QCraft targets 2026 mass production with Qualcomm's Snapdragon Ride

QCraft's urban NOA, running on Qualcomm's SA8650P, handled unprotected left turns and tunnel exits in live traffic — a milestone toward 2026 global deployment.

Qualcomm's Snapdragon Ride SA8650P powered live urban autonomous driving demonstrations in Wuxi, China, as partner QCraft targets global mass production of its Navigate-on-Autopilot solution by year-end 2026.

"QCraft's development on the Snapdragon Ride platform has entered the fast lane toward mass production," Dr. Dong Li, chief technology officer at QCraft, said during a keynote at Qualcomm's Automotive Technology and Cooperation Summit on June 5.

The SA8650P-equipped vehicles navigated unprotected left turns, mixed pedestrian-vehicle traffic, tunnels and congested road transitions with smooth, human-like control. QCraft completed development and on-road validation on both the SA8650P and higher-end SA8775P platforms in under a year since forming its strategic partnership with Qualcomm in September 2025. A higher-compute solution based on Qualcomm's QAM8797P platform is now in joint development.

The demonstration positions Qualcomm's Snapdragon Ride family — spanning the SA8650P, SA8775P and QAM8797P — as a credible challenger to Nvidia's Drive Thor and Mobileye's EyeQ platforms in the fast-growing ADAS and autonomous driving SoC market, projected to exceed $30 billion by 2030.

Mass production at scale

QCraft's QPilot assisted-driving solution has shipped on nearly 30 production models, with more than 50 additional models expected in 2026. Across its fleet, the system has logged more than 3.5 billion user-driven kilometers and over 100 million parking-assist uses. The company estimates the technology helps users avoid more than 146,000 potential accidents each year.

The system maintains an automatic emergency braking false-trigger rate of less than once per 500,000 kilometers — a critical reliability metric for automakers evaluating ADAS suppliers. By comparison, industry AEB false-trigger benchmarks from third-party tests typically range from one per 50,000 to one per 200,000 kilometers, according to published data from the Insurance Institute for Highway Safety.

World models and reinforcement learning

In his keynote, Dr. Li described the industry as reaching an inflection point toward general-purpose physical AI, with world models and reinforcement learning as the essential bridge. QCraft's cloud-based world model generates controllable, physics-aligned video for training, while a zero-shot engine uses natural language to synthesize long-tail and adverse-weather scenarios on command. The system runs low-cost closed-loop simulation for continuous reinforcement learning, allowing the AI to develop what Li called "defensive driving instincts."

The approach addresses a core challenge in autonomous driving: collecting enough edge-case data — a pedestrian in heavy fog, a construction zone at night — to train safe behavior. QCraft's synthetic data pipeline aims to reduce reliance on expensive real-world fleet testing.

Investor implications

For Qualcomm, the QCraft partnership provides a production-ready ADAS reference design that could accelerate Snapdragon Ride adoption among Chinese automakers, the world's largest EV market. Nvidia's Drive platform currently commands the highest-profile design wins in China, including Nio, Xpeng and Li Auto, while Mobileye supplies Volkswagen and Geely. Qualcomm's SA8650P, built on a 5nm-class process node (comparable to TSMC's N5), offers a mid-range compute option that could appeal to automakers seeking to balance performance and cost.

QCraft did not disclose pricing for its QPilot system or the per-chip cost of the SA8650P solution. The company's ability to scale from 30 production models to an expected 80-plus by end of 2026 will determine whether Qualcomm's Snapdragon Ride can capture meaningful share in a market where Nvidia's Drive Thor — sampling now with 2,000 teraflops of compute — targets 2025 production vehicles.

This article is for informational purposes only and does not constitute investment advice.