The strongest near-term case for VR is in training, surgical simulation and rehabilitation rather than gaming. Those are the uses where a headset replaces something expensive, dangerous or hard to schedule. Enterprise training has published cost figures. A composite-organisation study summarised by VR.org reports a 219% return over three years: $6.1 million in benefits against $1.9 million in costs for a composite organisation of 10,000 employees, 3,300 of them trained in VR, with payback in under six months. In surgery, a Cochrane review of laparoscopic training found that VR improves trainees’ technical skills, such as faster suturing and better accuracy, but that its clinical impact is not known. VR pays for itself where the task, the number of learners and the comfort limits all line up, and it does not pay for itself by default.
Where virtual reality already pays back
Enterprise training is the first category, with published cost figures. The same VR.org summary puts VR training at roughly $115 per person over three years, nearly 50% below instructor-led methods. It finds VR breaking even with classroom training at 375 learners and coming in 52% cheaper at 3,000 learners. These figures describe a composite organisation of 10,000 employees, not a pilot with a few dozen trainees.
High-risk and high-cost training is the second category. A 2025 panel paper on immersive technologies in training and healthcare argues that complex immersive systems are most effective where real-world practice would risk life, health or expensive equipment. The same paper notes that, unlike physical simulators, VR systems need less space, are cost-effective and can be scaled for wide adoption. Its examples run from astronaut training for in-flight molecular diagnostics to procedural training for clinicians. Manufacturing training follows the same logic of risk-free practice. A VR training system for industrial work, with gesture-based controls and adaptive feedback, reported significant gains in knowledge retention and task precision in its own evaluation. Treat that as a promising single-system result rather than a general rate.
Healthcare is the third category, and it splits in two:
- Clinician training. VR training appears to cut operating time and improve operative performance for trainees with limited laparoscopic experience, compared with no training or box-trainer training. The Cochrane authors are explicit that the effect on patient outcomes and on costs to healthcare funders is unknown.
- Patient treatment. AR and VR are used to treat patients in domains including pain management, mental health, neurological disorders, surgery planning, intraoperative procedures and post-operative rehabilitation. In rehabilitation, gamified VR exercises (the paper’s example is fighting a virtual dragon with a sword while doing prescribed limb movements) are reported to improve engagement and adherence.
Our examples of VR in healthcare cover the treatment side in more depth. A wider catalogue of virtual reality use cases and examples lists applications beyond these three categories.
Is VR gaming the future, or will training drive VR adoption?
Gaming has a future, but training and other business uses are more likely to drive funded VR deployments in the near term, because those buyers can justify a headset with a payback calculation. GSMA’s Cloud AR/VR whitepaper describes the gaming bind plainly. Standalone or phone-based headsets have limited processing and can only run mild games, while PC-tethered setups deliver a better experience but are large, not portable and costly. GSMA concludes that these problems have greatly limited the popularity of VR games.
The enterprise side has different pressures. GSMA notes that enterprise business models are highly sensitive to owning fast-depreciating assets, and that enterprises want to turn heavy local technology into light, cloud-based, shareable resources. A training department can make that kind of argument. A household buying for play does not.
Headset market data shows a recent decline in shipments. IDC reported that worldwide AR/VR headset shipments fell 28.1% year on year to 1.1 million units in Q2 2024, after a quarter hit by a weak economy, rising average selling prices and excess inventory at Sony. In September 2024, IDC still forecast growth from 6.7 million units in 2024 to 22.9 million in 2028, a 36.3% compound annual growth rate. A forecast like this frames the direction of the market. It does not show that any particular use case will succeed.
More immersion does not guarantee more learning
It is tempting to assume that a more immersive lesson teaches more. A controlled study tested that assumption on a science lesson in VR. Students in the immersive VR version reported much higher presence (d = 1.30) but learned less (d = 0.80), and showed significantly higher cognitive load on an EEG measure (d = 0.59). The authors’ reading is that VR can overload and distract learners, which leaves them less room to build understanding.
This does not contradict the training gains above. It narrows where they apply. VR.org quotes a PwC finding that VR learners are four times more focused and 2.3 times more engaged than classroom learners. VR earns its place when it practises a spatial, procedural or physical skill that slides and videos cannot rehearse. It earns it less often when it wraps conceptual content in a 3D scene. A project should name the skill it is meant to improve and test that skill. “Engagement” alone is not a goal. Our piece on VR in education looks at classroom uses with this lens.
Cybersickness and latency set the comfort ceiling
Cybersickness is still one of the primary barriers to prolonged, comfortable VR use. Qualcomm’s guide to reducing cybersickness reports that it affects 50% to 80% of VR users and calls it one of the largest barriers to adoption. The cause is a sensory mismatch: the eyes see motion in the viewport that the vestibular system does not feel. Symptoms include disorientation, eyestrain and nausea, and dizziness and poor coordination can continue after the session ends. For training, the cost is direct. Symptoms impair task performance and shorten how long people can stay immersed.
Latency matters for comfort too. GSMA gives a commonly targeted motion-to-photon latency of 20 ms for the full chain from head movement to updated display. It warns that users may feel dizzy if latency repeatedly exceeds that budget. GSMA also says the headset should process at least 90 frames per second, because at 90 to 120 fps the eye stops noticing screen jitter. Delivering 90 fps at the resolution a training scenario needs, on a standalone mobile chip or a shared cloud GPU pool, is a GPU performance engineering problem.
Cybersickness onset and progression also vary substantially between people, which matters for long sessions. A study of early eye-gaze dynamics found that fixation dispersion in the first two to four minutes was consistently associated with how discomfort later progressed, but not with final symptom severity. The authors suggest this could feed closed-loop VR systems that monitor discomfort and adapt before symptoms become severe. The FDA’s list of probable risks for medical AR/VR devices adds head and neck strain, cybersecurity and privacy risks, and unknown effects on vulnerable groups such as children. We cover these and other challenges of virtual reality separately.
Technology advances that will shape the future of VR
Cloud GPU rendering and thin-client headsets
Cloud rendering moves VR processing off the headset and into the cloud. GSMA describes cloud VR as running and rendering applications on cloud processing resources and streaming the view to thin-client headsets. Those headsets would be much cheaper than self-contained ones, could last longer and could give better battery life. The catch is infrastructure. Operators have to add GPU resource pools to their clouds and data centres, and GSMA notes they usually have no experience running that kind of hardware.
Latency does not go away. It moves to the network. For interactive AR/VR, GSMA puts end-to-end network latency below 10 ms and uses edge computing to shorten propagation delay. Its rule of thumb is that light in fibre takes about 1 ms round trip per 100 km. Moving GPUs closer to users lowers latency, but deployment and operating costs rise sharply the further out the infrastructure is spread. On the headset, asynchronous timewarp and asynchronous terminal-cloud rendering generate intermediate frames from local head motion, so motion-to-photon latency depends on the terminal rather than the network. Even so, GSMA reports that current systems meet latency needs for weak and moderately strong interaction but not for strong or ultra-strong interaction. Weakly interactive services such as on-demand and live-broadcast cloud VR already meet the requirement. Strongly interactive services such as cloud VR games do not.
Hand and eye tracking as input
Hand tracking now runs mostly on infrared cameras built into the headset, and devices such as the Meta Quest 3 add two RGB cameras for better tracking. Research also covers hand redirection, which uses change blindness so users can work in a small physical space while interacting with a larger virtual one. Eye tracking has a second job beyond input: the gaze research above suggests it could become the sensor that lets a system detect cybersickness building and adapt to it.
Displays and form factor
Display resolution is improving, but slowly. GSMA notes that headset resolution still lacks the detail of high-definition video, and that rendering at higher resolution needs more local compute, which makes the platform more expensive. For cloud VR, it sets a minimum of 2560×1440 pixels for entry-level terminals. GSMA expects the pace of the market to depend on breakthroughs in battery life, cooling, processor density and optics at mass-market prices. VR.org reports headset prices down roughly 60% since 2016.
How mixed reality changes where VR is used
Mixed reality brings virtual objects into the user’s physical surroundings instead of a sealed virtual world. Meta’s developer documentation defines MR as blending virtual objects with the user’s physical environment so that virtual elements interact with both the user and the space around them. The use cases Meta describes are telling:
- Classroom lessons where co-located students handle virtual historical artefacts, or take notes in a physical notebook while dissecting a virtual frog.
- Fitness sessions where a trainer’s avatar appears in the user’s living room via passthrough.
- Virtual offices where the user’s real keyboard stays visible through object tracking.
Each keeps part of the physical room in view: notes, the keyboard, the people nearby. IDC expects MR devices to make up over 70% of the total device volume IDC forecasts for 2028, and expects AR headsets such as HoloLens to give way to more affordable MR devices that cover some of the same uses. IDC also says that dedicated AR devices have proven hard to produce at scale while overcoming barriers of price, battery life, thermals and optics.
MR also changes the compute load. GSMA points out that cloud-based AR, unlike cloud VR, has to analyse incoming sensor data such as stereo camera feeds, so upstream bandwidth can match or exceed downstream. For the spatial side, see our explainer on mixed reality and how VR, AR and XR fit together, and for the see-through-glasses track, the future of augmented reality.
Deciding whether a VR project is worth funding
A VR project is worth funding when you can name the skill or treatment it improves, show that enough people will use it to pass break-even, and keep sessions within comfort limits on hardware you can manage.
| Question to settle | What to check | What tips it towards funding |
|---|---|---|
| Is the real-world alternative costly or risky? | Cost of live practice: equipment, downtime, safety exposure | Practice that would risk life, health or expensive equipment, where immersive training is most effective |
| Will enough people use it? | Learner numbers over the programme’s life against the cost of content | Volume well past the 375-learner break-even VR.org’s composite study reports against classroom training |
| Does immersion help this learning goal? | A pilot that measures the target skill, not presence or engagement | Evidence that immersion helps the target skill; one science-lesson study found more presence but less learning |
| Is outcome evidence needed? | Whether funders need patient or business outcomes, not training scores | Fields where gains in trainee performance are enough, given that clinical impact is still unproven for surgical VR |
| Can people stay in the headset long enough? | Session length, motion design, frame rate and latency on the target device | Stable 90 fps and motion-to-photon latency under 20 ms |
| Who owns and runs the fleet? | Device management, content updates, privacy and cybersecurity | A plan for the privacy and cybersecurity risks the FDA flags, and for how quickly headsets depreciate |
| Where does rendering happen? | On-device, local PC or cloud/edge GPU | Cloud or edge only where the network can meet the interaction level the scenario needs |
The second row is the one we would check first. In VR.org’s figures, VR’s cost advantage grows with learner numbers, from break-even at 375 learners to 52% cheaper at 3,000. A programme with a few dozen trainees a year sits far below the 375-learner break-even VR.org reports; one onboarding thousands of staff sits well above it. Before arguing about which device to buy, the funding question to answer is whether your volume is closer to the first case or the second.
FAQ
Is VR healthy for your brain?
The FDA lists probable risks of medical AR/VR devices, including cybersickness, head and neck strain, and unknown effects in children and other vulnerable groups. Qualcomm notes that dizziness and lack of coordination can continue after a session. In one study, immersive VR also raised cognitive load as measured by EEG.
Which technology is used in virtual reality?
A headset combines displays with tracking. Modern devices track head position as well as orientation (6DoF), because orientation-only tracking cannot capture changes in head position. Hand tracking typically uses built-in infrared cameras, and rendering runs on a local GPU or, in cloud VR, on remote GPUs that stream frames to the headset.
