Akamai Builds Distributed AI on Blackwell
Akamai is deploying thousands of Nvidia's new Blackwell GPUs to create a massive, distributed AI platform across its 4,200 edge locations. The move aims to offer low-latency AI compute as an alternative to centralized hyperscaler clouds. This is a major validation of Blackwell for production-grade, decentralized AI workloads like inference at the edge.
This build-out is a direct challenge to the centralized cloud model, targeting the 56% of organizations that cite latency as a primary barrier to scaling AI. Akamai is betting on its highly distributed network, which originated as a Content Delivery Network (CDN), to deliver faster, localized AI inference and fine-tuning closer to end-users and data sources. The move leverages Akamai's acquisition of cloud provider Linode for $900 million in 2022 and is part of a broader strategy, codenamed "Gecko," to embed full-stack cloud computing into its edge network. This initiative aims to establish cloud services as a third major pillar of Akamai's business, alongside cybersecurity and content delivery, with a goal of reaching $1 billion in cloud computing revenue by 2027. Nvidia's Blackwell platform, announced in March 2024, offers a significant performance leap over its Hopper architecture, with capabilities to run real-time generative AI on trillion-parameter models at up to 25 times less cost and energy consumption. The architecture combines two large, separately manufactured dies into a single chip produced by TSMC and is a key component in systems like the GB200 NVL72, a rack-scale system with 72 Blackwell GPUs. This partnership extends beyond just GPUs, integrating NVIDIA's BlueField-3 DPUs to enhance and secure data flows and AI workloads at the edge. The collaboration also includes a push into industrial and operational technology security, using a hardware-based Zero Trust model to protect critical infrastructure in sectors like energy and transportation. The strategy pits Akamai against hyperscalers like AWS, Microsoft Azure, and Google Cloud, which are also heavily investing in both purchasing NVIDIA GPUs and developing their own custom silicon (e.g., AWS Trainium, Google TPUs). While hyperscalers benefit from massive, centralized data centers, Akamai's distributed approach aims to win on latency and cost, claiming potential savings of up to 86% for AI inference compared to traditional clouds. The focus on inference is critical, as the AI market shifts from the initial training phase to the widespread deployment of models in production applications. Edge inference is projected to be the largest and fastest-growing segment of the AI inference market, which is expected to grow from over $100 billion in 2025 to more than $250 billion by 2030.
