As inference becomes the largest share of AI compute, proximity to users and data is turning location into a core part of infrastructure strategy.

Many inference workloads perform best close to the users, data and regulations they serve, which is changing where AI infrastructure gets built.

For the first wave of generative AI, infrastructure strategy was largely about scale. The priority was building the largest possible clusters to train frontier models, often in a small number of very large campuses located wherever power was abundant. Training still follows that logic, but the mix of AI workloads is changing, and with it the question of where AI infrastructure needs to be.

Inference Is Becoming the Majority of AI Compute

Deloitte predicts that inference, the work of running trained models to answer queries and power applications, will account for roughly two-thirds of AI compute in 2026. Inference differs from training in several ways, including latency sensitivity, data location and continuous demand, and each of them affects where capacity should sit. 

Training is a large but bounded job. A model is trained, and the cluster moves on to the next run. Inference, by contrast, is continuous: every query, recommendation and automated decision draws compute for as long as an application is in production, and that load grows with the number of users.

Deloitte also projects that the market for inference-optimized chips will exceed $50 billion in 2026. As inference becomes the larger share of demand, its requirements of consistent performance, responsiveness and proximity carry more weight in infrastructure decisions alongside raw scale.

Latency Makes Location a Performance Decision

For many inference workloads, distance is a performance variable. Every request that travels to a distant data center and back adds delay. A chatbot can absorb a small delay. For fraud detection, industrial automation, real-time recommendations or AI agents chaining many model calls together, small delays compound into a noticeably slower product.

Placing compute closer to end users reduces that round trip. Lower latency is one reason edge AI is growing, a market that spans devices, software and edge infrastructure and that Grand View Research values at about $24.9 billion in 2025, with a projected $118.7 billion by 2033 and a 21.7 percent compound annual growth rate from 2026.

Edge AI Includes Regional Data Centers as Well as Devices

The term edge is used loosely, so precision helps. Some edge AI runs directly on phones, vehicles and sensors. Deloitte expects most inference to remain in data centers, mainly large facilities, rather than shift to low-cost edge devices.

The open question is where those data centers sit. Regional, high-density facilities near population and business centers can shorten the distance between compute and demand while keeping the reliability and density of a professional data center.

Data Sovereignty Turns Location Into a Compliance Question

Location also matters for compliance. Enterprises in regulated sectors increasingly need to know where their data is processed as well as where it is stored. Gartner forecasts worldwide sovereign cloud infrastructure spending of $80 billion in 2026, up 35.6 percent from 2025, according to CIO Dive, with governments as the main buyers followed by regulated industries and critical infrastructure organizations.

When a query leaves a jurisdiction to reach a model, sensitive data can leave with it. Localized inference keeps both the data and the decisions made on it within the region that governs them, so the physical location of AI infrastructure becomes part of the compliance strategy alongside encryption, access controls and contracts.

Proximity Still Requires High-Density Infrastructure

Moving compute closer to users does not reduce the physical demands of AI hardware. Inference clusters still require high-density power, advanced cooling and reliable operations, which many regional facilities were not designed to provide.

The result is a practical challenge: locations that suit latency and sovereignty often lack AI-ready capacity, while places with abundant power are often far from users. Retrofitting existing properties is one way to bring that capacity closer to demand, and Vertical Edge’s conversion model, described in its urban data center announcement, is intended to offer faster time to revenue and lower capital intensity than ground-up construction.

Location Is Becoming Part of AI Infrastructure Strategy

For AI companies and enterprises planning their next deployments, the questions are expanding. Asking how much compute is needed is no longer enough. Teams also need to know where each workload must run. Distance from users and data, and the jurisdictions that apply, are now planning inputs. 

Vertical Edge, Vertical Data’s data center ownership platform, is developing a network of edge data centers near major North American metropolitan areas, pursuing retrofits, new builds and acquisitions to support low latency, secure localized inference and data sovereignty. As inference continues to grow, location is likely to matter alongside scale.