Rising AI Investments in APAC Confront Infrastructure Challenges
Artificial intelligence expenditure continues to climb in the Asia Pacific (APAC) region, yet many enterprises struggle to realize expected returns on their AI initiatives. A significant factor behind this difficulty is the inadequacy of existing AI infrastructure, which often fails to support the speed and scale required for real-world applications to perform inference effectively.
Industry research indicates that numerous AI projects fall short of their return on investment (ROI) goals despite substantial investments in generative AI (GenAI) tools. The underlying infrastructure limitations directly impact AI performance, operational costs, and scalability across APAC markets.
Edge Computing: A Strategic Response to Inference Bottlenecks
Akamai, in collaboration with NVIDIA and leveraging the latest Blackwell GPUs, has introduced Inference Cloud to tackle these challenges. The core concept is to bring AI inference closer to the end-users, rather than relying solely on centralized data centers located far from users. This proximity reduces latency, lowers operational costs, and supports AI applications that require immediate decision-making.
Jay Jenkins, CTO of Cloud Computing at Akamai, shared insights on the critical shift enterprises must make to overcome the inference bottleneck. He emphasized that while AI training cycles receive much attention, it is the inference stage that now dominates computational resource consumption and poses the most significant deployment challenges.
Infrastructure Gaps Hamper AI Project Success
Jenkins explained that the transition from AI experimentation to full-scale production is wider than many organizations anticipate. High infrastructure costs, latency issues, and difficulties in managing models at scale frequently hinder progress. Most companies still depend on centralized cloud resources and large GPU clusters, which become prohibitively expensive and slow, especially in regions distant from major cloud hubs.
Latency issues degrade user experiences and diminish business value, particularly as AI services demand split-second responses. Additionally, complexities arising from multi-cloud environments, stringent data regulations, and compliance requirements further complicate scaling AI projects.
Inference Demands Surpass Training in APAC
With AI adoption progressing beyond pilot phases, the daily inference workload—processing real-time data through AI models—is now the primary consumer of computing power across APAC. Diverse language needs, regulatory environments, and data contexts increase the pressure on centralized systems, which were not designed for such responsiveness and geographic distribution.
Advantages of Edge AI Infrastructure
Shifting AI inference to edge locations—closer to users, devices, or autonomous agents—significantly enhances response times and reduces the volume of data transmitted to centralized clouds, decreasing operational costs. This is particularly crucial for physical AI systems such as robots and smart city technologies that require millisecond-level decision-making.
Cost savings from localized AI deployments are notable. Akamai’s analysis reveals that enterprises in countries like India and Vietnam experience substantial reductions in expenses when running image-generation models at the edge, benefiting from more efficient GPU utilization and lower data egress fees.
Industries Leading Edge AI Adoption
Industries sensitive to latency, such as retail, e-commerce, and finance, are early adopters of edge AI. In retail, faster, localized inference improves personalized recommendations and search functions, reducing shopper abandonment due to delays. In financial services, quick AI-driven fraud detection, payment approvals, and transaction scoring are essential, making edge computing vital for maintaining regulatory compliance and operational speed.
Collaborations Between Cloud Providers and GPU Manufacturers
The growing demands of AI workloads have fostered closer partnerships between cloud service providers and GPU manufacturers. Akamai’s collaboration with NVIDIA exemplifies this trend, deploying GPUs, DPUs, and AI software across thousands of edge sites. This distributed “AI delivery network” enhances performance and helps address compliance by enabling local data processing aligned with varying regional regulations.
Security measures such as zero-trust architectures, data-aware routing, and fraud protection are integral to these edge AI infrastructures, ensuring robust defenses against cyber threats and maintaining data integrity.
Preparing for Agentic AI and Distributed Operations
Agentic AI systems, which perform multiple sequential decisions rapidly, require infrastructure capable of millisecond-level operations. APAC’s diverse connectivity and regulatory landscape pose challenges, but flexible edge deployments can meet these demands. Research indicates a growing reliance on edge services, with many enterprises expecting to integrate them alongside public cloud resources by 2027.
Managing AI workloads across numerous distributed locations necessitates advanced orchestration tools and comprehensive monitoring to track performance, costs, and errors consistently.
Future Considerations for Enterprises
As AI inference shifts toward the edge, organizations must adapt to more distributed AI lifecycles involving frequent model updates across multiple sites. While this increases data governance complexity, local processing facilitates compliance with diverse regulations.
Security will require heightened focus, as each edge location becomes a potential attack surface. Protecting APIs, data pipelines, and mitigating fraud or bot attacks are critical priorities, with financial institutions already leveraging Akamai’s security controls to safeguard operations.
In summary, the move to edge AI infrastructure is a strategic imperative for APAC enterprises seeking to control escalating inference costs, reduce latency, and unlock the full potential of AI applications in real-time environments.
Photo credit: Igor Omilaev
Fonte: ver artigo original
LiteLLM Ends Partnership with Delve Following Security Breach
Cloudflare’s new crawler rules expose 3 tensions in the OpenAI-ChatGPT race
Ilya Sutskever Highlights the Need for a New Paradigm in AI Learning
What Europe’s AI Education Experiments Reveal for Business Talent Strategies
