Share this:

Shifts in data center infrastructure services have been underway for more than a decade, one specifically that was set in motion in the 2010’s with the introduction of the DPU. The global data processing unit (DPU) market size was $553.96 million in 2021 and is projected to reach $5.5 billion by 20311. What compelling need is the driving force behind this surge in DPU technology? The answer starts with the hyperscalers who are pushing the technology envelope specifically when addressing efficiency and at-scale infrastructure challenges.

What is a DPU and how does it affect hyperscalers?

A DPU, or a Data Processing Unit, sits in an application server and offloads key infrastructure services such as networking and storage functions from the server CPU, freeing up 100% of the server CPU and memory resources to run a customer’s workloads.

Hyperscalers, who have traditionally relied on SDN/SDS software, have been innovating new ways to solve these efficiency challenges, which had been consuming 20-30% of server resources. One cloud vendor estimates it loses $27K in revenue per server from SDS/SDN overhead, with two-thirds of the loss related specifically to SDS.2 The hyperscalers, however, have addressed these challenges by deploying Nitro-like devices and proprietary software stacks, effectively leading the next wave of infrastructure services.

      • In the Early 2010’s, various startups form, introducing their own version of the DPU.


      • Amazon Introduces The Nitro Card in 2013, a DPU that together with the Amazon cloud and Amazon Machine Images (AMIs) combine to create the first cloud operating platform.


      • In Oct 2021, it is reported that Google and Intel will develop a new chip for the ‘Mount Evans’ Infrastructure Processing Unit3 to help hyperscalers offload infrastructure tasks from CPUs to IPUs, freeing up server CPU cycles for revenue-generating tasks.


      • In May 2022, AMD acquires Pensando4 seeing the value of DPU-based infrastructure, which “enables us to offer cloud, enterprise and edge customers a broader portfolio of leadership compute engines that can be optimized for their specific workloads” – AMD Chair and CEO Dr. Lisa Su


    • In Dec 2022, it is reported that Microsoft is in advanced talks to purchase Data Processing Unit vendor, Fungible5, to integrate its DPU into Azure’s public cloud infrastructure architecture. It signals that Microsoft has recognized the competing requirement for the same levels of efficiency that Amazon AWS and Google have achieved when deploying this new model. This relative inefficiency would become a competitive threat to the Azure business if not addressed.


Why Fungible, why now?

At the heart of Azure’s potential lack of competitiveness is the decision AWS has made over the last few years to move away from a software-defined server-storage architecture to one where infrastructure services such as storage, security, management and networking are offloaded onto DPU-like cards. This has driven tremendous efficiency improvements. It allows AWS to maximize the level of customer application workloads running on each deployed server. It enables every server to be managed at the lowest cost. And it ensures that servers can be protected and recovered quickly should there be a security attack. These are the hyperscaler efficiencies obtained by AWS’s evolution to a DPU based architecture called Amazon Nitro, derived from an acquisition of Annapurna Labs back in 2015.

It can safely be assumed that Microsoft will try to match these hyperscaler efficiencies with internal use of the Fungible DPU within Azure. This move to DPU-based architecture represents a critical infrastructure revolution from the previous generations of 3-tier external storage and software-defined infrastructure.


This is great for the cloud, but how does it apply on-premises?

If the big 3 public cloud hyperscalers are moving to DPU-based infrastructure architectures, where does it leave everyone else? Where does it leave those who do not have the massive resources required to build their own DPU-based architecture? How about other Cloud Services Providers (CSPs)? Or Managed Services Providers (MSPs) providing infrastructure solutions into their Enterprise customer base? Or the Enterprises themselves who would love to match hyperscaler operating efficiencies for Edge deployments, or in their Core data centers?

We have already seen this shift trickle into on-premises infrastructure. It started on the networking side with companies like Pensando who have introduced DPUs which offload networking functions from the server CPU. But what about the storage side?

The reality is that we are on the cusp of an infrastructure architecture revolution and everyone else – anyone who is not a public cloud hyperscaler – needs a Cloud Operating Platform that delivers hyperscaler equivalent operational efficiencies that matches the AWS Nitro approach.


Nebulon brings AWS-like Cloud Operating Platform on-premises with its DPU-based solution

Nebulon smartInfrastructure has pioneered the Cloud Operating Platform for everyone else. A DPU-like solution that allows customers to deploy, manage and protect their server-storage infrastructure with the hyperscaler operating efficiencies obtained by the Big 3. It is optimized for enterprises that have 10s to 1000s of edge site locations. It provides solutions for Enterprises that want to deploy, manage, and protect their Core and Edge deployments. It allows CSPs and MSPs to build their own value-added solutions. They can use Nebulon smartInfrastructure to match the Big 3 public cloud provider’s IaaS approach, whether deploying in their own data centers, at Edge or Core locations within their Enterprise customer base.


If you wish to match the hyperscaler level efficiencies available to the Big 3 public cloud providers for Edge, Core or IaaS solutions we would invite you to investigate Nebulon smartInfrastructure, and visit us on



2 Azure Accelerated Networking: SmartNICs in the Public Cloud, 2018:

Share this:

Author Image

Siamak Nazari


previously 3PAR Chief Architect, then an HPE Fellow and Vice President for Hybrid IT Infrastructure.