(This article is authored by Hua Juan Bu Sai Che, and published on Titanium Media with authorization.)
The intelligent driving industry is entering a phase where storytelling is no longer so easy.
A few years ago, a company that could deliver advanced features and showcase a compelling demo was often able to attract attention from automakers and investors. As assisted driving enters mass production, the competitive landscape is shifting: while algorithms remain critical, the ability to deliver solutions on schedule, rapidly adapt to different chips and vehicle models, and keep costs within a reasonable range is becoming the foundational capability for suppliers to secure projects and scale up.
This competition in supply-chain capabilities is playing out directly on the production lines of mainstream automakers. Recently, a factory‑fit mass‑production solution developed by CalmCar based on Horizon’s Journey 6B platform has been seeing intensive vehicle integrations and deliveries. Its intelligent driving system, which provides core forward‑looking perception algorithms, has already achieved mass‑production deployment on the Roewe Rongwei 07, the entire 2027 Buick LaCrosse lineup, and the all‑new Cadillac XT5 PHEV, while also entering large‑scale production phases in related projects at other OEMs.
To date, in the L2–L2+ level solutions segment, TianTong Weishi has secured designation letters for 198 vehicle models across 23 automotive brands, with 105 models from six brands already in mass production and delivered.
(Cadillac XT5 PHEV, Buick LaCrosse 2027 model, and Roewe Jiayue 07 correspond to the luxury SUV, mainstream joint-venture family sedan, and domestic-brand family sedan segments, respectively.)
Compared with the technical performance of a single project, what’s even more worth examining behind these figures is this: as the autonomous driving industry transitions from technology validation to commercialization, how does a company simultaneously manage an increasing number of vehicle models, chip platforms, and delivery milestones?
I. From Algorithm Competitions to Delivery Competitions
“In the first half of the intelligent driving industry, the key was whether you could actually build it; now that we’ve entered mass production, the challenge has shifted to delivering reliably—on hundreds of vehicle models, on automakers’ timelines, and at controllable costs.”
During the discussion, TianTong Weishi CEO Wang Xi remarked: “Getting a single prototype vehicle to run smoothly and ensuring that 100,000 production units installed in vehicles operate without issues are two entirely different capabilities.”
This is also a clear shift currently underway in the intelligent driving industry. In the early stages of R&D, top algorithm teams could continuously refine the user experience around a handful of vehicle models. However, once mass production begins, automakers differ significantly in their electrical/electronic architectures, sensor configurations, chip platforms, and software interfaces. As a result, an algorithm designed for one model must undergo extensive integration, validation, and calibration when deployed on new vehicles.
Cadillac XT5 PHEV
The more pressing challenge comes from time. For automakers, advanced driver-assistance systems are just one link in the overall vehicle development chain; any software delays could directly impact the vehicle’s start of production (SOP).
Wang Xi stated that the industry’s traditional “one project, one code” approach could still be sustained when the number of projects was relatively small; however, as the number of customers and vehicle models grew, so too did the costs associated with software branching, testing and validation, and ongoing maintenance. TianTong Weishi is currently seeking to reduce these costs through platform-based development.
According to official disclosures, on previously partnered chip platforms, the time from deployment to mass-production delivery for new models can be kept under six months; when working with a new chip platform, system deployment and tuning can be completed in about one month. Previously, on the Zeekr project, nearly 1,000 beta and production‑release software versions were rolled out across multiple vehicle models within a single year.
These metrics are difficult to equate simply with technological leadership, but they reflect another aspect of automotive supply-chain capability: the ability to manage multiple projects simultaneously while minimizing the marginal costs associated with new projects.
As advanced driver-assistance systems make their way into mainstream vehicles priced between RMB 100,000 and RMB 200,000, automakers can no longer afford to keep piling on ADAS costs indefinitely. Functionality, pricing, and delivery timelines must all align; accordingly, the industry is shifting from a mere race to push technological boundaries to a more pragmatic engineering‑driven approach.
II. Standardization: Determines Whether Scale Can Be Turned into Profit
For autonomous driving solution providers, the more vehicle models they serve, the easier it is—in theory—to achieve economies of scale, provided that the technology can be reused.
Otherwise, what appears to be software sales actually requires a sustained investment of substantial engineering resources. If each new vehicle model necessitates nearly a complete redevelopment, the more projects there are, the more labor costs will rise—making it difficult to realize the marginal benefits that software companies should enjoy.
Tian Tong Weishi proposes saying goodbye to the “one project, one code” approach, with the core goal being to increase code reuse.
Wang Xi revealed that the reuse rate of his software and AI algorithm models across different projects exceeds 70%; when considering only the toolchains and verification frameworks used for porting across chip platforms, the reuse rate can reach approximately 90%. The remaining work that requires vehicle‑specific handling is primarily focused on customer interfaces, real‑vehicle calibration, and adaptation to certain scenarios.
The automotive industry cannot achieve complete standardization. The real challenge lies in compressing the customizable components into as thin a layer as possible.
To this end, TianTong Weishi has progressively turned its data closed-loop, model training, and system integration processes into tools. Its CalmVolution development platform covers the entire workflow, from data management and model training to automated integration. According to official data, CalmForge can shorten model-training cycles by more than 30%, while its automated integration framework helps reduce redundant development efforts across different projects.
J6B domain controller
The scale effect is also beginning to show in the company’s business structure. From 2023 to 2025, Tian Tong Weishi’s L2–L2+ software licensing revenue increased from RMB 12.1 million to RMB 31.7 million; in 2025, the gross margin for L2–L2+ solutions rose from 32.4% in the previous year to 43.8%. The company explains that as more projects transition from the early development phase into the post‑mass‑production software licensing stage, the same software can continue generating revenue while new costs remain relatively limited.
This may be more important than simply looking at the number of units loaded. Only when the number of fixed locations is converted into mass production and further generates sustained software licensing revenue does the scale truly acquire commercial significance.
III. The Survival Space of Third-Party Autonomous Driving Providers
Another issue facing TianTong Weishi is just how much room remains for independent third-party intelligent driving companies in the future.
On one hand, an increasing number of leading automakers are strengthening their in-house development of intelligent driving systems; on the other hand, major third-party solution providers continue to expand their customer base. Suppliers positioned in the middle must once again demonstrate their value.
TianTong Weishi adopts a cross-platform, vendor-neutral approach. Wang Xi believes that automakers still need a multi‑vendor system to diversify supply risks and avoid being locked into a single chip or technology stack. “What customers want is the ability to switch platforms without having to change suppliers.”
As of now, TianTong Weishi’s customer base has grown from 222 at the end of 2023 to 330, while the revenue contribution from its largest customer has decreased from 62.1% in 2023 to 10.4% in 2025, and the combined revenue share of its top five customers has fallen to 35.3%. This reduction in customer concentration helps mitigate reliance on any single project.
However, this expansion has also brought additional pressure. From 2023 to 2025, the company’s trade receivables increased from RMB 88.7 million to RMB 547.8 million, with an average turnover period of approximately 300 days in 2025. According to the official explanation, the lengthening of payment terms is largely attributable to L4‑level business, while a portion of revenue is recognized predominantly in the fourth quarter, resulting in a relatively high year‑end receivables balance. The company expects operating cash flow to turn positive in 2026, though this remains subject to validation by subsequent operational data.
For autonomous-driving suppliers, scale doesn’t just mean securing more project awards and vehicle integrations. As the number of projects grows, accounts receivable, R&D spending, and capital tied up in operations may also increase accordingly. To assess a company’s operational quality going forward, it’s not enough to look solely at installed‑vehicle volumes; you must also evaluate software licensing revenue, gross margin, customer mix, and whether cash flow improves as scale expands.
IV. L4 and Going Global: Testing Capabilities for the Next Phase
Beyond L2–L2+ in China, TianTong Weishi is positioning its next phase of growth in L4 and overseas markets.
According to Wang Xi, L2–L2+ and L4 are not mutually exclusive. L2–L2+ delivers large-scale mass production, engineering expertise, and ongoing software licensing revenue, while L4 opens up new business scenarios and data sources.
In 2025, TianTong Weishi’s L4 solution generated revenue of RMB 374.5 million, accounting for 68.4% of its total annual revenue. To date, the company has disclosed L4 pre-orders covering more than 2,500 Robotaxis, Robobuses, and Robotrucks, with a combined contract value exceeding RMB 1 billion, scheduled for phased delivery over the next 3 to 5 years.
In overseas markets, the real test lies in the engineering capabilities accumulated over time. The company has currently secured production contracts for 87 overseas vehicle models, with 59 of them already in mass production abroad. During discussions, Wang Xi noted that once Chinese intelligent driving companies go global, algorithms tend to be the relatively easy component to replicate; the greater challenges lie in regulatory certification, local engineering, customer trust, and the delivery system.
This is also a challenge faced by Chinese intelligent driving suppliers. Overseas markets are not simply a replication of domestic solutions; differing regulatory frameworks, road conditions, and driving habits in each country all require re-adaptation.
Back in China, the intelligent driving industry is also entering a new round of selective consolidation. Technologies such as large-scale models, end-to-end systems, and VLA continue to push the industry’s technological boundaries, but competition among suppliers increasingly hinges on their ability to achieve real-world commercialization. Striking a sustainable business model that balances hardware cost control with software performance, while ensuring seamless project delivery, scalable operations, and high‑quality revenue, is becoming the key determinant of corporate competitiveness.
Tian Tong Weishi’s mass production and delivery of 105 vehicle models can be seen as a case in point of this shift. Going forward, the key question will be whether these engineering capabilities can be further translated into stable economies of scale. As software licensing revenue from L2–L2+ level solutions continues to grow and L4 operations enter the commercial deployment phase, the company will still need to demonstrate its ability to improve profitability and cash flow.
Meanwhile, in overseas markets, whether it can break free from a business model that relies heavily on Chinese automakers going global and truly gain entry into the supply chains of mainstream global automakers will also serve as a key indicator of its internationalization capabilities. (This article was first published on the TMTPost app; author: Hua Juan.)
For more exciting content, follow TMTPost on WeChat (ID: taimeiti), or download the TMTPost App.