AI-driven vehicle intelligence 2.0 has arrived, with the industry reaching a critical turning point for substantial implementation. On April 22, Horizon held its 2026 Product Technology Launch Conference, introducing China's first 5nm cockpit-drive integrated vehicle intelligence chip, "Starry Sky (Starry)." The chip incorporates the industry's first adaptive computing engine and the Fortress security physical isolation architecture, enabling the integration of intelligent driving, cockpit, instrumentation, and vehicle control domains. It is reported that the Starry chip will enter mass production and be installed in vehicles in the third quarter of this year. Simultaneously, Horizon released China's first vehicle intelligence operating system, "KaKaClaw," and an upgraded V1.6 version of HSD, the country's first mass-produced single-stage end-to-end assisted driving system. This marks Horizon's completion of a full-stack vehicle computing platform covering chips to software and intelligent driving to intelligent cockpit, officially leading China's smart vehicle industry into the "vehicle intelligence era."
Historically, smart vehicles utilized a dual-chip distributed architecture for cockpit and intelligent driving domains. This approach not only caused hardware redundancy, data transmission delays, and high per-vehicle costs but also severely limited the upgrade of intelligent vehicle experiences due to the independence and lack of deep collaboration between cockpit and driving systems. In recent years, major automakers have intensively promoted the mass production of cockpit-drive integration solutions. However, most current products remain at the level of "simple software-level coordination," failing to break the physical barriers between intelligent driving and cockpit domains and achieve true "intelligent integration."
Horizon's three new products—"Starry chip + KaKaClaw + HSD V1.6"—thoroughly establish the underlying computing platform for vehicle AGI, becoming a key force driving China's smart vehicles toward the central computing era. Integrated intelligence is entering an era of qualitative change, where cockpit-drive fusion is no longer just a cost-saving technical option but the core path for vehicles to become "vehicle intelligence agents."
Current smart vehicles mainly use distributed AI model architectures, where sensors like vision and acoustics operate independently from storage and transmission modules, with each functional module having its own dedicated "brain." This results in low coordination efficiency and computational waste. Breaking this fragmented pattern to achieve true cockpit-drive integration requires both software architecture innovation and underlying hardware support.
Dr. Yu Kai, founder and CEO of Horizon, emphasized that the future direction involves a centrally integrated AI model that unifies multimodal sensor data—such as visual, acoustic, facial features, and gestures—into one large model, which simultaneously outputs capabilities for autonomous driving decisions, cockpit entertainment responses, and human-machine interaction. Such a centralized AI model architecture relies on high-computing-power, highly integrated hardware foundations. Horizon's Starry series chips are pivotal hardware for realizing "vehicle intelligence agents."
The Starry series includes the Starry 6P and Starry 6H models. The Starry 6P, built on a 5nm process, features a 20-core CPU (over 500K DMIPS), a 3000GFLOPS dual GPU, a 4-core BPU Nash, LPDDR5x bandwidth of 273GB/s, and peak computing power of 650TOPS. It supports local operation of edge-side large models, simultaneous display on 6–12 screens, and deployment of cockpit digital AI and high-level assisted driving models. As China's first cockpit-drive integrated vehicle intelligence chip, the Starry chip adopts a single-chip integrated architecture, merging cockpit and driving domains, unifying computing power scheduling, and sharing DDR memory, thermal management, and domain control systems. This not only breaks the hardware barriers between intelligent driving and cockpit domains but also significantly reduces costs, such as DDR expenses.
Reports indicate that per-vehicle costs can be reduced by 1,500–4,000 yuan, with R&D delivery cycles shortened from 18 months to 8 months, and integrated cockpit-drive hardware and software delivery time reduced by 56%. Dr. Yu Kai highlighted that, amid current memory shortages and price surges, vehicles based on the Starry chip require only 28–40GB of DDR, compared to 48–64GB for traditional solutions, resulting in cost savings of 2,000–3,000 yuan from memory alone. The Starry series has completed tape-out and real-vehicle validation and has secured intended mass-production collaborations with several mainstream automakers and Tier 1 suppliers, including
Horizon aims to become the "Wintel of the Physical AI Era," believing that the three essential elements for vehicle intelligence are personality, skills, and memory. Only by enabling vehicles to actively perceive, make autonomous decisions, and continuously evolve can true "intelligent integration" be achieved, rather than mere "functional叠加." To this end, Horizon launched KaKaClaw, China's first Agentic Car OS. Dr. Yu Kai stated, "The future in-vehicle OS should not be just a software version but an intelligent agent that understands user intent, mobilizes vehicle resources, and proactively completes tasks." He added that whoever provides the best operating environment for autonomous driving and local large models will become the "Windows" or "Android" of vehicles.
Unlike traditional cockpit or driving OS, KaKaClaw centers on intelligent agents, integrating full-stack capabilities for intelligent driving, cockpit, vehicle control, and interaction. This upgrades vehicles from "passively responding to commands" to "actively understanding intent and autonomously planning and executing," turning the concept of vehicle intelligence into a tangible user experience. Users can issue natural language commands via voice, and the KaKaClaw system can concurrently schedule physical agents, digital agents, and edge-cloud large models across driving and cockpit domains, realizing an intelligent interaction paradigm of "task as a service." Notably, KaKaClaw will adopt an open deployment model, supporting not only Horizon's chip platforms but also other mainstream market chips.
Horizon's HSD, China's first mass-produced single-stage end-to-end assisted driving system, released its V1.6 version with significant upgrades in driving, parking, and safety. Dr. Yu Kai announced that Horizon will release HSD V2.0 within the year, further enhancing the practicality and safety of the intelligent driving system. Through deep collaboration between HSD and KaKaClaw, Horizon has successfully created an advanced Chinese version of FSD+Grok, transforming vehicles into intelligent agents that "can drive, chat, remember, execute, and grow."
With the launch of KaKaClaw and the Starry chip, Horizon stands out as one of the few players in the industry mastering hardware, software, intelligent driving, and cockpit domains. Dr. Yu Kai outlined Horizon's goal to become the "Wintel of the Physical AI Era," following a integrated hardware-software approach to provide the underlying computing platform and infrastructure for automotive AGI.
The ultimate goal for smart vehicles is to become physical AI agents with autonomous perception, decision-making, execution, and learning capabilities. Achieving this hinges on the architectural shift from distributed to centralized systems and the upgrade from functional intelligence to AGI, which is the only path to overcoming the stagnation in vehicle intelligence. Horizon's strategy is clear and firm: building on the millions of mass-produced Journey series chips, and centered on the Starry chip, KaKaClaw, and HSD V1.6, to create an integrated hardware-software and cockpit-drive vehicle computing platform, establishing the underlying link to vehicle AGI.
"Vehicles are the largest physical AI platform, and autonomous driving is the first application of physical AI," Dr. Yu Kai noted. He stated that vehicles are just the first step; Horizon's ultimate goal is to advance toward general physical AI, merging physical and digital AI to ultimately achieve AGI. Horizon predicts that 2025 to 2030 will be a period of rapid development for physical AI, with autonomous driving, as the first application, leading the integration with digital AI. Post-2030, the two will continue to merge in broader scenarios like factories and homes, ultimately progressing toward AGI, with smart vehicles likely becoming the earliest mature carriers in this process.
Over the past few years, Horizon has released five generations of Journey series chips, securing a leading position in the domestic intelligent driving computing platform market with millions of units mass-produced. Data shows that as of August 2025, cumulative shipments of Horizon's Journey series chips exceeded 10 million sets, maintaining the top spot among domestic intelligent driving chips. Additionally, Horizon's HSD has secured partnerships with over 10 automaker brands and more than 20 vehicle models domestically and internationally. In the next 3–5 years, HSD aims to achieve mass production of millions of units.
From the mass production of Journey chips to the breakthrough in cockpit-drive integration with the Starry chip, and the full-stack collaboration of KaKaClaw and HSD, Horizon is filling technological gaps in China's central computing field through innovation, providing a practical and scalable Chinese solution for the development of global smart vehicles.
Comments