The spatial computing enterprise pitch has had two years to prove itself. The pattern in 2026 is now clear enough to talk about with some confidence: a few use cases have produced genuine repeat purchases and budget continuity, and a much larger set have produced one-off pilots that didn’t get renewed.

Worth being specific about which is which.

What’s actually sticking

Three categories have generated meaningful repeat enterprise spending.

Industrial training, particularly in scenarios that are dangerous, expensive, or impossible to recreate in the physical world. Mining, oil and gas, manufacturing line training, and complex equipment operation have produced repeat purchases because the alternative is genuinely worse — incident risk, lost production time, equipment damage. The training programmes that have stuck share a common pattern: they were designed by industrial training professionals working with VR developers, not designed by VR developers being told what training looks like. The content quality matters more than the hardware, and the operations who treated this as a training programme that happened to use VR have done better than operations that treated it as a VR programme that happened to be about training.

Architectural and design review for high-value built environments. Hospitals, infrastructure projects, large commercial fitouts. The use case is concrete: stakeholders walk through a planned space at scale, identify issues that don’t appear on plans, and reduce expensive late-stage changes. The ROI is measurable, the buyer is identifiable, and the use case has produced repeat purchasing. The pattern that works is integration with the existing BIM and architectural workflow, not standalone VR review tooling.

Surgical and procedural simulation in healthcare. The training content quality has improved enough that medical educators are integrating it into formal curricula. The use case is differentiated from gaming-grade hardware needs and has its own product category developing. Capital purchases by teaching hospitals and surgical training networks have been quiet but consistent.

What hasn’t stuck

A larger number of use cases have produced enthusiastic pilots and quiet retirements.

VR for collaboration and remote meetings. The pitch has been pursued since 2017 and hasn’t worked at any pilot. Comfort, calendar friction, software ecosystem fragmentation, and the general superiority of Zoom plus screen sharing for actual work have all conspired against it. There are exceptions for specific design and engineering review workflows, but those are properly described as design review, not collaboration. The general meeting use case is dead, and operators who keep pitching it should stop.

VR for general training where the alternative is a video. The pilots that compared “watch this video” against “complete this VR module” have routinely shown either no measurable learning gain or a small gain that doesn’t justify the per-seat cost. The training categories where VR works are the ones where video doesn’t work either. Where video works, video wins on cost.

VR sales tools for property and retail. The early enthusiasm for showing customers virtual properties or virtual stores has produced limited repeat business. Customers can mostly visualise what they need to from photographs and videos. The added value of VR doesn’t justify the friction. Some niche applications work — high-end off-plan property, particularly for international buyers — but the broader use case hasn’t scaled.

Office productivity and AR overlays for general knowledge work. The pitch that AR glasses would replace monitors for common office work hasn’t survived contact with real users. The form factor isn’t comfortable enough for an eight-hour day, the resolution doesn’t match good monitors, and the workflow advantages haven’t materialised. The interesting AR work is happening in domain-specific applications, not in general productivity.

What enterprise buyers are doing differently

The buyers who have built genuinely successful spatial computing programmes share a few habits.

They picked use cases where the alternative was actually worse, not where the alternative was just adequate. The “wouldn’t this be cool” pilots have mostly failed. The “we have a real problem and this is the best available solution” deployments have mostly worked.

They invested in content development as a separate discipline from hardware procurement. The hardware is comparatively easy to procure. The training content, design tooling, or domain-specific application is where the work is, and where most failed programmes have under-invested.

They built institutional capability rather than relying on vendor delivery. Programmes that depended on a single VR vendor for ongoing content and operational support have generally fared worse than programmes that built internal capability and treated the vendor as a tooling supplier.

They sized expectations realistically. The vendors who promised transformation produced disappointment. The buyers who expected modest measurable gains produced sustainable programmes.

What I’d watch

Three things over the next 18 months.

The shipment of substantially lighter, longer-battery-life enterprise headsets. Form factor improvements alone could reopen use cases that have been ruled out by comfort and battery limitations. Several manufacturers have signalled meaningful hardware updates for 2026; whether they actually land matters.

The maturity of authoring tools for enterprise spatial content. The cost of producing high-quality training content has been the binding constraint for many programmes. AI-assisted content authoring tools are advancing quickly and may bring the cost down enough to expand the addressable use cases.

The emergence of clearer vertical-specific platforms versus general-purpose hardware. The horizontal “spatial computing platform” pitch has produced limited results. Vertical-specific solutions — surgical training, mining safety, industrial design — have produced clearer value. Whether the platform players can build the vertical depth, or whether vertical specialists can scale across markets, will shape the next product cycle.

The honest summary for 2026: spatial computing in the enterprise is a real but narrower business than it was pitched as. The use cases that work, work well. The ones that don’t have had enough time to show that they don’t. Buyers who can tell the difference are spending sensibly. Buyers who can’t are still being sold dreams.