Akamai Deploys NVIDIA Blackwell GPUs at the Edge

Akamai has acquired thousands of NVIDIA Blackwell GPUs as part of a strategy to expand AI inference capacity across its global distributed cloud network.

The company plans to deploy the GPUs throughout its infrastructure of more than 4,000 edge locations, bringing high-performance AI compute closer to where data is generated and applications are used.

Rather than concentrating AI workloads in a small number of hyperscale facilities, Akamai’s approach embeds inference capacity throughout its distributed platform. The goal is to reduce latency for applications that require real-time processing while lowering the network costs associated with moving large volumes of data to centralised data centres.

The move reflects growing demand for infrastructure that can support AI models once they move from development to production environments.

Distributed infrastructure for AI inference

AI infrastructure has largely focused on centralised clusters used for model training. However, organisations are increasingly prioritising inference – the stage where trained models generate results in real-world applications.

Akamai says deploying GPUs across its distributed infrastructure will allow enterprises to run inference workloads closer to end users and connected devices. By doing so, the company expects to reduce latency by as much as 2.5 times while also lowering bandwidth costs.

Adam Karon, COO and General Manager of Cloud Technology Group at Akamai Technologies, says: “While hyperscalers continue to push the boundaries of AI training, Akamai is focused on meeting the unique demands of the inference era.

“Centralised AI factories remain essential for building models, but bringing those models to life at scale requires a decentralised nervous system.

“By distributing inference-optimised compute across our global fabric, Akamai isn’t just adding capacity. We’re providing the scale, at minimal latency, that is required to move AI from the laboratory to the street corner and the hospital bed – where the work happens, where the data lives and where the ROI is realised.”

According to research cited by the company, more than half of organisations identify latency as the primary barrier to scaling AI applications. Placing GPU infrastructure closer to the point of use addresses this constraint for workloads that require immediate responses.

Supporting latency-sensitive workloads

Akamai’s edge deployment strategy is designed to support applications where even small delays can affect performance or outcomes.

Examples include autonomous systems, digital health platforms, industrial automation and financial fraud detection. In these environments, infrastructure proximity can determine whether data can be processed quickly enough to trigger real-time decisions.

The rollout of Blackwell GPUs will enable enterprises to run inference workloads directly within Akamai’s distributed cloud platform rather than relying exclusively on hyperscale data centre environments.

The architecture dynamically routes workloads to the most suitable GPU clusters within the network, enabling organisations to fine tune models and deploy large language models closer to regional users.

This approach can also help address data sovereignty requirements by keeping data processing within specific geographic regions while maintaining performance.

Rethinking AI infrastructure beyond hyperscale

Akamai’s investment highlights a shift in how organisations are approaching AI infrastructure design. While hyperscale providers continue building large training clusters, enterprises are also looking for cost effective ways to run AI models in production environments.

By distributing Blackwell GPUs across its network, Akamai aims to offer an alternative architecture where AI workloads can be executed nearer to the source of data.

The company says businesses may be able to reduce AI inference costs by as much as 86% compared with traditional hyperscale infrastructure. The Blackwell architecture itself is designed to support high throughput workloads while improving energy efficiency, making it suitable for deployment across geographically dispersed sites.

For developers, the platform provides access to GPU capacity without needing to manage infrastructure across multiple regions. Enterprises, meanwhile, gain predictable performance when delivering AI-driven services to large numbers of users.

As organisations integrate AI into operational systems across healthcare, manufacturing, retail and financial services, infrastructure requirements are shifting towards low latency compute and distributed capacity.

Akamai’s deployment of Blackwell GPUs reflects that transition as data centre infrastructure increasingly extends beyond traditional facilities into edge environments.