May 04, 2026

Your teams are already using AI. The more pressing issue is whether your infrastructure can keep up.

Picture this: a sales lead asks for a forecast update during a client call. Your analyst pulls it up, but the summary takes several seconds to load. The moment passes, and the client notices.

Small delays like this can affect how consistently AI tools fit into everyday work. They rarely appear as major failures. More often, they show up as minor slowdowns that repeat across teams, meetings, and workflows.

According to McKinsey’s State of AI in 2025 research, 88% of enterprises report regular use of AI across operations, supporting contract review, forecasting, meeting summaries, sales analytics, and operational planning. AI capabilities are increasingly embedded throughout the workday, and the systems supporting them play a growing role in how consistently they perform.

For companies expanding AI usage across departments, the key question isn’t whether teams are using AI. It’s whether the underlying hardware is designed to support those workloads.

The real cost of routing everything through the cloud

Cloud AI helped accelerate enterprise adoption by providing scalable access to powerful models without requiring significant upfront infrastructure investment. For model training, large-scale analytics, and complex reasoning tasks, cloud processing continues to play an important role and remains essential to most modern AI strategies.

But when employees rely on an AI assistant for productivity throughout the day, a cloud-only architecture can introduce tradeoffs:

• Responsiveness: Tasks like summarization and transcription often require sending requests to remote servers and waiting for results to return. In environments with inconsistent connectivity, these delays become more noticeable.
• Data movement: Processing information through external infrastructure means certain data must travel across networks, which can introduce additional governance considerations.
• Cost variability: Increased AI usage across employees and workflows can increase cloud compute consumption, which may affect how organizations forecast and manage costs.

These factors are manageable individually. As AI adoption expands across departments, they become more relevant to long-term infrastructure planning.

What changes when AI runs locally

On-device AI enables certain workloads to run directly on the computer instead of requiring a round trip to external servers.

Modern business laptops and tablets designed for AI workloads include a dedicated neural processing unit, or NPU. The CPU handles general computing and the GPU handles graphics, while the NPU is optimized for specific AI tasks.

NPUs can support capabilities such as:

• Meeting transcription that processes audio locally
• Real-time captioning and audio enhancements during calls
• Document summarization within applications
• Local processing for certain analytics and reporting tasks

For teams that run on tight schedules and tighter deadlines, this type of local processing can make a noticeable difference.

Surface for Business Laptops and 2-in-1 computers incorporate this architecture by distributing supported workloads across the CPU, GPU, and NPU, alongside Windows 11 Pro and Microsoft 365 Copilot ¹ experiences. That foundation can also support tools like Copilot Agents ² for automation scenarios, like routing information or assisting with repetitive tasks.

Data control as a performance strategy

When AI tools interact with sensitive business data every day, security can’t be an afterthought. It has to be built into the system from the start.

Consider what’s at stake: a contract reviewed by an AI assistant, a financial projection summarized for a board presentation, or a customer communication drafted with AI support. Each involves data that organizations are legally and operationally responsible for protecting. Running supported workloads locally can reduce the need to transmit certain data externally, which may simplify some governance considerations.

Surface for Business computers incorporate Microsoft’s chip-to-cloud security architecture, providing layered protection across hardware, firmware, software, and identity. This architecture helps support device security and enables organizations to manage sensitive information within their existing security frameworks. Because AI workloads can run locally, features are able to continue operating even when network connectivity is limited.

A cost structure that scales with AI adoption

AI adoption rarely grows at the pace organizations originally plan for. A tool that starts with one team spreads to several. A single automated workflow quickly becomes many. Each expansion can add cloud compute costs that weren’t fully anticipated at the start.

Combining cloud-based processing with on-device AI capability can provide additional flexibility. Running workloads locally may help reduce reliance on constant cloud processing while still preserving access to advanced AI services when they’re genuinely needed.

Computers equipped with NPUs can also support additional AI features as software evolves, which helps organizations prepare for future workloads such as predictive analytics and natural language processing that will become increasingly central to how teams work.

The infrastructure decision behind AI success

Teams that get the most value from AI aren’t always the ones with the most sophisticated models. They’re the ones whose hardware is designed to support AI workloads consistently, securely, and across the environments where work happens.

Surface for Business computers combine NPU acceleration, Windows 11 Pro, and Microsoft chip-to-cloud security as part of a hardware platform designed to help organizations run AI locally, support consistent performance under sustained demand, and scale with greater confidence.

Organizations looking for faster responses, stronger data protection, and more predictable costs as AI scales can explore how Surface for Business is designed to support all three across every team and work environment.