Momenta IPO: How a Chinese Startup Chose L2 Over the Waymo Path

Momenta went public on the Hong Kong stock exchange on July 8, 2026. After the first day of trading, the company was valued at 700 billion Hong Kong dollars (roughly $90 billion). The venture-backed startup, founded in 2016, took the opposite path from the rest of the autonomous driving industry.

Strategy: L2 Instead of L4

In 2018–2019, the entire AI industry was betting on L4 (full autonomy with no driver). Waymo received a $175 billion valuation, while PonyAI and Yushu Zhixing raised hundreds of millions. Capital talked up heroic hands-free driving.

CEO Cao Xudong chose L2: driver assistance (adaptive cruise control, lane keeping, partial steering). People considered it niche and boring. Funding in 2019 was slow. Doubts spread inside the company.

But Cao believed in the "data flywheel": if a system can collect data from millions of cars, it converges quickly toward perfection. Waymo's L4 fleet is limited (expensive cars, not many of them). L2 can be deployed at once across all production cars.

Results: 3,000 Times More Data

By 2026, the strategy had won.

• Market share: 60% of all L2 systems deployed on production cars in Europe
• Specific models: 50%+ for each of 210+ customized automotive platforms
• Deployed in the real world: 100+ million cars running Momenta software
• Team: just 1,000+ employees
• Margin: rose over three years from 17.5% → 49% → 71.6%

Cao told investors about a visit to Tesla to try Musk's FSD (Full Self-Driving). Tesla does this with 200–300 engineers, while Momenta uses 1,000+. He read that as inefficiency. Privately, he vowed: "We need to be faster and more efficient."

Engineering Efficiency as Strategy

All of Momenta's successes are tied to engineering efficiency. The company invested in automation tools: a CI/CD pipeline, road-testing robots, and data versioning. Cao personally tracks the top 10 tools.

One example is the "rt spa" tool, which automates the full road-testing cycle. Cao named the tool himself, led the first 10 architecture meetings, and attended every iteration. The result: an algorithm researcher presses a button and 4 hours later gets a finished quality report on the build.

The second is a system for preparing builds for release. How do you decide whether a new version is good? You have to look at 1,000+ metrics. Momenta automated this through a dashboard.

Competition and Challenges

The L2 market is no longer a niche. Competitors include Huawei, Horizon Robotics, Qingzhou, Waymo and Baidu Apollo. Prices for L2 systems have fallen. Qualcomm and NVIDIA compete on chips. New entrants (Wenmiao, Elementai) are trying to undercut Momenta on price.

Cao sees the future in software-hardware integration and world models for L4, but that still lies ahead (2027–2028).

What This Means

Momenta shows that the same principles hold in the AI industry as in software and hardware engineering: data > model size, iteration > a single breakthrough, engineering culture > genius. The company won not because it had the best scientists, but because it chose the right scale of data and invested in production engineering. It is a lesson for the whole AI industry: not everything has to be won in the lab.