After five years of rapid development in China's new energy vehicle sector, key features like driving range, fast charging, and batteries have become increasingly standardized. There is now a consensus within the industry that intelligent technology will be the decisive factor in the next phase of competition.
However, the cost structure of intelligent hardware is reaching a bottleneck. Equipping a vehicle priced around 200,000 yuan with both a dedicated autonomous driving System-on-Chip (SoC) and a separate cockpit SoC is becoming increasingly difficult to justify financially in the mainstream price segment. This dual-chip approach necessitates two sets of chips, systems, and cooling solutions.
The industry's response is cockpit and driving integration, merging the two chips into one. While this concept has been discussed for two years, most solutions remained theoretical until the 2026 Beijing Auto Show. This event served as a turning point, with companies including Qualcomm, Horizon Robotics, and Black Sesame Technologies showcasing their latest integrated solutions. The show even featured a dedicated exhibition area for this technology for the first time. Centered around Qualcomm's single 8775 chip, at least five Tier 1 suppliers presented their respective domain controller products.
The density of available solutions is now sufficiently high. However, a closer look reveals divergent strategic focuses—some are competing for chip market share, others for software definition rights, and still others to establish the foundational platform for整车智能.
This concentrated emergence is no coincidence. As intelligent driving progresses from highway Navigation on Autopilot (NOA) to more complex urban NOA, functional demands are rising while vehicle prices are under pressure. Furthermore, rising memory chip costs have increased the urgency for consolidation. The architecture of two separate chips, each requiring its own memory, is becoming a luxury. Advancements in manufacturing processes now allow single chips to provide sufficient computing power, creating a window of opportunity that has just opened.
While the number of available solutions is already abundant, mass production and delivery are only just commencing. The ensuing competition will not remain at the chip level, and the window for this market shakeout may be much shorter than the industry anticipates.
The limitations of the old dual-chip architecture extend beyond mere expense. The computing power utilization of autonomous driving chips is consistently below 30%, as most everyday scenarios do not require high-performance computing, whereas the Graphics Processing Unit (GPU) on the cockpit side is often fully loaded. With each chip maintaining its own redundancy, the overall resource wastage is significant. Horizon Robotics' Chief Architect pointed out on April 23, 2026, that a computing system's resource utilization should not exceed 70% to avoid instability. Dividing the system into two parts inherently creates more redundancy than a single system. Simply by eliminating static partitioning, integrated systems can reduce memory usage to two-thirds of the dual-system requirement.
"The more fragmented the system, the more memory it requires," he stated, "Therefore, integration is essential."
Rising DDR prices make the dual-chip approach even less economical. Automotive-grade memory is facing structural shortages; since September 2025, DDR4 prices have surged over 150%, and DDR5 prices have skyrocketed by 300%. An executive from Li Auto warned as early as late December that the 2026 supply fulfillment rate might be less than 50%. Integrating two memory sets into one not only saves materials but also provides a competitive edge in a tight supply environment.
Cost reduction figures are already being validated by mass production. At the auto show, Desay SV revealed that the 8775 solution can reduce costs by 20% to 30%. Horizon Robotics reported saving 50% on space, 30% on components, and reducing costs by 1,500 to 4,000 yuan.
Where do these savings go? They are redirected to the user experience. The Arcfox Alpha T5 is the first mass-produced model equipped with the 8775 integrated solution. A single chip with 144 TOPS simultaneously handles both cockpit functions and autonomous driving, incorporating urban NOA, highway NOA, and cross-floor memory parking, with a starting price of 105,800 yuan. This marks the first time urban NOA has been introduced in an SUV priced around the 150,000 yuan level. With a traditional dual-chip architecture, fitting these features into this price bracket would be impossible. The savings in Bill of Materials (BOM) costs are directly translated into making advanced driving capabilities more accessible.
According to forecasts, the compound annual growth rate (CAGR) for China's integrated cockpit and driving market from 2026 to 2030 is projected to be 36%, with a 3.6-fold growth potential by 2030. The driving force is not just cost reduction but also the potential for functional democratization enabled by those savings.
Beyond economics, there is an architectural imperative. The evolution of Electronic and Electrical Architecture (EEA) from distributed to domain-controlled to central computing makes cockpit-driving integration a prerequisite for centralization. The requirements for Level 3 human-machine co-driving are more direct; seamless takeover and handover between autonomous driving and the driver require microsecond-level communication running on the same chip. Separating these domains makes it difficult to reduce latency sufficiently.
The chief architect elevated this to a principle of physics, citing the history of computing. From the ENIAC with over 8,000 vacuum tubes failing every few hours, to today's vastly more complex single-chip computers that rarely fail. "What is the essence of the computer industry? It is integration, integration, and more integration. The question is not whether it is needed, but whether the level of integration is high enough and the number of components has decreased." Vehicle computing now has the opportunity to consolidate onto a single chip. The time window for dual-chip architectures is narrowing.
Different companies are approaching this integration challenge from different starting points. Qualcomm entered the automotive space via the cockpit. Its 8775 chip incorporates virtualization isolation and an ASIL-D grade safety island to add ADAS capabilities. Qualcomm does not develop application-layer software, focusing instead on selling chips and the platform. Tier 1 suppliers like Desay SV, Zoyte, ChinaTSP, Huayang, and Hosem are all developing solutions based on this single chip, competing for business.
Qualcomm profits from chip sales without engaging in software, a strategy that offers broad coverage but requires differentiation to come from software layers added by solution providers. Paths are already diverging, with some focusing on native multi-modal models and others on integrating cockpit, driving, and chassis domains.
The roster of vehicles featuring the 8775 chip expanded at the auto show, including models from joint ventures like Buick and Dongfeng Nissan—a significant signal indicating joint ventures sourcing intelligent solutions from Chinese suppliers.
Qualcomm itself is advancing rapidly. Its next-generation Snapdragon 8797 chip was unveiled at the show, boasting 700 TOPS per chip, a 12x improvement in NPU performance, and also supporting integration. Leapmotor announced the global debut of a dual-8797 system, and Qualcomm and Li Auto completed the mass production process for the 8797 in just 14 months. Sources indicate over 15 automakers plan to launch new vehicles with the premium platform this year.
The 8797's power is not solely for autonomous driving. It already enables on-device operation of a 30-billion-parameter Mixture-of-Experts (MoE) large language model, running both driving and cockpit models simultaneously—a demand far exceeding traditional split architectures. As the 8775 enters mass production, the 8797 is already poised to take over. This rapid iteration pace itself acts as a barrier to entry.
Horizon Robotics is taking a different path. Its founder and CEO used an analogy, comparing the integration challenge to climbing the same mountain from two different slopes, noting that moving from cockpit to driving is the steeper climb. Horizon's strength lies in autonomous driving. Just before the auto show, it launched the Starry 6P, China's first natively designed integrated cockpit-driving chip, built on a 5nm process with 650 TOPS. Unlike Qualcomm's virtualization, Starry uses a physical isolation architecture called "Fortress," separating the two domains at the hardware level with independent computing resource pools, preventing cockpit failures from affecting driving. The chip also includes an "adaptive computing engine" for dynamic resource allocation, shifting power towards cockpit rendering during normal cruising and towards driving in complex scenarios.
Horizon is pursuing a full-stack approach, developing the chip, driving software, cockpit OS, and more. iCAR is confirmed as the launch partner. Horizon claims its solution saves 50% space, 30% components, reduces costs by 1,500-4,000 yuan, and cuts the R&D cycle from 18 months to 8 months. However, each additional layer requires proving its value to automakers. Cockpit graphics rendering and OS ecosystem richness are barriers Qualcomm built over a decade.
Black Sesame Intelligent also has its unique approach. Its Wudang C1200 was designed from the ground up for integration, offering architectural purity as an advantage, though potentially limiting its customer base initially.
Currently, automakers need suppliers to solve problems beyond just architecture and chips; competition is moving upstream.
The competition is intensifying above the chip layer. At the Beijing Auto Show, Horizon Robotics launched not just a chip but also its KaKaXia OS, marking its official entry into the OS domain. It aims to be a platform provider, not just a chip seller, challenging Qualcomm's established cockpit software ecosystem. Its success will depend on performance in mass-produced vehicles.
Zoyte is taking another route, focusing on the model layer above the chip. It launched a native multi-modal foundation model pre-trained on universal physical world规律, aiming for cross-scene, cross-country reusability, transitioning from a solution provider to a model company.
Thundersoft and Zhicheng Yuan also showcased their cross-domain operating systems. The software layer above the chip is becoming as crowded as the chip layer itself.
Some are already ahead. Banma Zhixing, based on Qualcomm's 8397 and 8797 platforms, has deployed a large language model in vehicles, performing inference, decision-making, and planning entirely on-device, functioning offline in 90% of scenarios with data staying within the vehicle. This exemplifies the software differentiation occurring within the Qualcomm ecosystem; with the same chip, superior model performance and a more complete user experience will determine which Tier 1 supplier succeeds.
As one executive noted, everyone is searching for a "silver bullet." However, nearly two decades after the iPhone's launch, no Android device has truly matched its交互流畅性. Moreover, automotive safety is critical, and integrating cockpit and driving cannot be solved simply by deploying a large model.
In the absence of a silver bullet, what is the key? Horizon's founder points to ecosystem, stating that an open ecosystem is a game for the strong, requiring the ability to empower partners across chips, supply chain, and software. Horizon's president offered a simpler view: the key is ensuring every partner willingly collaborates a second time. He cited the "Apple supply chain" as an example, where value creation naturally fostered an ecosystem.
The industry's trajectory may have a reference point. The founder noted that the smartphone era ultimately had two major chip suppliers plus Apple's in-house design. The president added that final players in enterprise markets are few; "once a better option exists, nobody wants the second best."
The fact that five Tier 1 suppliers are differentiating on the same Qualcomm chip demonstrates that the key differentiator is not the chip itself, but the OS and model layers above it. As the number of chips reduces from two to one, the value contribution of software becomes even more critical.
The battle for a place in this evolving market is fierce. Chip companies and Tier 1 suppliers are vying for definition rights, and automakers are not standing still either. NIO has upgraded its central computing platform from a separated architecture to multi-domain fusion. Its self-developed chip handles both driving and some cockpit tasks, enabling cross-domain compute scheduling. While not strictly a single-chip solution, the direction is identical: breaking domain boundaries for free compute flow. NIO claims its chip's mass production cost is about 45% lower than NVIDIA's Orin-X, saving approximately 10,000 yuan per vehicle, demonstrating that in-house chip development can yield greater cost reductions than supplier solutions.
XPeng merged its autonomous driving and intelligent cockpit centers into a "General Intelligence Center" last year, signaling organizational integration ahead of product fusion—a typical precursor.
Suppliers are facing pressure from both ends. Above, automakers are moving down into the chip layer. Below, new entrants are pushing integrated solutions below the 100,000 yuan price point using domestic chips. ECARX has developed a single-chip platform integrating cockpit, driving, and parking based on its own "Lóngyīng No.1" chip. Cockpit-driving integration may transition from a premium feature to a standard offering faster than expected.
Scaling production also faces bottlenecks. Global memory giants are shifting capacity to High Bandwidth Memory (HBM), exacerbating structural shortages in automotive-grade memory. An executive noted that integration can reduce DDR usage, but only if the chip is sufficiently integrated; otherwise, memory constraints will hinder scaling. Safety certification is another slow variable, as full ASIL-D level validation can take longer than chip development itself.
Ultimately, the final stage of competition in cockpit-driving integration will not be about chip specifications, but about organizational capability and engineering culture.
Integration was intended to simplify—reducing chips, saving BOM costs, simplifying wiring and cooling. Yet, while hardware simplifies, competition in the software layer intensifies. Five Tier 1 suppliers are pursuing five different directions on the same chip, each aspiring to be more than just a hardware integrator. Whether they can truly transform into software companies remains an open question. Cockpit-driving integration simplifies the architecture but accelerates the industry shakeout.
Industry forecasts suggest a five-year growth window for this technology. However, the actual window of opportunity may be much shorter. The few thousand yuan saved in BOM costs will not merely impact profit margins by a few percentage points; it will determine whether advanced functions like urban NOA can be democratized from the 300,000 yuan segment down to 150,000 yuan, and further to 100,000 yuan.
The day a universal solution pushes this threshold below 100,000 yuan, any solution provider without mass production and delivery will not be playing catch-up—they will already be out of the game.
Comments