Photo by Raghav Kalia on Pexels

Driven by escalating inference costs, enterprises across the Asia-Pacific (APAC) region are rapidly shifting their AI infrastructure towards edge computing solutions. Despite consistent growth in AI investment, many companies face challenges in translating AI projects into tangible results, often hampered by suboptimal infrastructure ill-equipped for real-time inference at scale.

Akamai is tackling this issue with its Inference Cloud, leveraging the power of NVIDIA Blackwell GPUs to bring AI decision-making closer to the end-user. This strategic shift aims to minimize latency and optimize costs. Jay Jenkins, CTO of Cloud Computing at Akamai, stresses the critical need for enterprises to re-evaluate their AI deployment strategies, emphasizing that inference has emerged as the primary bottleneck.

Jenkins points out the significant disparity between AI experimentation and widespread deployment. The limitations of centralized cloud architectures and extensive GPU clusters, particularly high costs and latency issues in geographically distant regions, are becoming increasingly apparent. Edge infrastructure offers a compelling alternative, boosting both AI performance and cost-efficiency by reducing data transit distances and facilitating quicker model responses. Retail, e-commerce, and finance sectors are leading the charge in edge inference adoption, driven by their demand for instantaneous decision-making.

Collaboration between cloud providers and GPU manufacturers is becoming increasingly vital to accommodate the growing demands of AI workloads. Furthermore, Jenkins underscores the escalating importance of security, with zero-trust architectures and data-aware routing becoming fundamental requirements. The rise of agentic AI and automation necessitates infrastructure capable of operating at millisecond speeds. Enterprises must proactively prepare for a more decentralized AI lifecycle, effectively managing models across diverse locations while ensuring robust data governance and security protocols.