Xiaomi Phase II Passed Inspection, Don't Rush to Claim Doubled Capacity

The most valuable information in this article is that Xiaomi Auto's Phase II factory didn't receive a "production qualification certificate," but rather a construction engineering archive acceptance opinion within a chain of compliance. Approval passed on September 2nd, but there's still planning acceptance, housing verification filing, equipment joint debugging, production line ramp-up, vehicle consistency, and supply chain stability before true mass production. The headline says capacity is expected to double in the future; my first reaction was that this demo is far from mass production.

I work in embodied intelligence and have seen too many beautiful videos. Robotic arms grabbing objects, humanoid robots walking, unmanned delivery vehicles weaving through cones—in the lab, everything looks successful. But once they enter parks, workshops, or residential communities, problems arise: uneven ground, weak signals, battery degradation, no maintenance personnel, and model changes requiring re-labeling data. Last month I tested YiKa Smart Car's unmanned delivery vehicle system—drawing maps in the backend, approvals, testing. The car isn't big, but running the process revealed that the real trouble isn't whether it can move, but whether the scenario can close the loop.

Xiaomi's Phase II is the same. It's not "the factory is built," but that the project has begun undergoing engineering archive-level review. This node is boring but critical. It breaks a grand narrative into verifiable, auditable, and financeable nodes. I flipped through several reports. Public metrics state that after Phase II production begins, the two factories provide a total rated annual capacity of 300,000 units; some say annual capacity exceeds 300,000 units, laying the foundation for this year's full-year delivery target of 350,000 units. The numbers aren't entirely consistent, which is actually normal. Capacity planning is planning; delivery is delivery.

Headline Narrative	Actual Node	Core Metrics Startups Should Watch
Capacity doubled	Archive acceptance passed	Weekly stable offline volume
Factory completed	Fixed asset formation	Unit manufacturing cost
Annual capacity >300k	Planning metric	Yield rate and first-pass yield

This table looks simple, but many startup teams don't want to look at it. Everyone loves talking vision, platform, ecosystem, and the next Tesla. What are the core metrics? It's not the annual capacity on the PPT, but whether the same product can be delivered stably for four consecutive weeks, whether failure rates rise with batches, and whether after-sales costs eat up gross margins.

Xiaomi team's execution deserves a separate mention. From land acquisition, construction, planning acceptance, to archive acceptance, the pace isn't slow. Public plans stated Phase II completion in June, official production in July/August. Yet archive acceptance approval was only seen in September; obviously, there were bottlenecks in between. However, delays in engineering projects are common. The key is whether nodes are transparent, responsibilities are clear, and problems are broken down into solvable sub-items. What startups fear most isn't slowness, but slowness that can't be explained.

Another trap in hardware startups is treating capacity as a business model. Cars aren't sold via PPT; they're sold via delivery. Doubling capacity doesn't mean doubling profit. The more cars, the more inventory, finance, insurance, repairs, spare parts, and channels amplify. Public news shows Xiaomi Auto has 352 stores in 97 cities and is continuing to open more. Channel expansion and factory production must match; otherwise, cars roll off the line but users can't pick them up, or after-sales can't keep up after pickup, and reputation will backlash.

The embodied intelligence track currently has a bunch of bubbles like "build the robot first, wait for the scenario." Some talk about general-purpose robots during fundraising, household services, industrial handling—every market sounds huge. But I care more about where the first dollar comes from, whether the first customer is willing to pay for labor savings, and whether the first delivery cycle can be compressed to an acceptable range. I used to think as long as algorithms are strong, hardware would catch up sooner or later. My thinking has changed now: algorithms are the entry point; manufacturing and delivery are the barriers.

So seeing Phase II pass acceptance, I won't get excited about "doubling," nor mock "not yet in production." It shows Xiaomi pushed the hardest part of project management forward one step. What's truly worth learning is breaking big goals into acceptance forms, milestones, responsible persons, and cash flow nodes. For startups, shout fewer slogans and watch the beat more. Wait until weekly offline volume, first-pass yield, and delivery cycles are stable for four consecutive weeks before talking about mass production.