Akamai maps distributed cloud shift for APAC AI workloads

Asia-Pacific enterprises are reassessing their cloud and security architectures as artificial intelligence projects move from trials into production, according to Akamai.

The transition is increasing demand for computing capacity and real-time data processing. Infrastructure must also operate closer to users, applications and enterprise data.

Sean Li, Managing Director, Asia Pacific, and Senior Vice President of Sales at Akamai, said enterprises across the region were moving AI budgets from experimental projects to production deployments.

Akamai identified four changes shaping infrastructure decisions. AI is progressing from assistance to action as it becomes embedded in applications and workflows. Companies are reorganising operations around the technology, while more business value is moving to the layer connecting models, agents, data and applications.

Cloud demands

Cloud infrastructure has generally been designed for websites, applications and centralised computing. Production AI introduces different requirements, including low-latency inference and GPU-accelerated processing.

Jay Jenkins, Chief Technology Officer, Cloud Computing at Akamai, outlined the need for distributed processing that brings inference closer to users and data. This approach can reduce the distance travelled by requests and improve response times for real-time applications.

Asia-Pacific presents a complex infrastructure environment. Markets have varying levels of cloud maturity, connectivity and data centre capacity. AI workloads are also placing additional pressure on computing and energy resources.

Data sovereignty creates another consideration. Companies must balance application performance with local requirements for data residency, security and compliance.

Processing AI workloads closer to users could improve fraud detection, personalised banking, commerce recommendations and media processing. Keeping inference near the underlying data can also reduce the movement of sensitive or regulated information.

Akamai is positioning its distributed cloud network as an infrastructure layer spanning centralised facilities, regional cloud locations and edge sites.

Centralised infrastructure can support model training and large-scale experimentation. Regional cloud resources can address market-level performance and data residency needs. Edge infrastructure can handle real-time inference closer to users, devices and applications.

Akamai's work with NVIDIA covers Blackwell GPUs on Akamai Cloud and NVIDIA AI Grid intelligent orchestration. The companies aim to place and manage AI workloads across distributed GPU infrastructure as demand changes.

"Inference has become the most compute-intensive phase of AI - demanding real-time reasoning at planetary scale. Together, NVIDIA and Akamai are moving inference closer to users everywhere, delivering faster, more scalable generative AI and unlocking the next generation of intelligent applications," said Jensen Huang, Chief Executive Officer, NVIDIA.

Traffic patterns

Jenkins said production inference was generating different traffic patterns from AI training. Instead of large transfers between centralised GPU clusters, Akamai is seeing smaller, continuous requests distributed across more locations. These workloads are sensitive to latency and need to be processed near users and data.

The largest changes in Asia-Pacific are emerging among digital-native businesses, with eCommerce companies adopting AI quickly. Jenkins expects multi-agent systems to increase demand for different types of computing capacity across multiple locations.

This shift is influencing Akamai's network and compute investments. The company plans to deploy thousands of NVIDIA Blackwell GPUs across its distributed infrastructure.

Akamai has also built Inference Cloud, which it describes as the first global implementation of NVIDIA AI Grid. The platform routes inference workloads across its edge network.

"This allows us to bring compute closer to where the data and users are located, helping customers improve latency while also making AI workloads more cost efficient," said Jay Jenkins, Chief Technology Officer, Cloud Computing, Akamai.

Agent risks

AI adoption expands the enterprise attack surface across applications, APIs, data pipelines, infrastructure and autonomous agents.

Reuben Koh, Security Technology and Strategy Director, Asia-Pacific and Japan at Akamai, identified identity and access management as a key issue as AI systems become more autonomous.

Sensitive information can enter AI services through prompts, processing pipelines and third-party tools. AI applications also rely heavily on APIs connecting models with business data and operational systems.

Other risks include prompt injection, data poisoning and model misuse. Access controls become more consequential when agents inherit user privileges or gain permission to act across business systems.

Autonomous agents can generate continuous interactions across software services, mobile applications and connected devices. Machine-to-machine exchanges may also occur without direct human involvement.

Security systems therefore need to attribute actions to individual users or AI agents. Organisations also require visibility into prompts, responses and data transfers as they occur.

Akamai Workforce Protector, formerly LayerX, is designed to govern interactions across AI, software-as-a-service, web and private applications. Akamai acquired LayerX and integrated its technology into the security portfolio.

The platform uses an extension for Chrome, Edge, Firefox and Safari, allowing companies to retain their existing browsers and network architecture.

It captures the context surrounding prompts, text entries, clipboard activity, file transfers and browser plug-ins. Policies can allow, warn, redact or block an interaction based on the user, application, data type and assessed risk.

A management console centralises policy creation, investigations and reporting. The platform's cloud intelligence component analyses applications, identities, extensions and AI activity.

Workforce Protector can discover AI applications, desktop tools and agents used across an organisation. It can also attribute activity to human and AI identities, creating records for investigations and compliance reporting.

Companies can enforce read-only sessions and watermarking on unmanaged devices used by employees or contractors. The system also assesses browser extensions based on their permissions, behaviour and risk scores.

These interaction-level controls complement network and endpoint security. Secure access service edge and security service edge platforms focus on network traffic and connectivity. Endpoint tools cover files and device processes. Browser-level monitoring adds context for prompts, clipboard activity and other fileless interactions.

Customer deployment

GoVeda has migrated AI workloads to Akamai Cloud to support searches across more than 220 million patent publications worldwide. The patent search provider uses G8 CPU compute for inference and plans to add RTX 6000 Pro GPUs.

The deployment also uses managed containers, object storage, block storage and managed databases. GoVeda reported a 30% performance improvement and a 20% reduction in infrastructure costs after the migration.

"Akamai's infrastructure just works. It's stable, reliable, and allows us to scale as our needs evolve," said Cheng Tai, Chief Executive Officer and Co-Founder, GoVeda.