NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

Megvii's founder sold 235 electric cars in a single month while his software rode in a million

Yin Qi co-founded Megvii at 23. Now he chairs an automaker and a large-model startup at once, while the car business he took over keeps shrinking.

2026-11-16 · 1164 words · NeuroAI
Megvii's founder sold 235 electric cars in a single month while his software rode in a million

Two numbers sit in the same company's disclosures in September and October 2026, and they do not belong together.

In September 2026, Qianli Technology (千里科技) sold 6,252 vehicles. Of those, 235 were new-energy vehicles (新能源汽车), down about 95% from a year earlier. In October, its chairman told a conference in Chongqing that the company's assisted driving (辅助驾驶) system had passed one million cumulative installations across 16 models, with 93.9% of equipped cars actually switching it on.

Selling 235 electric cars and shipping software into a million is not a growth story. It is a company in the middle of changing what it is — and the person running the change is Yin Qi (印奇), who built his reputation on facial recognition.

The first curve

Yin co-founded Megvii (旷视科技) in 2011 with two Tsinghua University classmates, Tang Wenbin (唐文斌) and Yang Mu (杨沐). He was in his early twenties and had just started graduate study at Columbia University. Megvii became one of the four Chinese computer-vision companies the industry grouped as the "AI four dragons" (AI四小龙), a label that described a business model as much as a technology: sell face recognition and visual algorithms to banks, phone makers, and city surveillance projects.

That model had a structural problem everyone could see and few could fix. Algorithms were sold as projects, priced per deployment, and the customer's budget cycle set the revenue rhythm. Margins looked like software; cash flow looked like systems integration.

Yin's answer, thirteen years later, was not another algorithm. It was a car.

Buying a car company with a balance sheet attached

In November 2024 he became chairman of Lifan Technology (力帆科技), a Chongqing automaker that went through bankruptcy restructuring in 2020 and, per Chinese financial press, was taken over by Geely and Chongqing state capital. The company was renamed Qianli Technology in February 2025 and adopted an "AI plus car" (AI+车) strategy; it trades in Shanghai under the ticker 601777.

The logic he has repeated in public since is consistent: a car is the only embodied product with enough volume, enough sensors, and enough willingness to pay to make AI a mass-market component rather than a project line item. He calls this physical AI, and in his telling the intelligent vehicle is where it lands first.

What he bought, though, was not a clean sheet. He bought a legacy manufacturer with factories, obligations, an internal-combustion business, and a controlling shareholder that is also its biggest customer and, according to Chinese financial press, its largest supplier.

Where the software comes from

The assisted driving system is called Qianli Haohan (千里浩瀚). Launched in March 2025, it has been pushed into Geely group brands — Zeekr, Lynk & Co, Galaxy, smart — which is exactly why the install base grew fast and why the number deserves care: a system shipping inside a partner's cars is a supplier relationship, not consumer demand for Qianli.

Per Chinese financial press, Geely consolidated its driver-assist teams in 2025 — the Zeekr ADAS group, Megvii's automotive unit, and part of Geely's research institute — into Qianli's autonomous-driving subsidiary, forming a team of roughly 2,000 engineers. Read plainly: the software capability was assembled inside the group and then listed through a company Yin chairs. That is a legitimate way to build a business, and it makes Qianli's fortunes unusually dependent on one customer's product plan.

The company has also pushed the regulatory front. In March 2026 it said Haohan received UN R171 approval and described it as the first Chinese driver-assistance system cleared for the European market — a claim worth treating as a company statement until the approval document is read directly.

The second company

In January 2026, Yin added another chairmanship: StepFun (阶跃星辰), the Shanghai large model (大模型) developer. The announcement came on the same day the company disclosed a B+ round of more than RMB 5 billion (≈US$694 million / HK$5.4 billion), which local business press described as the largest single round for a Chinese model company in twelve months.

The pairing is the interesting part. StepFun supplies models; Qianli supplies a vehicle to put them in. Yin has described it in interviews as a brain and a body, and the strategic bet is that a model company without a hardware channel cannot price its intelligence, while a car company without its own models cannot differentiate.

Announced intentions include a personal agent (个人智能体) in the car and phone-to-car handoff in the first quarter of 2027 — a dated, checkable promise.

The numbers that do not fit the story

Here is where the profile has to stop being flattering.

Qianli's own monthly production and sales filing for September 2026 shows total vehicle sales of 6,252, down 50.25% year on year, and new-energy sales of 235, down 94.69%. Nine-month cumulative sales were 66,903, down 7.83%. Some of that is deliberate: a legacy petrol line being run down. But a 95% drop in the category the whole story is built on is not a rounding error.

Meanwhile the software figures move fast enough to be hard to track. A Xinhua report in July 2026 cited 500,000-plus installations across 19 models at 92% activation; three months later the chairman cited one million across 16 models at 93.9%. Different denominators, different moments, all company-reported. Vehicle count and model count moving in opposite directions suggests the definitions are not stable.

None of this means the strategy fails. It means the company is currently a tier-one supplier with a car factory attached, valued on a software story, and the two halves are growing at very different rates.

Honest limitations

This profile relies on Qianli's own filings and on statements its chairman made at a public conference, both of which favour the company's framing. The install-base and activation numbers are self-reported and definitionally slippery; I could not reconcile the July and October figures, and I have said so rather than picking one.

I did not interview Yin, and there is no quotation from him here beyond paraphrases of positions he has stated publicly. His stake purchase, the Geely relationship, and the internal team consolidations are described from Chinese financial press; I did not obtain the underlying transaction documents. I also could not independently verify the UN R171 approval.

The sales decline may be largely a wind-down of unprofitable legacy models. I do not have the segment-level margin data that would settle it, and the company's profitability by segment is not something I could confirm from primary disclosure in the time available.

What readers can do now

  • Read the monthly production and sales filing on the Shanghai exchange before believing any transformation narrative; it is published within days of month end and it is unemotional.
  • Separate installations from adoption. A system installed in a partner's vehicles proves a supply contract, not that buyers chose it. Ask how many were activated by owners who paid for the feature.
  • Keep a dated list of promises. The 2027 first-quarter agent-in-car target is specific enough to check; most transformation stories are not.

Related coverage

Cite this article

NeuroAI (脑机网), "Megvii's founder sold 235 electric cars in a single month while his software rode in a million," 16 November 2026, https://neuroai.site/a/na-people-yin-qi-second-curve

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Builders” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories