Intelligence at the edge: How in-room compute is transforming collaboration

The architectural shift

As AI accelerates across the collaboration ecosystem, the previously plain corporate meeting room is undergoing its most significant architectural shift in more than a decade.  What began as simple AI tools—virtual backgrounds, noise suppression, meeting transcription—has now begun to migrate into the meeting room itself.  At ISE 2026, the conversation is no longer about if edge AI will matter in such environments, but how it will reshape meeting room technology and the very definition of a “smart meeting space”.

The drivers are clear: with hybrid work patterns now established as the new normal, organizations want richer experiences, tighter data governance, improved performance, and consistent quality across potentially thousands of rooms.  Cloud-based collaboration platforms including Zoom Workplace, Cisco Webex, Microsoft Teams Rooms, and Google Meet continue to evolve, but the next wave of innovation is happening inside the hardware centre stage at the front of the room.

This is the era of in-room compute—specialized processors for AI workloads integrated into collaboration bars, cameras, microphones, and the AVoIP endpoints themselves.

Why is AI migrating from the cloud to the room?

Over the past couple of years, AI features have penetrated conferencing platforms.  Examples include ambient noise reduction, beam-forming microphones for voice isolation, AI for intelligent framing and auto-tracking, but now also gesture-aware meeting controls, live transcription, translation and speaker attribution, plus highly accurate contextual summaries after the meeting concludes.

Initially, the cloud was the only place with sufficient compute to deliver these experiences at scale.  But in 2026, several trends are pushing capabilities closer to the edge:

• Lower latency and real-time responsiveness.  Camera tracking, voice direction detection, and spatial audio rendering function best when processed instantly.  In-room AI eliminates the network hop to the cloud, reducing delays that sometimes degrade user experience.

• Data privacy and regulatory pressures.  Many global organizations—especially in finance, healthcare, and government—prefer that sensitive A/V data never leaves the room.  Edge AI enables advanced features to be applied locally, or at corporate network level, without routing raw data offsite, further avoiding the requirement to encrypt and decrypt streams.

• Cost and bandwidth efficiency.  AI processing in the cloud scales with usage and increases overall service provisioning costs.  At large enterprise volumes, it is far more efficient to perform video enhancement, noise suppression, and encoding on a dedicated room appliance.

• Heterogeneous platform support.  Rooms now support multiple UC platforms, guest users, and BYOD (bring-your-own-device) scenarios.  Edge-based intelligence can help ensure consistent capabilities regardless of which collaboration platform is active.

The rise of the Intelligent Collaboration Bar

The collaboration bar has steadily become the workhorse of corporate meeting spaces.  Equipped with several microphones, speakers, a large camera, and now dedicated AI processors (NPUs – neural processing units), today’s collaboration bars are expected to perform tasks previously done only in cloud data centres. These include:

• Spatial audio processing: microphone beamforming that adapts dynamically to the speaker’s position, voice separation from ambient noise, specialist detection for hybrid-room acoustics, even screening out external sounds from beyond the meeting room boundary.

• Advanced video analytics: person detection and participant counting, auto-framing, eye-contact correction, also now multi-camera orchestration and stitching, all executed locally on the device.

• Some new bars support subtle gesture recognition—raising a hand to request to speak or using simple motions to control meetings and collaboration tools without appealing to a physical control surface.

• Local transcription and summarization.  While cloud platforms still offer more expansive language models, edge devices increasingly provide sufficient compute for baseline transcription and notetaking with no external data dependency.

For remote participants, a similar set of AI tools will run locally on PCs, and especially so on those that contain an on-board NPU to help offload AI tasks.  Overall, this distributed intelligence creates a more consistent user experience across UC platforms, even those joining remotely.

How much AI compute is sufficient?

Collaboration technology installed in meeting rooms must have a lifespan of several years in order to be cost-effective, yet UC platforms are continuously innovating and improving.  So, equipment manufacturers are now building products containing larger NPUs, anticipating a steady migration to edge compute as AI features in those UC platforms evolve and are progressively optimised.  This means the compute requirements must be overspecified to cope with an AI future not yet fully defined; and that adds cost to the hardware.

Concurrently, the other consideration is that there are several products serving each room.  Many of these will individually include NPUs, multiplying the AI compute capacity overall within the room.

TOPS (trillions of operations per second) is one of the key metrics for NPU performance; this provides a useful, although somewhat inexact, measure of AI compute within the hardware.  With multiple PTZ cameras, a smart whiteboard, AI PC, plus a collaboration bar, it’s not inconceivable that meeting rooms will have several hundred TOPS of AI compute capability, but again with expectation of this distributed capacity being underutilised.

Microsoft have specified that 40 TOPS of NPU performance is now the baseline requirement for an AI Copilot PC.  Yet the same hasn’t been declared for collaboration technology.  Indeed, it’s unclear whether the UC platform vendors will coalesce around a minimum TOPS figure for each component, or for in-room compute overall.  But this would certainly assist those vendors serving the sector in defining appropriate yet future-proofed product specifications.

As part of this, Futuresource are expanding research to examine the AI performance of each hardware component within the meeting room.  As always, it's a complex ecosystem with many moving parts; nevertheless, the results aim to decipher what level of AI compute is considered necessary for each type of meeting space, and how this might be apportioned to service typical requirements.

The meeting room itself becomes the platform

AI at the edge is no longer a future consideration it’s an architectural shift.  Collaboration bars and meeting room technologies are becoming more intelligent, with context-aware devices that work in tandem with cloud-based UC platforms.  This distributed model delivers, higher performance, increased privacy and a reduced cloud dependency which in turn provides improved user experience and even greater cross-platform continuity.

As hardware vendors introduce more powerful NPUs and as UC platforms refine their hybrid AI frameworks, the meeting room becomes more autonomous, more adaptive, and more aligned with the real-time needs of hybrid work.  This year marks a turning point: the future of intelligent collaboration will be built not just in the cloud, but directly in the room itself.

For more information and reports visit here. 

About Futuresource

Futuresource Consulting provides the insights that power the world’s leading technology and media companies. For more than 30 years the firm has combined rigorous data, sector expertise and a forward-looking view of market change. Its syndicated research, consulting services and industry partnerships span consumer electronics, entertainment, Pro AV, education and emerging technologies.

www.futuresource-consulting.com

Press Contact: Nicola Finn, Marketing Manager, Futuresource Consulting, nicola.finn@futuresource-hq.com