AI-Driven Video Growth Forces Cloud Infrastructure Strategies to Account for Compression

Gaëlle Martin-Cocher, InterDigital

Edouard François, InterDigital

On the Frontier of Media Coding Advancement

WASHINGTON, DC, UNITED STATES, November 10, 2026 /EINPresswire.com/ -- Video traffic is entering a new phase of growth as artificial intelligence expands the number of systems that create, process, store and consume visual data. The shift could force cloud providers and enterprise technology leaders to reconsider how video compression factors into infrastructure capacity, network design, computing requirements and energy consumption.

During a recent executive vidcast interview, Gaëlle Martin-Cocher, senior director of media systems at InterDigital, and Edouard François, who leads InterDigital research into 2D video compression, discussed how these changing video workloads could reshape cloud infrastructure requirements. Their analysis focused on the growing convergence of video and AI traffic and the role more efficient compression technologies, including Versatile Video Coding, or VVC, could play in managing the resulting demands on storage, compute and network resources.

Historically, much of the economics surrounding internet video has been driven by human consumption. Streaming services, videoconferencing platforms and social media applications primarily moved video downstream toward viewers. AI and machine-to-machine applications are changing that traffic model by introducing new sources of video and increasing the amount of visual information moving in both directions across networks.

That distinction could become increasingly important as organizations deploy computer vision, robotics, digital twins, AI-generated media and multimodal artificial intelligence systems.

Martin-Cocher said the infrastructure implications extend beyond storage. Cloud environments increasingly need sufficient processing and networking capacity to handle AI and video workloads simultaneously. More efficient compression can affect each of those requirements because reductions in the amount of data being moved or retained can accumulate throughout the lifecycle of video assets.

For his part, François expects machine consumption of video to add another significant source that will contribute to the exponential surge of this category of traffic. This is because human video consumption is constrained by the amount of content people can watch, he said, while machines face fewer inherent limits, an asymmetry that could drive machine-to-machine video usage to grow much faster than human-generated and consumed traffic.

Compression Becomes a Cloud Capacity Variable

Video codecs have traditionally been evaluated primarily according to their ability to balance picture quality, bandwidth and computational requirements. Growing infrastructure constraints could broaden that calculation.

VVC can provide compression improvements of roughly 30% to 50% compared with its predecessor, High Efficiency Video Coding, or HEVC, according to François. He also placed typical VVC compression improvements at approximately 20% compared with AV1. Those gains can translate into lower storage and network requirements for cloud computing providers when applied across sufficiently large video workloads.

The significance of those gains extends beyond the initial reduction in file size. At cloud scale, the same video asset may consume infrastructure resources repeatedly as it is stored, replicated, transported, retrieved and ultimately processed or viewed. The implication is that compression efficiency can affect costs at multiple points across the life of video-associated assets.

That operational impact becomes much more significant once the entire lifecycle is considered, according to Martin-Cocher, who argued industry leaders need to account for it fully in infrastructure planning.

A video asset may be encoded once and stored for months or years. Or, copies can be distributed through content delivery infrastructure, transferred repeatedly across networks and ultimately decoded millions of times. Similar patterns with video are emerging outside conventional entertainment applications in enterprise archives, surveillance systems, AI training data and industrial digital twins.

Further complicating matters, traditional consumer video architectures have been heavily optimized for downstream distribution. AI and machine-to-machine applications can generate significantly more upstream traffic as cameras, sensors, devices and users transmit video to cloud environments for analysis or send requests that result in newly generated video.

This places pressure on cloud processing capacity and network infrastructure at both ends of the transaction. Rapidly growing real-time machine applications introduce yet another challenge because of tightly defined latency windows.

Video Quality Takes on a Machine Dimension

The relationship between compression and quality is also changing as machines become video consumers. Traditional video quality measures have largely focused on human perception. Computer vision systems can place different demands on an image because analytical models may depend on edges, textures and other fine visual details to identify objects or interpret events.

Martin-Cocher said advanced codecs can help preserve those details while reducing the amount of data that needs to be transmitted, introducing another requirement for trade-off analysis by infrastructure planners. Compression improvements can be used to reduce bitrate and file size, improve visual quality at a comparable bitrate, or establish a balance between the two objectives.

The appropriate choice, she said, will depend on the workload. A streaming platform focused on controlling distribution costs may prioritize bitrate reduction. An industrial computer vision application may place greater value on retaining image details needed for machine analysis. Robotics applications may have to balance image fidelity against latency and network capacity. Meanwhile, synthetic images and video can contain characteristics similar to computer-generated or screen content.

Infrastructure Economics Extend to Energy Consumption

The growth of AI infrastructure has intensified scrutiny of data center energy consumption. Video adds another variable because storage, processing and network transmission all consume resources. While more advanced codecs can require greater computational effort during encoding, which creates an upfront processing and energy cost, the economics change substantially when content is encoded relatively few times but stored, replicated, transmitted and decoded repeatedly.

The same lifecycle math applies to energy costs, according to Martin-Cocher. For long-lived assets, she said, the initial encoding cost can be amortized across many subsequent infrastructure operations. Streaming libraries, video-on-demand archives, surveillance footage, enterprise archives, AI data sets and digital twins are examples of workloads where this effect could become significant.

Legacy archives present a different economic problem. François noted that organizations may not need to routinely re-encode large collections of existing content simply because a newer codec becomes available. The computing requirements and operational disruption can make such projects difficult to justify. The more immediate opportunity, he suggested, is therefore likely to involve newly created video entering storage and distribution systems.

It is this distinction that makes codec selection an infrastructure decision with long-term consequences. Video encoded today may remain within enterprise or cloud environments for years, carrying its storage and bandwidth requirements with it throughout its useful life.

The issue is likely to become more consequential as AI evolves toward increasingly visual and multimodal forms, Martin-Cocher said, expanding both the volume of video moving through digital infrastructure and the number of times that video must be processed, stored and exchanged.

Conclusion

The emerging agentic economy will very likely amplify the asymmetry between human and machine consumption. If machine-generated and machine-consumed video grows accordingly, managing the size and movement of that data becomes an infrastructure economics issue.

The result is a broader role for video coding. Decisions that once centered largely on media quality and distribution will increasingly influence the cost, performance, energy consumption and resource requirements of cloud and AI infrastructure itself.

Airrion Andrews
Mindshare Capture

Legal Disclaimer:

EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.