Chinese automakers are self-developing EE architectures but still rely on overseas high-end SoCs. Domestic chips face long verification cycles and adaptation challenges; Chenzhi Semiconductor partners with Elektrobit to break the deadlock.
Beneath the booming landscape of China's new energy vehicle industry, there runs a taut, hidden undercurrent.
While outside attention and hot money remain fixated on the sensory thrills of refrigerators, TVs, and oversized rear seats, the real contest in the deep end has already switched scripts. Over the past two years, dozens of new models have been launched into the domestic market every month, dropping like dumplings into boiling water. Amid this frantic cadence of new releases, leading OEMs are accelerating their pivot toward cross-platform, cross-model software platforms and have begun designing the next generation of electronic/electrical (E/E) architectures in-house. The blueprints are indeed drawn by their own hand. But when the pen hits the paper, they discover that the "pencil" they need to precisely sketch out the underlying logic—high-end system-on-chips (SoCs) with high compute power and high reliability—is still firmly gripped by overseas giants.
This is by no means a simple story of "domestic substitution" in the supply chain. It is about who truly holds the "right to define" the underlying architecture of the next generation of vehicles. As the soul of the automobile shifts entirely from mechanical engagement to code-driven instruction, the close coordination between chips and foundational software directly dictates the vehicle's responsiveness and iteration efficiency. If the grand vision on the blueprint cannot be unleashed without reservation on domestic hardware, then China's automotive industry breakthrough in the second half of the intelligent driving race remains a dance performed with a hand at its throat. The window of opportunity for the industry to find this hard-core domestic pencil has already shrunk to an extreme degree.
Farewell to the "Black Box"
To understand why this "pencil" is so scarce, you first have to peel back the technical reality of Software-Defined Vehicles (SDV).
As vehicle architecture shifts from distributed to centralized, the car's functions are no longer a monolithic slab welded shut at the factory, but a living organism that can be infinitely extended through software. Functions once realized through pure hardware are now being taken over in batches by vast, complex software code. In the past, cars were stuffed with dense bundles of wiring and hundreds of independent ECUs. Now, these scattered control points are converging toward central computing platforms.
To be honest, the development logic of the global automotive industry over the past few decades has been rigidly compartmentalized. OEMs bought sealed "black boxes" from Tier 1 suppliers, and as long as basic communication between these boxes remained functional, that was sufficient. But in today's Chinese market, this inefficient and expensive playbook has completely ground to a halt. The involution of the domestic auto industry is going through phased evolution: the first phase was competing on cost and point technologies; now, the industry has formally entered the second phase—competing on efficiency and ecosystems. To squeeze out maximum efficiency, both hardware and software must be fully white-boxed. OEMs are reaching downward, chipmakers are reaching upward, and everyone is crossing boundaries to find new ecological niches.
But amid this vigorous reconstruction of blueprints, the support from domestic foundations looks thin. Even by the most optimistic estimates, China's overall automotive chip self-sufficiency rate is below 20%, and the localization rate for high-end automotive-grade chips is even lower. These figures mask a severe structural imbalance: what has been localized so far is mostly power devices, while in the realm of complex SoCs and high-end MCUs—which truly set the ceiling for intelligent capability—the localization rate still hovers at a feeble single-digit percentage.
The OEMs have drawn the new blueprints themselves, so why is the adoption of domestic high-performance chips still so slow?
The crux of the problem isn't a lack of willingness, but the objective existence of a "validation wall." The automotive industry demands extremely high safety standards. A complex SoC, from first silicon to actual mass production in a vehicle, faces an extremely long validation cycle and a massive amount of adaptation work. These chips carry complex underlying software. When an OEM switches to a domestic solution, it not only has to contend with stringent audits for functional safety and cybersecurity, but also has to throw massive manpower at re-running the entire software adaptation stack for the new architecture.
In a landscape where architecture iterations must be completed every two to three years, this enormous time cost and adaptation workload become a mountain pressing down on automakers. No one is willing to risk project delays to take on such engineering complexity. Put plainly, this is a calculus of time, and the math is crystal clear.
Forging a Chinese-Made "Hard-Core Pencil"
The opening often comes from a crisis ripping things apart.
The global chip shortage storm that swept through 2020 shattered the once-ironclad trust in the global supply chain. A flood of finished vehicles that couldn't be delivered due to missing chips forced domestic OEMs to seriously evaluate local startups. Panic on the demand side and the localization wave on the supply side converged in the same window of time.
Chenzhi Semiconductor, founded in 2023, landed squarely in this dangerous yet tantalizing gap. Most startups would play it safe and start with simpler MCUs, but Chenzhi went straight for the hardest bone, targeting high-difficulty SoCs. Since domestic automakers are abandoning the inertia of old architectures and trying to seize the right to define future architectures, they inevitably need a chip partner that can match their aggressive iteration pace and understand local demands.
Chenzhi's flagship C1 series chips focus on communication and control fusion, combining high bandwidth and low latency to directly address the pain point of high-real-time cross-domain control. This is no paper talk. From chip bring-up to clearing functional safety (ISO 26262) and cybersecurity (ISO 21434, etc.) certifications, every milestone demands blood, sweat, and tears. The slow progress of localizing such complex chips is also due to the enormous workload invested in software adaptation.
To stamp a seal of trust on this domestic pencil, Chenzhi chose a pragmatic path: leveraging external strength. In December 2025, Chenzhi formally signed a strategic cooperation agreement with Elektrobit, one of the world's top three automotive foundational software giants.
But a chip passing tests in the lab isn't enough. The true test of an automotive-grade SoC lies in whether it can seamlessly mesh with underlying foundational software and deliver stable performance under realistic complex workloads. Elektrobit has 35 years of mass-production adaptation experience in embedded software, with a software stack covering over 600 million vehicles. That means they've seen enough "corner cases" to know where a chip might stumble under extreme scenarios. Running Chenzhi's C1 chip through Elektrobit's software validation framework in advance is like rehearsing all the potential compatibility landmines at the chip definition stage, long before it ever reaches a vehicle. This deep "chip + software" pre-integration is not a simple 1+1. It shifts the adaptation risk—which would normally fall on the OEM or Tier 1 downstream—forward to the early stages of chip development.
Of course, Chenzhi is not alone. Domestic SoC makers like Horizon Robotics, SemiDrive, and Black Sesame Technologies are also accelerating their efforts, each with different entry points and technical focuses.
Winning Endorsement from the Global Top-Tier Ecosystem
But a hard pencil alone still can't produce a good blueprint.
In the context of software-defined vehicles, chip hardware is increasingly becoming a vast and silent foundation. Without top-tier software to orchestrate it with precision, customers can't unlock the chip's true value. Chinese automakers certainly dazzle the world with their innovation at the application layer, but in the invisible ecological accumulation of underlying operating systems and foundational middleware, there remains a clear time gap versus Europe's old-line industrial powers.
Looking at its open-source implementation and kernel adaptation, this combination of "local chip + international software" is far from mere PR posturing. Elektrobit itself is undergoing a deep localization transformation, shifting from simply selling European standard components to China, toward building local architectural capabilities. Demand for AUTOSAR in the domestic market is exploding. In the past, a car only needed it on two or three ECUs; now, virtually all mainstream ECUs have it as standard.
In this complex software-hardware coordination battle, Elektrobit and Chenzhi did something extremely clever: they front-loaded the adaptation work that would normally fall on the OEM. By deeply pre-integrating Chenzhi's SoC with Elektrobit's foundational software, they pushed the synergy between the two to its limits. When customers receive AutoNexKit—the high-performance automotive network development kit jointly launched by Chenzhi and Elektrobit—the underlying communication and safety scheduling mechanisms are already fully wired up. Elektrobit provides one-stop technical support on Chenzhi's underlying chip, dramatically compressing the joint debugging cycle and achieving true plug-and-play.
This precisely hits the automaker's pain point, slashing validation costs and workload. Are there hidden concerns with this model? Of course. The roots of the underlying software ecosystem still lie in Europe. But at least in the current close-quarters battle, this is the optimal solution for domestic chips to clear the trust threshold and seize the window for vehicle integration.
The Architectural Dividend of Surplus Reliability
Once technological momentum builds up enough, spillover is only a matter of time. While sweating over the pearl of the automotive sector, this pencil, tempered by automotive-grade rigor, has already set its sights on embodied intelligence and the low-altitude economy.
Look at it in depth: a car is essentially a robot on four wheels. Looking at the evolution of underlying logic, the robotics industry today is in the same primitive stage the automotive industry was in its early days. Most robots today have a "smart brain," but the motion control of their torso and joints is cobbled together with scattered, low-end MCU chips, with no unified, efficient architecture in place. As embodied intelligence gains more and more joints, the demands for motion planning and millisecond-level real-time communication will skyrocket exponentially.
This is a prime hunting ground for domestic high-end SoCs. The extreme reliability demanded by automotive-grade chips may be overkill for today's robotics industry, but that's precisely the core advantage for seizing the market position during this window.
AutoNexKit, co-launched by Chenzhi and Elektrobit, is essentially a prototype "robot central control system" already validated for automotive mass production. The architecture evolution from distributed ECUs to centralized domain control took the automotive industry twenty years to complete—the robotics industry doesn't need to retrace that path. AutoNexKit's pre-integrated Hypervisor virtualization technology allows a single Chenzhi C1 chip to simultaneously run Linux (handling AI tasks like path planning) and a high-safety, high-real-time OS (handling hard real-time tasks like joint servo control, torque closed-loop, and real-time bus communication), with hardware-level isolation between the two domains ensuring deterministic responses for safety-critical tasks remain undisturbed. The Classic AUTOSAR layer is naturally suited for robot joint motor driver software and CAN/Ethernet bus protocol stacks. This means the real-time control of dozens of joints in a humanoid robot can be centrally orchestrated by the C1, evolving "a swarm of independently operating mini-brains" into "a unified, coordinated central nervous system."
Choosing this precise moment to enter embodied intelligence is fundamentally about staking a claim before the ultimate architecture of the robotics industry solidifies. It's a preemptive declaration of the right to define the underlying logic on another vast blueprint about to be unrolled.
Final Thoughts
When breakthroughs in point technologies start hitting physical ceilings, the true friction point of industrial competition has long shifted to squeezing out systemic efficiency and pushing ecosystem foundations to their limits.
Building an integrated "chip + software" platform has one most practical significance: clearing away the messy underlying complexity so OEMs can pour all their ammunition without reservation into differentiated battles at the application layer. The mighty march of China's automotive industry cannot be built on quicksand where the underlying logic is controlled by others.
From passively accepting black boxes, to drawing their own architecture diagrams, to pulling in world-class international software to endorse domestic SoCs, this road is full of invisible, smokeless battles at the foundation level. By keeping the pencil that draws the blueprints firmly in their own hands, Chinese automakers can avoid being forced to walk away from the table in the ultimate game of full-domain software-defined vehicles.