cd ~/
Quest 3LaunchScaleVPSMeta

Quest 3 Launch: VPS at Consumer Scale

Quest 3 brings VPS to mainstream consumers - the largest deployment yet and what we learned.

Evyatar Bluzer
3 min read

Quest 3 launched with VPS enabled by default for MR experiences. It is our largest deployment by far, and the first week made that concrete.

Launch Numbers

500K+ devices activated in the first week. 2M+ VPS queries served. 50K+ unique locations attempted, with an 87% success rate in covered areas. We have never operated at anything close to this scale.

What Worked

The infrastructure held. We invested heavily in scaling through Q2 and Q3, and the launch load came and went without incident. On-device caching earned its keep: popular locations are cached on the headset, and 40% of queries got served locally. Developers were ready too - launch apps worked, and the first-party MR experiences performed well. As for the user experience, when VPS works, users do not notice it. That is the good outcome, even if it makes the work invisible.

What Didn't

Coverage gaps hurt the most. I said before Quest Pro that coverage gaps would be the main user-facing failure and that shipping at 88% success was good enough. Both held: 87% at consumer scale, and the loudest complaint is about where VPS is not. Plenty of users are in unmapped areas, and "VPS not available" is a frustrating first impression. Outdoor sunlight is still challenging; accuracy drops in bright conditions. I gave sunlight robustness three months at the MVP launch. That was twenty-seven months ago. Onboarding confuses people - users do not understand what VPS does or when it is working. And power consumption came in higher than target in continuous MR mode.

Three calls that came due at launchThree horizontal rows, one per earlier call, each a line from the month it was made to a dashed line at this post with a verdict at the right; the middle row carries a second, labeled point at March 2023; the top row, sunlight robustness, is highlighted.Jan 2022Jan 2023Sunlight robustness: three monthsSelf-serve mapping portalCoverage gaps; 88% is good enoughthis postMarch 2023 liststill challenging, 27 months ondated four times, now a lessonheld: 87% at consumer scale
Three calls that came due at launch: sunlight robustness was given three months in June 2021, the self-serve portal has been dated four times since then, and the September 2022 call that 88% success was good enough held at 87%.

User Feedback Themes

On the positive side: "Content stays in place - feels like magic", "Navigation actually works", "Shared MR with friend was amazing". On the negative: "Doesn't work at my house" (not mapped), "Glitchy outside in the afternoon" (sunlight), "Drains battery too fast". The negatives map directly onto our known-issues list, which at least means nothing surprised us.

Rapid Response

In the first month we expanded coverage to the top-requested areas, improved sunlight robustness with a server-side model update, shipped better error messaging, and pushed a power optimization patch. This is the advantage of server-side components: we can update them without a device update.

What did consumer scale teach us?

The long tail is really long - the locations users want are incredibly diverse, and we cannot map everything. Good enough beats perfect: 80% accuracy everywhere beats 95% in limited areas. Failure UX matters as much as success, because how VPS fails determines how users perceive it. And self-service is essential, since we cannot manually map everywhere users want to be. The March launch list had a self-serve mapping portal on it, the fourth time I have put a date on that tool since June 2021; it appears in this post as a lesson, not as a launch feature.

Looking Forward

Quest 3 establishes VPS as core infrastructure. The expansion list from here: more coverage through crowdsourcing and partnerships, better indoor performance (still behind), and glasses integration, the next frontier. Three years from concept to consumer scale.

Comments