU-as-a-Service Actually Mean for Customers?

Duos Technologies launched GPUaaS offerings with a contract to deploy 2,304 NVIDIA GPUs (graphics processing units, the specialized chips that power AI model training and inference) across its edge data center platform . This is significant because it means customers don't need to buy expensive GPU hardware outright. Instead, they can rent computing capacity on an as-needed basis, similar to how cloud computing works for storage and servers.

The GPUaaS model addresses a real pain point in the AI industry: GPU hardware is expensive, scarce, and requires significant infrastructure investment. By offering GPU capacity as a service, companies can lower the barrier to entry for smaller organizations that want to run AI workloads but can't justify the capital expenditure of purchasing and maintaining their own hardware. The contract is expected to generate significant recurring revenue over multiple years with strong margin contribution, according to the company's financial disclosures .

How to Evaluate Edge Data Center Solutions for Your AI Needs

• Latency Requirements: Determine whether your AI workload needs responses in milliseconds (edge deployment) or can tolerate delays of seconds (centralized data center). Real-time applications like autonomous vehicles or industrial automation require edge infrastructure.
• Cost Structure: Compare upfront hardware costs against monthly GPU rental fees. Edge GPUaaS models eliminate capital expenditure but involve ongoing service fees; calculate your break-even point based on expected usage.
• Geographic Distribution: Assess whether your customers or operations are concentrated in one region or spread across multiple areas. Distributed edge networks serve geographically dispersed demand more efficiently than centralized facilities.
• Power and Cooling: Edge data centers typically require less total power than mega-facilities but still need reliable electricity and cooling. Verify that potential deployment locations have adequate infrastructure.
• Vendor Lock-in Risk: Evaluate whether GPUaaS providers support multiple GPU platforms (NVIDIA, AMD, etc.) or lock you into a single vendor's ecosystem.

The financial results tell a compelling story about market demand. Duos Technologies generated approximately $27 million in full-year 2025 revenue, up from roughly $10 million in 2024 . More importantly, the company's edge data center and GPUaaS businesses are driving this growth. The company also established Duos Technologies Solutions, a new division focused on sourcing, logistics, and fulfillment services for data center equipment, which generated approximately $10 million in new backlog within its first quarter of operations .

This momentum reflects broader industry recognition that AI infrastructure is becoming decentralized. While OpenAI's recent $122 billion funding round emphasizes the importance of compute as a strategic advantage, the company is also diversifying its infrastructure strategy beyond a single provider . OpenAI now partners with multiple cloud providers including Microsoft, Oracle, AWS, CoreWeave, and Google Cloud, as well as multiple chip platforms including NVIDIA, AMD, AWS Trainium, and Cerebras . This multi-vendor approach mirrors the industry-wide shift toward distributed, flexible infrastructure rather than monolithic mega-facilities.

The edge data center model also addresses energy efficiency concerns that have dominated recent AI infrastructure discussions. By processing data closer to where it's generated, edge facilities reduce the energy wasted on long-distance data transmission. A 4.8-megawatt facility serving regional demand consumes far less total energy than routing all inference requests to a distant mega-data center and back.

For companies evaluating their AI infrastructure strategy, the emergence of viable edge data center and GPUaaS options represents a genuine alternative to the traditional centralized model. The technology is no longer theoretical; companies are deploying it at scale, generating revenue from it, and attracting significant capital investment. Whether edge infrastructure becomes the dominant model or coexists with centralized mega-facilities, the diversity of options available to AI developers and enterprises is expanding rapidly.