Horizon Robotics Deploys HSD V2.0 Amid Customer Chip Self-Development Risk
Horizon Robotics rolled out its largest-ever software upgrade on June 30, but the update arrives as the Chinese autonomous driving chip supplier faces a structural reckoning: its two biggest customers are building the very technology they currently buy from it.
The HSD V2.0 over-the-air update — spanning six capability dimensions, 18 new features, and 25 experience refinements — is the most comprehensive iteration since the system's launch. Initial deployment has begun on iCAR V27 vehicles, with a broader rollout to other partner automakers to follow. At face value, the release demonstrates Horizon's continued software execution. Viewed against the company's 2025 financials and a rapidly shifting competitive landscape, however, it reads more like a defensive signal than a growth catalyst.
Markets have taken note. Horizon's positioning as China's leading domestic alternative to Nvidia in automotive domain controllers is being stress-tested simultaneously on three fronts: customer vertical integration, deepening losses, and a maturing market where willingness to pay for intelligent driving features is declining.
Upgrading Performance Where It Counts for Mass-Market Buyers
The V2.0 release targets three practical use cases that directly affect daily driver experience in Chinese urban conditions. On highway and city driving, Horizon claims a 56% improvement in autonomous mileage without human intervention and a 167% gain in handling contested traffic scenarios — merges, roundabout entries, and close-range lane changes that previously triggered overly conservative system responses.
On parking, the system abandons reliance on pre-mapped standard bays. It now autonomously identifies usable space in irregular configurations common to older residential compounds across Chinese cities — a meaningful differentiator in a market where parking infrastructure quality varies dramatically.
Safety response has also been overhauled. The environmental perception model previously reserved for intelligent navigation has been extended to emergency braking, emergency steering, and unintended acceleration prevention. Horizon states overall system reaction speed improved 20%, with response latency 34% faster than the second-ranked competitor — a metric the company did not attribute to a named rival.
All enhancements were achieved within the existing technical architecture, with no new chip platform required. That constraint is both a strength — proving software-layer value without hardware dependency — and a limitation, as it underscores the ceiling imposed by current silicon.
Key Customers Are Developing the Chips They Currently Purchase
The most acute structural risk facing Horizon centers on BYD and Li Auto, which Horizon CEO Yu Kai confirmed on a recent earnings call remain the company's highest-volume customers by shipments. BYD's introduction of its proprietary Xuanji A3 automotive-grade chip has generated market concern that order volumes could migrate in-house over time.
Yu Kai's recent visit to BYD headquarters was interpreted as a diplomatic effort to stabilize the relationship, and he reiterated publicly that both automakers remain top-tier Horizon customers. Yet the underlying dynamic is unmistakable. BYD's chip self-development is not an isolated move — NIO, XPeng, Li Auto, and Tesla have all entered chip design to varying degrees. A Gartner forecast from 2021 projected that 50% of the world's top ten automakers would be designing their own chips by 2025. That threshold has effectively been reached.
Horizon's forward guidance — over 100 vehicle models equipped with its chips expected to launch in the coming year — suggests near-term volume stability. But each incremental self-development announcement by a major OEM narrows the addressable market for third-party suppliers structurally, regardless of short-term order books.
Losses Widen as R&D Burns Past Revenue
Horizon's 2025 annual results reveal the financial cost of competing in this environment. Full-year revenue reached RMB 3.758 billion (US$521.9 million), while net loss reached RMB 10.469 billion (US$1.454 billion) — a sharp reversal from a RMB 2.347 billion (US$326 million) profit recorded in 2024. Research and development expenditure surged 63.3% to RMB 5.154 billion (US$715.8 million), equivalent to 137.1% of total revenue. For every renminbi earned, the company spent RMB 1.37 on R&D.
Margin pressure is set to intensify further. To defend mid-market share, Horizon is cutting the price of its Journey 6P flagship chip by 15% — a volume-for-margin trade-off that will compress hardware gross margins in the second half of 2026. Multiple institutional analysts estimate that, assuming no deterioration in Horizon's competitive position, the company's earliest realistic path to profitability is 2028.
The loss trajectory is not unusual for companies in the capital-intensive phase of automotive intelligence platform development. However, the combination of accelerating R&D spend, pricing concessions, and a contracting third-party supplier opportunity window creates a narrower runway than the company faced two years ago.
Nvidia Dominates the High-End Segment Horizon Needs Most
The third structural challenge is competitive positioning within the domain controller chip segment — the highest-margin, highest-growth tier in automotive intelligence hardware. According to NE Times data for April 2026, Nvidia held 50.9% of domestic domain controller chip installations in China, with Horizon second at 13.6%.
Nvidia's share is anchored by its Orin-X and Thor platforms, which are the silicon of choice for premium models from Li Auto, NIO, and Xiaomi. Horizon's Journey 6 series ships primarily in mid-tier 6M and 6E configurations targeting family-segment vehicles, with only the 6P attempting to compete at the premium end. The compute ceiling and high-end customer pipeline gap between the two companies remains substantial.
User demand dynamics compound the challenge. A 2025 intelligent driving white paper published by Autohome Research Institute, drawing on surveys of more than 2,600 consumers, found that willingness to pay for intelligent driving features has declined measurably compared to 2022. The market is transitioning from early-adopter enthusiasm to mainstream price sensitivity — precisely as Horizon needs to justify premium positioning.
Software Iteration Buys Time; Structural Answers Remain Outstanding
HSD V2.0 is a competent, well-executed software release. It closes real experience gaps in urban driving, irregular parking, and safety response, and it demonstrates that Horizon's engineering organization can deliver meaningful iteration within a stable technical framework. The initial rollout to iCAR V27 owners provides a controlled proving ground before broader deployment.
What a single OTA cannot accomplish is neutralizing the three structural forces converging on Horizon's business model: customer self-sufficiency in chip design, a loss profile that requires sustained investor confidence through at least 2028, and a competitive ceiling imposed by Nvidia's dominance in the high-value domain controller segment.
Yu Kai has articulated a long-term thesis anchored in integrated hardware-software development and mass-market deployment of full-domain intelligent driving. The strategic logic is coherent. Execution, however, will require Horizon to demonstrate that its integrated model generates defensible differentiation — not just incremental feature updates — before its largest customers complete their vertical integration journeys.
The Chinese intelligent driving industry has moved decisively from a growth narrative phase into a margin, moat, and proprietary-value competition. Horizon is not yet on the wrong side of that transition. But HSD V2.0, however solid, is a holding action rather than a resolution.
Related Coverage:
Horizon Robotics Launches Unified Architecture to Challenge Tesla's FSD Dominance in China