Is Spatial Computing Finally an Enterprise Decision, Not a Bet?

For a decade, "enterprise spatial computing" meant an expensive pilot that never scaled. In the last eighteen months, that stopped being true for a specific, testable set of reasons, not a vibe shift. Here's what actually changed, and how to tell whether your industry has crossed the line or is still being sold a demo.

Ask most operators over 40 what they think of enterprise AR and you'll get a version of the same story. Google Glass launched to fanfare in 2013, got rebranded within a year, retreated into an enterprise-only edition, and Google finally killed that too in 2023. Magic Leap raised over 2 billion dollars, sold roughly 6,000 units of its first headset in six months, pivoted hard to enterprise, laid off a thousand people, then laid off 75 more in 2024, and watched its valuation fall from 4.5 billion dollars to 2 billion. Microsoft's HoloLens found real industrial niches but never escaped "promising pilot" at scale.

That history is why "spatial computing" still triggers a reflex among CFOs: a technology that eats a budget line and returns a case study, not a return. The skepticism isn't irrational. It's pattern-matched, correctly, against a decade where the hardware, the software, and the economics all arrived years apart, and every company that bet on all three lining up got burned waiting.

That's the story most procurement conversations are still having in 2026. It's just no longer the accurate one.

Three things happened in the same window that never happened together before: the economics stopped requiring a bet, the platform stopped being borrowed, and the use cases stopped being demos.

Apple Vision Pro 2 launched in February 2026 at $2,499, a thousand dollars cheaper than the original, running an M5 chip with twice the on-device AI inference of its predecessor. That's not a price a company pays hoping costs fall later. Roughly 475,000 Vision Pro units have shipped since 2024, and by most estimates 75 percent went to enterprise buyers, more than 50 of them Fortune 100 companies. Enterprise-grade headsets from other vendors now run £2,000 to £4,500, priced and procured like specialized industrial hardware, not a moonshot.

visionOS 26, released in September 2025, closed the enterprise gap that kept IT and security teams cautious for two straight product cycles: a Protected Content API that blocks screenshots of confidential material, team device sharing so a shared pool of headsets can be managed like shared laptops, and a persistent Spatial Scenes framework for environments that survive being put away and picked back up. None of that existed a year earlier, and all of it is the unglamorous infrastructure that turns a device into a platform IT can actually approve.

And the use cases stopped being "imagine if." Spatial training environments are measurably cutting onboarding time 20 to 40 percent and removing the need for physical prototypes where a prototype is expensive to build twice. Remote expert collaboration, a field technician's headset streaming a first-person view to a specialist who annotates it in real time, runs on named, mature platforms like PTC Vuforia and TeamViewer Frontline, not custom pilots. Surgeons are rehearsing procedures on patient-specific 3D anatomy before the actual operation.

Call the point where all three conditions hold at once the Decision Line: the moment a technology stops needing a bet on the future to justify buying it now, because its economics, its platform, and its proof already stand on their own. Spatial computing crossed it sometime in the last year, not because the technology got more exciting, but because it got boring in exactly the ways that make a CFO comfortable.

Crossing the Decision Line doesn't mean every company should buy, and the market itself is sending a more complicated signal than the hype suggests. Meta exited enterprise Quest sales entirely in February 2026, choosing to concentrate on consumer headsets and its Ray-Ban smart glasses instead, whose $2.15 billion in 2025 revenue passed Quest hardware revenue for the first time. Apple's own Vision Pro unit volume actually dropped, to roughly 85,000 units in 2025, after a production halt at manufacturer Luxshare. Rising enterprise share and falling total volume are both true at once, and they say the same thing from different angles: the market is consolidating around who has a repeatable use case, not expanding toward everyone.

That consolidation matters differently depending on where a company sits. For manufacturing, healthcare, and construction, the industrial digital twins, the surgical rehearsals, the remote maintenance calls, the Decision Line has already been crossed, and the conversation is which vendor, not whether. For retail, media, and most consumer-facing product teams, the honest answer is closer to what it was in 2022: interesting, not yet load-bearing. The visionOS app ecosystem has grown past 3,000 titles, but most are still 2D ports rather than true spatial applications, which tells you exactly where platform depth exists and where it's still being built.

The uncomfortable middle is everyone else: teams whose leadership saw a Fortune 100 case study and want the same result without the same use case underneath it. That's the group still placing a bet dressed up as a decision.

The Decision Line gives a concrete test instead of a gut check. Before a spatial computing line item goes into next year's budget, three questions should have real answers, not projections:

1. 1.Does the unit economics work at today's price, not tomorrow's? If the business case depends on hardware getting cheaper, it's still a bet.
2. 2.Does the platform have enterprise features, or borrowed consumer ones? Device management, content protection, and compliance support are the difference between an IT-approved deployment and a shadow-IT pilot that never scales.
3. 3.Is there a named, repeatable production use case in the same industry, not just an adjacent one? A retailer citing a manufacturer's digital twin win is citing someone else's Decision Line, not their own.

If any answer is no, the honest move is a scoped pilot, not a platform commitment, the same three-to-five-device, one-high-value-use-case approach that's already proven out in the field. That's not a lesser strategy. It's a bet and a decision, told apart on purpose instead of by accident.

We've made a version of this argument before about AI agents: the pilots that reach production are built around a use case specific enough to prove, not a capability impressive enough to demo. Spatial computing is having the same conversation, a decade into its own version of that pattern, just with headsets instead of agents. The same discipline shows up in how we think about design system investment: start with one principle, one team, one proof, and let momentum earn the next commitment instead of trying to buy certainty upfront. And the cost of skipping that discipline looks the same wherever it shows up. Building early on infrastructure that isn't ready yet quietly compounds into debt nobody budgeted for, whether that infrastructure is an AI coding tool or a platform still finding its enterprise features eighteen months ago.

Spatial computing didn't become real because the demos got better. It became a decision because the boring parts, the pricing, the device management, the compliance APIs, finally showed up. That's the signal worth watching, in this technology and the next one.