Extended Reality in Remote Work: A Practical Shift

Extended reality earns its place in remote work only for spatial, interactive tasks: the evidence on fit, comfort, adoption, network and GPU needs.

Extended Reality in Remote Work: A Practical Shift
Written by TechnoLynx Published on 15 Apr 2025

Extended reality suits remote work best when a meeting needs interaction, coordination or a shared object people point at and discuss. In one controlled study, embodied VR meetings produced physiological synchrony comparable to face-to-face collaboration and clearly stronger than video conferencing. The same study’s authors conclude that video may be enough for less interactive, information-sharing tasks. Against that sit real costs: headset comfort, thin native software, colleagues who join without a headset, organisations that never get past the pilot, and the network and GPU capacity needed to keep latency down. We would pilot only where the task fits and that capacity is realistically available.

What extended reality adds that a video call cannot

ETSI defines extended reality (XR) as the umbrella term for virtual reality, augmented reality and mixed reality, along with immersive technologies not yet developed. VR places the user entirely inside an interactive virtual environment. AR adds spatially registered digital content to the real world in real time without blocking it out. Choosing between them for a given job is a separate decision, covered in our guide to choosing between AR, VR and XR for a use case. For remote work, the useful distinction is simpler: do colleagues share a space as bodies, or as rectangles on a screen?

Video calls fail in specific ways. A 2025 arXiv survey of video conferencing’s shortcomings uses Riedl’s (2021) definition of Zoom fatigue: physical and cognitive exhaustion from heavy use of video conferencing tools. The survey traces much of that fatigue to missing or distorted social cues. Eye contact is poor, a gallery of faces feels like being watched, faces appear larger than they would at a comfortable distance, and there are no side channels for recovering from misunderstandings.

Three of the failures the survey examines are these:

  • Gaze. The camera and the display are rarely in the same place, so you cannot tell who is looking at whom, and one camera cannot line up with many participants at once. In an early study the survey cites, a gaze-awareness display let users finish a task with 55% fewer turns than an audio-only system.
  • Shared objects. Talk about a diagram or a model is full of references like “that one” and “over there”, and the survey argues that XR’s ability to place virtual objects in a shared view equips it unusually well to support them.
  • Delay. Studies the survey reviews treat 500 to 650 ms as significant delay. At that level conversation suffers from overlaps, gaps and missed turn-taking. In immersive VR, Becher et al. (2020) found that raising added delay from 300 ms to 450 ms clearly reduced mutual understanding and task performance.

The survey’s authors note evidence that non-verbal cues can compensate for delay, and they expect XR to be particularly good at conveying location and gesture. The survey also finds that most of the XR remedies it reviews do not use head-mounted displays at all. Google’s Project Starline, for example, uses 3D capture, spatial audio and a stereoscopic display, and participant surveys found it notably superior to standard video for presence and engagement. For a remote team, “XR” need not mean a headset on every head.

Does a VR meeting really come close to meeting in person?

Partly. In a study published in Scientific Reports, 69 students in 23 groups of three did creativity tasks in one of three conditions: face to face, over Zoom, or in the Spatial VR application on Oculus Quest 1 headsets. Heart rate variability synchrony, a marker of social cohesion, was strong in VR and face to face and significantly weaker on video. Spatial presence in VR was comparable to face to face and significantly higher than on video.

VR did not close every gap. Creative group performance and social presence were highest face to face and significantly lower in both the VR and video conditions. The avatars had hand tracking but no face or eye tracking. Better tracking might narrow the remaining gap, but the study does not test that. The fair reading is that VR recovers the bodily, spatial part of being together, while in-person contact still wins on output.

A field study points the same way. In a study of remote teams meeting in VR (International Journal of Human-Computer Studies, 2023), VR meetings produced higher social presence but no better group development. The authors found VR collaborative spaces well suited to social events, brainstorming and seminars. Practicality was the biggest concern about switching, and people who joined by video without an avatar could feel excluded.

Which remote work tasks justify extended reality

Extended reality is the better fit over video for tasks that need interaction, coordination or a shared spatial object. For tasks where people mainly pass on information, the Scientific Reports study’s authors say video may suffice. The table sorts common remote work tasks using the evidence cited in each row.

Remote work task Better fit Why
Brainstorming and creative ideation in small groups XR (embodied VR) VR matched face to face and beat video on physiological synchrony, though not on measured creative output; the 2023 field study of VR team meetings found VR suited to brainstorming
Team socials and seminars XR, where everyone has a headset Higher social presence in VR meetings; video joiners may feel excluded
Design review around a 3D model or physical layout XR or AR Pointing and “that one” references depend on a shared object, which XR can overlay
Training for hazardous or costly-to-rehearse tasks XR Industrial adopters name dangerous situations as the classic VR/AR training case
Standardised training across sites XR, if reused at scale Digital training lets everyone learn the same way on their own schedule; cost advantage depends on reuse (see costs below)
Status updates, briefings, one-to-one check-ins Video Video may suffice for less interactive, information-sharing tasks
Mixed meetings where some attendees join without a headset Video Mixed-device meetings feel asymmetric and non-avatar joiners can feel excluded

Design review rests on that shared-object argument, and our article on augmented reality and 3D modelling for design covers the modelling side. Industrial adopters call training the universal anchor use case, and our article on virtual reality in manufacturing processes gives examples.

Solo knowledge work is a mixed case. In a long-term interview study of professionals who had used Apple Vision Pro regularly for at least six months, all but two said it made them more productive, mainly because of the extra display space and fewer distractions. One said it sped up nothing, and two others saw no use case for it. That evidence concerns individual work. It does little for the team problem video calls create.

Where XR breaks down for remote teams: comfort, software, cost and adoption

Comfort and cybersickness

Comfort limits long sessions. In one comparative study, 16 participants worked in VR for five days, eight hours a day, and also in a physical desktop setup. VR scored significantly worse on most measures, with concerning simulator sickness, below-average usability, and two participants dropping out on the first day because of migraine, nausea and anxiety. Vision Pro users who had grown used to their devices still criticised weight and comfort, and the most comfortable head strap turned out to be highly personal.

Guidance on session length still needs refining. A meta-analysis of exposure duration and cybersickness in head-mounted displays calls for prospective studies with standardised protocols and time-resolved symptom measurement before exposure guidance can be refined, even as the industry pushes towards all-day wear. UC Today, citing a 2024 industrial ergonomics study, puts the share of VR users experiencing some cybersickness at 50 to 80%. That is a secondary report, so treat the range as indicative rather than settled. Until better guidance exists, we would schedule XR for particular meetings rather than for the working day, and make opting out free of stigma.

Native software and existing workflows

Software remains a persistent limit. Long-term Vision Pro users named the lack of native applications and the difficulty of fitting XR into existing workflows as persistent limits, and most of them mainly mirrored their Mac screen instead of using native apps for long periods. Every participant used a physical keyboard. On collaboration, one participant said many shared experiences were limited because few people own the device, and another said shared sessions through SharePlay worked only in certain apps.

The colleague without a headset

Mixed-device meetings will be commonplace. The 2025 video conferencing survey expects heterogeneous systems, with participants on different technologies in the same meeting, to be commonplace. Vision Pro users described persona-based calls with people on plain video as asymmetric, and one participant would join video calls only if every attendee was wearing a headset. Horizon Workrooms allowed up to 50 participants, at most 16 of them in Quest headsets and the rest on standard video. A pilot needs a plan for these people from the first day.

Total cost and the scale problem

Device prices are only part of the cost. UC Today reports a PwC VR training study finding that VR training was 52% cheaper than classroom delivery at around 3,000 learners, with very different economics below that. The same piece notes that licences sized for a pilot often change once devices are shared, users rotate or more sites come online, and that programmes limited to a single department rarely reach the reuse at which cost per use falls. It also cites Stanford research in which under five minutes of VR motion data identified 95% of users in a pool of 511, which turns governance of spatial data into real work.

Platform risk

On 15 January 2026, Meta announced it would stop selling Horizon managed services and commercial Meta Quest SKUs from 20 February 2026, with support for existing customers until 4 January 2030. A systematic review of generative AI in XR also flags a lack of unified standards and interfaces between XR systems.

The Pilot Trap

Organisational readiness, more than the technology, decides whether XR lasts. A Technical University of Munich study based on 17 expert interviews between May and October 2025 describes a “Pilot Trap”: organisations run successful isolated XR pilots but struggle to turn them into sustained operations. Its authors found the binding constraints were organisational and financial, such as change management, aligned KPIs and political resistance, not technical maturity.

Integrators in that study said adoption needs three things together: operational champions, IT departments able to run the infrastructure, and executive sponsorship. Most organisations manage only one or two. Adopters advised involving IT, cybersecurity, data protection and works councils from the start, and defining KPIs with the department that owns the work before the pilot runs.

How much GPU and network capacity a shared extended reality workspace needs

A shared XR workspace streams video both ways and has to meet tight latency targets. That puts XR’s network needs between mobile broadband and ultra-reliable low-latency service classes. The numbers set the architecture before any headset is bought.

Network

In simulation, cellular capacity fell sharply when the per-user rate rose from 30 to 45 Mbps. In a dense urban simulation of 21 macro cells, each with 64 transmit antennas and 100 MHz of bandwidth, the 95th-percentile capacity was 4.2 VR/AR users per cell at 30 Mbps per user and 1.6 users at 45 Mbps. A 50% rise in the per-user rate cut supported users by about 62% (from 4.2 to 1.6).

Local networks need specific settings for NVIDIA’s Omniverse Spatial. NVIDIA’s Omniverse Spatial prerequisites recommend latency under 20 ms and require under 40 ms, with WiFi 6 (802.11ax) at 5 GHz or 6 GHz, the server wired to the router, and server and client on the same subnet for local development. ETSI says AR needs latency in the order of several milliseconds and that edge computing is necessary to deliver it. Plan for edge rendering, regional servers or on-device rendering. With several users, synchronisation adds another load: the generative AI and XR review notes that even small delays or inconsistencies between users can disrupt social presence.

Server GPUs

NVIDIA’s XR streaming runtime needs graphics GPUs with hardware video encoders, and compute-only GPUs do not qualify. NVIDIA’s CloudXR Runtime, an OpenXR-compliant runtime for streaming XR to remote clients, requires a Blackwell (recommended) or Ada generation GPU with graphics driver and NVENC support, a minimum of two NVENC sessions and three or more recommended. Supported models include GeForce RTX 5090, 5080 and 4090, the RTX 5000/6000 and RTX PRO 5000/6000 Blackwell series, and L40 and L40S server GPUs. Compute-only GPUs without graphics capabilities, such as A100 and H100, do not meet that requirement. For an Omniverse Spatial server, NVIDIA specifies one RTX 6000 Ada or an equivalent with three hardware encoders and AV1 support.

NVIDIA’s virtual workstation sizing guide says the RTX PRO 6000 Blackwell Server Edition supports up to 48 concurrent vGPUs per GPU. The same guide assigns 16 to 32 GB profiles to complex CAD, high-end 3D and high-quality rendering, notes that Blackwell may need more framebuffer than Ada or Ampere for the same workload, and recommends a proof of concept on the target GPU, OS, application, display setup and vGPU release. That guide covers virtual workstations rather than stereo XR streaming, so we would measure real per-user headroom with the encoders in use.

Rendering can also be split across GPUs. NVIDIA’s multi-GPU XR sample estimates about 35 MiB of colour and depth data per frame for a Meta Quest 3 (2064 x 2208 per eye), which takes about 1.4 ms to transfer over PCIe 4.0 and 0.7 ms over PCIe 5.0. For a Varjo XR-4 the figures are about 110 MiB, 4.4 ms and 2.2 ms. Running perception models on the headset itself is covered in our article on perception pipelines on an XR headset.

Capacity checklist before a pilot

  1. Can every pilot user reach the rendering server within the 20 ms recommended (40 ms required) latency budget from where they actually work?
  2. Do home and office networks offer WiFi 6 at 5 GHz or 6 GHz, with the server wired?
  3. Are the server GPUs graphics-capable, Ada or Blackwell generation, with enough NVENC sessions for the planned number of concurrent streams?
  4. Has a proof of concept on the target GPU and software stack set the framebuffer per user, instead of borrowing a profile from an older generation?
  5. Is there a plan for participants joining on plain video, and does the platform support them?

Should we pilot XR for remote work or stay with video?

Pilot extended reality only where the expected outcome is plausible and measurable, and stay with video everywhere else. TechnoLynx’s position for generative AI engagements is that the first step is checking whether the expected outcome is plausible at current capability, and that repeatable, menial tasks are far more likely to have measurable ROI. Applied to XR, that means three tests, all of which must pass:

  • Task fit. The task is interactive, creative, spatial or a hazardous-to-rehearse training task, not information sharing.
  • Plausible scale. The content or workspace will be reused across enough people or teams for cost per use to fall, rather than staying inside one department.
  • Owners in place. An operational champion, IT co-design and executive sponsorship are lined up, with KPIs agreed before the pilot starts.

If any test fails, video plus better meeting habits is the honest answer for now. If all three pass, we would start by measuring the latency path and GPU encoder headroom with a handful of real users at their real desks.

The training findings support the organisational tests. The Munich study found that the value of XR training does not come from the technology itself but from standardised procedures, repeatable curricula and measurable competency, and that organisations treating XR as a technology pilot tend to underperform.

Back See Blogs
arrow icon