APAC Enterprises Shift AI to Production, Driving Distributed Cloud Demand

Across Asia-Pacific, the conversation about artificial intelligence has shifted from whether companies should experiment with it to how they will actually run it. Enterprises are moving AI projects out of the lab and into production systems—a transition that is forcing a fundamental rethinking of how computing infrastructure needs to work.

When AI was experimental, it fit reasonably well into the cloud architecture that companies had already built. Those systems were designed for websites, applications, and centralised computing. But production AI demands something different. It needs low-latency inference—the ability to process requests and return answers in real time—and it needs GPU acceleration to handle the computational load. More importantly, it needs to happen close to where users are and where data lives, not in a distant data centre thousands of kilometres away.

Sean Li, managing director for Asia-Pacific at Akamai, observed that enterprises across the region are now redirecting AI budgets from experimental projects toward actual deployments. This shift is reshaping how companies think about their infrastructure. AI is no longer an assistant that helps humans make decisions; it is becoming embedded in applications and workflows, making decisions and taking actions on its own. Business value is moving to the layer where models, agents, data and applications connect. Companies are reorganising their operations around this reality.

The Asia-Pacific region presents a particularly complex environment for this transition. Cloud maturity varies widely across markets. Connectivity and data centre capacity are uneven. Energy resources are under pressure. And data sovereignty requirements—the legal obligation to keep certain information within national borders—add another layer of constraint. A company cannot simply route all its AI inference through a single global facility and call it solved. It has to balance performance with compliance, speed with security.

Akamai and its partners are positioning a three-tier approach: centralised infrastructure for model training and large-scale experimentation, regional cloud resources for market-level performance and data residency, and edge infrastructure for real-time inference close to users and devices. The company is working with NVIDIA to deploy thousands of Blackwell GPUs across this distributed network, creating what amounts to a global inference grid. NVIDIA's Jensen Huang framed the challenge plainly: inference has become the most compute-intensive phase of AI, and it demands real-time reasoning at planetary scale.

The traffic patterns are already changing. When companies train AI models, they move enormous amounts of data between centralised GPU clusters. When they run inference in production, they see something different: smaller, continuous requests distributed across many locations, each sensitive to latency and needing to be processed near where the data and users actually are. Digital-native businesses—particularly eCommerce companies—are adopting this approach fastest. Jay Jenkins, Akamai's chief technology officer for cloud computing, expects multi-agent systems to drive even greater demand for distributed computing capacity across multiple locations.

But moving AI closer to users and data introduces new security challenges. Autonomous agents expand the enterprise attack surface across applications, APIs, data pipelines and infrastructure itself. Identity and access management becomes critical when AI systems become more autonomous and inherit user privileges or gain permission to act across business systems. Sensitive information can enter AI services through prompts, processing pipelines and third-party tools. Prompt injection, data poisoning and model misuse are real risks. Agents can generate continuous interactions across software services and connected devices, sometimes without direct human involvement.

Akamai's response is Workforce Protector, a platform that governs interactions across AI, software-as-a-service, web and private applications. It works as a browser extension, capturing the context surrounding prompts, text entries, clipboard activity, file transfers and browser plug-ins. Policies can allow, warn, redact or block interactions based on the user, application, data type and assessed risk. A management console centralises policy creation, investigations and reporting. The platform can discover AI applications and agents used across an organisation, attribute activity to human and AI identities, and create records for investigations and compliance.

GoVeda, a patent search provider, has already moved its AI workloads to Akamai Cloud to support searches across more than 220 million patent publications. The company reported a 30 percent performance improvement and a 20 percent reduction in infrastructure costs after the migration. As more enterprises follow this path, the infrastructure layer connecting models, agents, data and applications will become the battleground for both performance and security.