Akamai deploys NVIDIA Blackwell GPUs for APAC distributed cloud AI

This expansion signals a shift from centralized AI training to distributed inference, where low latency is critical for real-time media processing and commerce recommendations. By placing high-performance compute at the edge, Akamai addresses the specific data sovereignty and connectivity challenges unique to the fragmented APAC market. For the streaming ecosystem, this infrastructure supports more sophisticated personalization and fraud detection without the latency penalties of backhauling data to central hubs. Watch for how quickly digital-native eCommerce and media firms adopt these multi-agent systems to automate complex user interactions.

Akamai's push into distributed AI inference places it in direct competition with hyperscalers that have been building their own edge and inference capabilities. In May 2025, Akamai announced a strategic partnership with NVIDIA to deploy GPU-accelerated inference across its globally distributed platform, targeting enterprise customers that need sub-100-millisecond response times for production AI workloads. The company has positioned its Inference Cloud as a complement to centralized training clusters, arguing that inference workloads benefit more from proximity to end users than from the massive parallelism of hyperscale data centers. This positioning aligns with broader industry movement toward edge inference, where latency-sensitive applications like real-time video personalization and conversational commerce require compute closer to the user.

The business case for distributed inference is being reinforced by enterprise demand for data sovereignty and regulatory compliance across APAC markets. In June 2025, Akamai reported that its cloud infrastructure services revenue grew 35% year over year in Q1 2025, driven in part by enterprises migrating AI workloads away from centralized hyperscale providers to meet local data residency requirements. The company's acquisition of Linode in 2022 provided the foundational compute layer, and the addition of NVIDIA Blackwell GPUs transforms that platform into an inference-optimized edge network. Competitors are responding: Telco edge computing strategy is shifting to meet these local AI inference requirements, while Microsoft Azure has been extending Azure AI inference capabilities to its edge zones in partnership with NVIDIA.

Technical benchmarks for edge inference deployments are still emerging, but early data suggests meaningful latency gains for streaming and media workloads. NVIDIA's Blackwell architecture delivers up to 30x inference performance improvement over the prior Hopper generation for large language model workloads, which translates directly into faster token generation for real-time applications like content recommendation and automated video tagging. For streaming platforms, the combination of Blackwell-class compute at edge locations means personalization models can run inference in under 50 milliseconds, a threshold that Akamai has cited as critical for maintaining viewer engagement during live events. The GoVeda partnership mentioned in the APAC expansion represents an early proof point, demonstrating that regional content platforms can deploy systems without relying on centralized cloud regions that may be thousands of kilometers from their end users.