The autonomous driving market in 2026 appears to be booming. The list of third-party solution providers is growing longer by the day, including Huawei Qiankun, Momenta, Horizon Robotics, Zhuoyu, and DeepRoute.ai, while automakers' self-developed solutions are also emerging endlessly.
Urban NOA (Navigation on Autopilot) has rapidly trickled down from high-end models to vehicles priced at over CNY 100,000 or even lower, with penetration rates steadily climbing.
However, behind this prosperity lies a paradox: as solutions proliferate, suppliers capable of achieving distinct differentiation are becoming fewer.
Once standardized solutions leveled the threshold for urban NOA usability, many offerings became increasingly similar in basic capabilities. The perceived differences among users have narrowed, leading automakers to prioritize utility, cost-effectiveness, reliability, and ease of implementation during selection. This is not technological stagnation but an inevitable phenomenon after technology advances to a certain stage.
In the second half of the autonomous driving race, the focus has shifted from "can it be done" to "is it stable, cheap, compliant, and scalable."
01 Homogenization Pressure Beneath the Prosperity
In the past few years, autonomous driving was still in the phase of showing off muscle.
Who opened cities first, whose takeover mileage was longer, whose end-to-end model had more parameters, or who stacked more LiDARs could all serve as temporary advantages.
Suppliers could tell distinctly different stories through technical routes, sensor configurations, and algorithm architectures.
By 2026, everything changed.
End-to-end architectures, BEV perception, Transformer/large model routes, and world models have transitioned from core technologies of a few players to mainstream standards. The gap between top-tier solutions in complex urban scenarios and long-tail handling is narrowing rapidly.
The experiential difference among the first tier has shifted from "whether it works" to subtle variations in "how well it works."
Moreover, national standards such as the "Safety Requirements for Combined Driving Assistance Systems of Intelligent Connected Vehicles" (GB 47955—2026), scheduled for implementation in 2027, impose clear requirements on functional safety, test coverage, OTA control, and DMS monitoring. Safety and compliance have become hard constraints.
Suppliers must invest significant resources into verifiable, certifiable, and traceable systems rather than engaging in unconstrained innovation.
The highly unified safety baseline naturally compresses the space for aggressive yet difficult-to-explain differentiation.
Driven by these factors, many solutions are becoming increasingly similar in core capability packages, featuring comparable sensor combinations, close computing power ranges, and converging algorithm frameworks.
What automakers purchase looks more and more like standardized autonomous driving modules rather than unique souls.
02 From Diversity to Convergence
The evolution of autonomous driving technology exhibits clear phased characteristics.
In the early stages, the debate centered on perception routes, discussing whether to choose pure vision or multi-sensor fusion, and whether to install LiDARs and how many. In the middle stage, the focus shifted to architecture debates, with greater interest in modular architectures versus end-to-end routes. Now, the emphasis in the autonomous driving field has moved toward finer details such as compliance and optimization.
Although end-to-end models are powerful, pure black boxes still pose challenges regarding safety explainability and liability delineation. Consequently, hybrid approaches like "end-to-end + rule-based fallback" and "large model + traditional modules" have become common compromises.
On the chip side, mid-to-high computing platforms (tens to hundreds of TOPS) can already cover mainstream L2+ to urban NOA needs, making cost-performance ratio the key factor.
LiDAR prices have dropped significantly. Short-range blind-spot fillers and some mass-production models have reached the USD 200 level, while the average price of forward main radars has fallen to the hundreds-of-dollars range. LiDAR is no longer a symbol of high-end status but an optional configuration.
When underlying technology toolkits converge, upper-layer differentiation becomes difficult to achieve.
When suppliers promote that they use the latest large models or have stronger perception, automakers and users often react with, "Others are similar too. What about the price? Production experience? After-sales support?"
Data closed loops further accelerate this convergence.
Top players accumulate real-world road data through large-scale mass production, enabling rapid model iteration. If small and medium-sized players lack sufficient data, they rely on public datasets or simulations, easily falling into the dilemma where solutions appear advanced but deliver homogenous or even inferior experiences upon deployment.
Once a data flywheel is formed, the cost for latecomers to catch up rises exponentially.
03 Automaker Self-Development and Supply Chain Restructuring: Are Suppliers Caught in the Middle?
A pressure more direct than technological convergence comes from the client side.
Leading automakers (BYD, XPeng, Li Auto, NIO, Huawei-affiliated brands, etc.) view high-level autonomous driving as a core competitive advantage, investing massive resources in full-stack or core self-development.
The logic behind self-development is clear: master data, define experience, form brand differentiation, and reduce long-term dependence on external parties. As a result, a significant portion of the high-value, high-premium market for advanced solutions has been captured by automakers' self-developed efforts.
The remaining market mainly consists of mid-tier automakers, joint-venture brands, and players needing to quickly address their intelligence shortcomings.
These clients have very realistic demands: costs must be low enough to support the democratization of autonomous driving and price wars; delivery must be fast without delaying vehicle programs; safety and compliance must be robust to avoid becoming risk sources; and ideally, solutions should be cross-platform and cross-chip compatible to prevent vendor lock-in.
Thus, suppliers find themselves in a dilemma.
To meet these requirements, solutions must become more standardized, platformized, and easier to integrate, which ironically weakens differentiation. Pursuing deep customization and unique experiences requires enormous R&D investment, long cycles, and high risks, which small and medium-sized clients may not accept.
Market concentration is also rising. Top players like Huawei, Momenta, Horizon Robotics, Afari Technology, and Zhuoyu have occupied the majority of the third-party high-level autonomous driving share, forming a clear head effect. Small and medium-sized suppliers are either squeezed out or barely survive in niche scenarios, low-level solutions, or specific regions.
With price wars ensuing, profits are continuously compressed.
04 Triple Squeeze: Cost, Safety, and Standards
Why has differentiation become harder? Three forces are at play.
Cost is the most direct impact.
As autonomous driving shifts from a selling point to a standard feature, automakers are unwilling to pay excessive premiums for common capabilities. Suppliers must aggressively cut costs while maintaining performance, focusing hardware selection, algorithm optimization, and supply chain management around cost efficiency.
Under cost pressure, suppliers prioritize core capabilities that allow for scalable reuse, sacrificing non-essential differentiated designs for platformization and integration efficiency.
When the same batch of chips, sensors, and algorithm frameworks are repeatedly combined by multiple solution providers, natural convergence occurs.
Safety and liability tighten the other end.
Once assisted driving or higher-level functions are involved, any aggressive strategy poses huge risks. Suppliers are forced to choose safer paths between aggression and conservatism, leading to mediocre user experience styles.
Standards and testing squeeze the remaining space.
National standards, track tests, and data compliance requirements are becoming increasingly strict, with complete safety closed-loop capabilities becoming entry barriers. Pure algorithm or pure hardware players lacking systematic capabilities are marginalized. Remaining competitors operate under the same set of rules, leaving less room for unconventional differentiation.
With these three forces overlapping, autonomous driving has shifted from a technology race to a system competition.
Simple algorithm highlights or sensor stacking can no longer constitute lasting moats.
05 Data and Scale Become New Moats but Also Exacerbate Homogeneity
What truly remains capable of widening the gap is data and scale.
Whoever has larger installed volumes, more real-world mileage, and broader scenario coverage iterates models faster and solves long-tail problems better.
Players like Momenta accumulate million-vehicle data through extensive design wins; Huawei forms user mindshare through its ecosystem and brand; Horizon Robotics lowers deployment thresholds through integrated chip-plus-algorithm solutions.
These are all advantages derived from scale.
However, scale advantages themselves drive homogenization.
To cover more automakers, price points, and vehicle models, top suppliers must turn solutions into highly platformized, replicable products.
The better the platformization, the smaller the space for differentiation.
Even if automakers adopt the same supplier, it is difficult to create fundamentally different experiences at the underlying level. They can only make limited adjustments in tuning, interaction, and scenario preferences.
Data closed loops bring another issue: training data sources are becoming increasingly homogeneous, consisting mostly of public data, similar road conditions, and analogous annotations. Consequently, model outputs naturally converge.
06 How Can Suppliers Still Achieve Differentiation?
It is undeniable that technological convergence is a reality, but it is not unsolvable.
Differentiation in the second half of the autonomous driving era has shifted from technical parameters to systems and ecological niches.
Some differentiate within scenarios.
Some specialize in complex urban streets, others polish stability for highway long-distance travel to the extreme, and some focus on specific climates, road conditions, and user habits. Small but deep scenario barriers are often harder to replicate than broad, general-purpose solutions.
Others differentiate through cost and efficiency.
If high-level capabilities can be achieved with lower hardware costs, lower computing consumption, and faster adaptation cycles, this itself is competitiveness. Especially in the mid-to-low price segment, cost-performance ratio is the biggest differentiator.
Software-hardware integration and platformization represent another path.
If chips and algorithms collaborate deeply, possess cross-chip and cross-platform adaptation capabilities, and come with complete toolchains and development support, automakers' deployment thresholds and risks decrease significantly, fostering stickiness. Horizon Robotics generally follows this direction.
Data and iteration speed remain effective.
Truly turning installed scale into a data flywheel and maintaining an iteration pace ahead of others results in users perceiving continuous improvement. This is what sustains the gap.
Furthermore, collaboration relationships can evolve from delivery to co-creation. Solutions need not be closed black boxes; opening up some capabilities allows automakers to participate in tuning and scenario definition, helping them build "my autonomous driving" rather than "the supplier's autonomous driving." Simultaneously, expanding territory to overseas markets, commercial vehicles, and Robotaxis widens the value space.
Taking a step further, suppliers can move from selling solutions to selling services.
If capabilities such as simulation, testing, data management, OTA operations, and safety certification support are delivered alongside the solution, cooperation transforms from a one-time design win into a long-term partnership.
In the autonomous driving field, those who ultimately survive and succeed in differentiation are mostly players who clearly understand their position and strive for excellence within it.
Some willingly act as electronic screws, some serve as neutral platforms, some become experts in specific scenarios, and some act as cost killers.
Those attempting to dominate all markets are most likely to fall into homogenization.
07 Differentiation Has Not Disappeared, It Has Just Changed Battlefields
The fact that autonomous driving solutions are proliferating while suppliers struggle to differentiate does not signal industry failure but rather maturity.
When a technology transitions from a scarce capability to a basic one, differentiation naturally shifts from the technology itself to cost efficiency, system reliability, data scale, ecosystem synergy, and scenario understanding.
Similar to the smartphone industry, where Android and Qualcomm combinations are highly homogeneous, brands still distinguish experiences through imaging tuning, system interaction, and ecosystem services.
Automotive autonomous driving will follow a similar path.
For suppliers, the true test is not whether they can produce a solution that looks different, but whether they can find and hold their position amidst comprehensive competitions in cost, safety, scale, and service.
The second half has begun. Flashy solution launch events will become fewer, while solid engineering, scalability, and systematic capabilities will become increasingly valuable.
Those who survive the wave of homogenization and achieve true differentiation will be the ones left standing.