Horizon Unveils All-in-One Vehicle Intelligence Stack

Horizon’s Starry chip becomes a full-stack vehicle brain.

Horizon Robotics pulled off a rare feat for a Chinese chipset maker on April 22, announcing a trio of products that complete a full-stack play: the Starry cockpit-driving intelligent chip, the KaKaClaw agentic vehicle operating system, and the HSD V1.6 full-scenario assisted driving system. In one launch, Horizon signals it intends to be more than a hardware supplier—an end-to-end platform for in-car intelligence. The Starry chip rides on a 5nm automotive-grade process and is billed as the computing backbone for both cockpit and autonomous driving workloads, backed by KaKaClaw and HSD V1.6 to run on top of it.

At the center is Starry 6P, described as delivering 650 TOPS of BPU computing power and 273 GB/s of memory bandwidth. Horizon emphasizes a unified memory architecture and software stack that replaces traditional, domain-separated subsystems with centralized computing. The claimed advantages are striking on paper: 50% reduction in space usage, lower per-vehicle costs by RMB 1,500–4,000 (roughly $210–$560), and a dramatic cut in development timelines—from about 18 months to 8 months. Horizon also says the integration cycle for cockpit and driving functionality is shortened by 56%, a time gain that matters as automakers push for faster model lifecycles and feature updates.

The company frames the Starry chip as the linchpin of a complete stack—from silicon to software to cockpit experience—making Horizon one of a handful of players capable of delivering end-to-end intelligent-vehicle compute. The KaKaClaw agentic operating system is meant to orchestrate perception, planning, and control, while HSD V1.6 promises full-scenario assisted driving across a range of road conditions. Horizon asserts that Starry has already attracted mass-production partners, a signal that China’s auto-intelligence ambitions are moving from pilots to production lines.

For automakers and suppliers, the implications are multi-faceted. Centralizing compute in a single stack could simplify vehicle integration, reduce cabling and do-away with some domain-specific ECUs, and potentially lower overall heat and power envelopes if optimized properly. But it also concentrates risk: a single vendor’s roadmap and software ecosystem would increasingly drive both cockpit and driving experiences, elevating the importance of long-term support, tooling, and migration paths for legacy models.

Two practical realities stand out for the supply chain. First, the 5nm automotive-grade node, while powerful, relies on highly constrained foundry and packaging ecosystems. Sourcing wafers, test, and advanced module assembly at automotive-grade scales will test Horizon’s commitments to ramp and reliability, especially as automakers push for ongoing OTA-driven feature expansions. Second, the 50% space reduction and cost drops hinge on the seamless integration of hardware with KaKaClaw and HSD V1.6. Any mismatch in software compatibility, toolchains, or validation cycles could erode those speed gains, underscoring a persistent risk in full-stack strategies: dependency on a tightly coupled platform.

From a market perspective, this move intensifies China’s domestic competition in auto AI compute. Horizon’s architecture mirrors a broader push to domesticize more of the vehicle AI stack—chips, OS, and driving software—reducing exposure to foreign suppliers and accelerating local innovation. Global buyers with China exposure will need to factor Horizon’s cadence into sourcing and program roadmaps, balancing the potential cost and time savings against ecosystem maturity and long-term supplier risk.

What to watch next: (1) whether Horizon can translate 8-month development cycles into sustained production throughput and feature updates across diverse vehicle architectures; (2) how automakers validate Starry’s performance across real-world driving scenarios beyond the lab; (3) the pace at which the KaKaClaw/HSD ecosystem expands, including developer tooling, certification, and third-party integration; (4) how capacity constraints at high-volume automotive nodes influence timelines and pricing as orders scale.

In short, Horizon’s trio of products marks a bold bet on a Chinese, end-to-end intelligent-vehicle platform. If the full-stack promise holds in volume production, the calculus for global automakers and chip suppliers could tilt toward a more China-centric, internally developed compute stack for cars—whether as a preferred supplier or a competitive alternative.

Sources

Horizon Unveils Trio of Breakthrough Products: Chip, OS, and Smart Driving System