
Xiaomi Smart Driving Doesn't Make Money from Software—So What Does It Make Money From?
Xiaomi Smart Driving Doesn't Make Money from Software—So What Does It Make Money From?
On October 21, at the 2026 World Intelligent Connected Vehicles Conference, Lei Jun repeated "without safety, everything else is zero" three times.HyperOS 1.18, rolled out a week earlier, is the product footnote to that statement—smart driving was renamed "Full-Domain Navigation Assist," enabling door-to-door driving all the way to the elevator entrance and charging-pile parking spots. The old questions remain: hardware maxed out, software free, betting on scale—how far has this road gone?

Promotional image for Xiaomi Auto OTA 1.18 urban navigation assist. Source / Xiaomi Auto official
Technical Route: Building the Foundation, or Selling Permissions
Lei Jun laid his cards on the table: Xiaomi's five-year cumulative hardcore R&D investment reached 105.5 billion yuan, with over 200 billion yuan expected in the next five years, covering chips, AI, and embodied intelligence; global AIoT connections exceed 1.1 billion devices, and the human-car-home full ecosystem is the trump card behind its heavy investment in smart driving. Applied to smart driving, the YU7 comes standard across all trims with LiDAR, 4D millimeter-wave radar, and a 700TOPS Thor chip, with end-to-end models trained on tens of millions of clips, further layered with a VLM vision-language model.
Xiaomi goes the opposite way—using the most expensive perception hardware to deliver a consistent experience across all trims and a data closed loop, bearing the cost on its own shoulders first. But the first-generation SU7 standard version has no LiDAR, and the 1.18 smart driving upgrade has nothing to do with it; "standard across all trims" is a promise for new models. The smart driving team exceeds 1,000 people, road testing exceeds 10 million kilometers, and the self-developed XLA cognitive large model has been deeply integrated with smart driving and will be rolled out in batches later.
Deployment Rhythm: Spread Breadth First, Then Fill in Depth
Breadth was spread two years ago: on October 30, 2024, "drivable nationwide" was pushed out across the country; depth came later, with the YU7 receiving 1.18.0 on October 12, 2026, starting parking-spot-level navigation, and on October 19, the SU7 and YU7 got the full rollout, with parking-spot-level navigation expanded to elevator entrances and charging piles, closing the loop on the last 100 meters. Breadth first, depth later.
The most noticeable change in 1.18 is that it "shuts up": lane-change voice prompts are removed, and acceleration/deceleration and detouring are tuned to be gentler, settling the account of "not detailed enough after rollout" round after round.

Xiaomi SU7 cockpit HAD environment perception interface and HUD head-up display. Source / Xiaomi Auto official
Commercialization: Not by Subscription, but by Scale
Xiaomi has not turned smart driving into a subscription; it is standard across all trims with free features, and the software is not charged for separately. In September, deliveries across all models hit a new high of 40,015 units (SU7 15,751, YU7 11,800, and Pengcheng exceeding 12,000 in its first month), and October is still at the 40,000 level, but the YU7 delivery cycle exceeds 30 weeks—demand is not the bottleneck, production capacity is.
Free standard features plus scale dilution mean monetization relies on "selling more cars and selling higher trims"Work from Creator Alliance, reviewed and approved.
Physix Frontier