XPeng has created a standalone robotaxi business unit, according to local media reports on March 23, turning what could have been another autonomous-driving update into a more consequential organizational move. The Chinese EV maker is not only talking about Level 4 autonomy in principle; it is assigning tier-one status to a unit that will handle product definition, integration, testing and operations, while pointing to passenger-carrying demonstration runs in the second half of 2026. That combination of structure and timing suggests China’s smart-EV competition is starting to shift from ADAS feature races toward the harder work of building commercial robotaxi operations.
This is an organizational upgrade, not just another autonomy announcement
The most important fact in the March 23 reports is not simply that XPeng wants to do robotaxis. Local outlets including CLS and Jiemian said the company formally established a Robotaxi business department on March 23 and gave it tier-one organizational status inside the company. CnEVPost, citing a person familiar with the matter, said the new unit will coordinate product definition, project integration, R&D testing and day-to-day operations. That matters because those are not the functions of a speculative research project. They are the functions of a business being prepared for deployment.
The timing also fits a broader internal restructuring already under way. CnEVPost noted that XPeng merged its autonomous-driving center and smart-cockpit center in February to form a new general intelligence center. Jiemian framed the March 23 robotaxi move as the application-layer follow-through to that February reorganization: first build a more unified technical base, then create a dedicated operating unit to push commercialization faster. In other words, XPeng is not presenting robotaxis as a side experiment. It is building a structure that connects core AI development to an eventual service model.
The timeline is what gives the move its commercial weight
XPeng’s March 20 fourth-quarter 2025 earnings call had already given the market a concrete timetable. During that call, chairman and CEO He Xiaopeng said the company planned to begin passenger-carrying demonstration operations for robotaxis in the second half of 2026 and hoped to move toward operations without safety operators in 2027. Those two dates are what make the March 23 organizational change more important than a routine China EV autonomy headline. A new department matters much more when it arrives with a near-term operational schedule rather than a vague multi-year ambition.
The same reporting cycle also added details that make the roadmap easier to read. CLS and Jiemian both said XPeng plans to launch three fully self-developed mass-production robotaxi models in 2026, while gradually opening its Robotaxi SDK to partners and building an ecosystem with Amap, the mapping and mobility platform owned by Alibaba, identified as its first global ecosystem partner. That combination is notable because it suggests XPeng does not want to stop at building a self-driving vehicle stack. It also wants to shape interfaces, distribution and operations around that stack.
There is still an important limit to what has been disclosed. Public reporting on March 23 did not lay out a detailed city-by-city rollout plan, expected fleet size, or the regulatory sequence required for scaled commercial service. So the right way to read the second-half 2026 milestone is as a demonstration-operations target, not as proof that XPeng is about to run a large fully commercial robotaxi network. The company has provided a timeline and an internal structure, but not yet the full operating blueprint.
XPeng is trying to attack L4 from the OEM side, not the startup side
That is where XPeng’s profile becomes interesting. Unlike a pure-play robotaxi startup, XPeng already has a mass-production EV business, a growing in-house chip story, a national retail footprint and an existing intelligent-driving product pipeline. On the March 20 earnings call, the company said cumulative shipments of its Turing AI chips had surpassed 200,000 units, and it expected its broader lineup to transition toward its in-house chips starting in the second quarter of 2026. Those facts do not prove robotaxi success, but they do show that XPeng is trying to build autonomous driving on top of vehicle-scale manufacturing and compute deployment rather than outside of them.
Jiemian’s March 23 report added another reason the story travels well in English: the technical framing is specific. The outlet said XPeng’s second-generation Vision-Language-Action, or VLA 2.0, model has become a core support layer for the robotaxi push, shifting from a “vision-language-action” chain toward a more direct “vision-to-action” architecture. Jiemian also said XPeng claims response latency below 80 milliseconds and a 12-fold improvement in inference efficiency. Even allowing for company framing, those numbers matter because they show XPeng wants investors and readers to view robotaxis not only as a vehicle product, but as the next application of its broader AI stack.
XPeng’s GX large six-seat SUV adds a real-world testing anchor to that narrative. CnEVPost reported in February that the GX began normalized L4 autonomous-driving road tests, and the March 23 reports said the model had already obtained Guangzhou road-testing permission and was continuing public-road testing. The GX is expected to launch in the second quarter of 2026. That does not mean XPeng has already solved robotaxi operations, but it does mean the company is trying to connect its L4 ambition to a real vehicle program rather than to a presentation deck.
The bigger signal is that China’s EV race is being reorganized around operations
For the past several years, China’s intelligent-driving race has often been framed around assisted-driving features: urban NOA rollouts, hardware cost reductions, lidar counts, map capability and how many consumer vehicles could offer higher-end ADAS. XPeng’s March 23 move points to a different phase. Once a company creates a unit whose responsibilities explicitly include operations, the competitive question starts to change. It becomes less about who can advertise the flashiest autonomy demo and more about who can organize vehicles, software, partners, compliance and daily fleet execution into a repeatable service business.
That shift matters because XPeng is an OEM, not a company that began life as a robotaxi operator. If a Chinese EV maker with its own vehicle programs and supply-chain scale is willing to create a dedicated robotaxi department now, the market signal is that commercial Level 4 operations are becoming important enough to deserve their own P&L logic, management chain and ecosystem strategy. In that sense, the March 23 announcement is not only about XPeng. It is also about how China’s leading smart-EV companies may be starting to reorganize for the next battleground after assisted driving.
At the same time, this is still a story that requires discipline in wording. The most detailed March 23 information came through local media reports citing company materials or people familiar with the matter, rather than through a standalone global press release from XPeng. The 2027 goal for operation without safety operators should be treated as a target, not an accomplished fact. And because rollout cities, fleet scale and approval details are still limited in public reporting, it would be premature to describe XPeng as already running a scaled robotaxi business.
Even with those caveats, the signal is strong. On March 23, XPeng did more than say it believed in robotaxis. It elevated robotaxis into a standalone business structure, linked that structure to a second-half 2026 passenger-carrying demonstration plan, and tied the whole effort to a broader VLA- and chip-based AI architecture. That is why the story matters beyond one company update: China’s smart-EV race is no longer only about making assisted driving better inside private cars. It is increasingly about which companies can turn that stack into commercial autonomous operations.
Related coverage on 1M Reviews
- XPeng Posts First Quarterly Profit and Heads to Mexico as China’s EV Startup Race Turns Into a Margin-and-Export Story
- Xpeng G6 Extended-Range SUV Brings VLA to Mass Market
- Nvidia pulls BYD and Geely into DRIVE Hyperion as it maps a Level 4 robotaxi timeline
Sources
-
CLS (Independent financial media) — Exclusive: XPeng sets up an independent business unit to accelerate Robotaxi rollout as the industry enters a race for scaled deployment (2026-03-23)
– https://www.cls.cn/detail/2321076
– Key takeaways: Local media first reported XPeng’s March 23 internal announcement creating a standalone Robotaxi unit, describing its responsibilities, tier-one status, 2026 demonstration timeline, and ecosystem plans. -
CnEVPost (Independent English EV media) — Xpeng sets up robotaxi unit to accelerate commercial rollout (2026-03-23)
– https://cnevpost.com/2026/03/23/xpeng-sets-up-robotaxi-unit-accelerate-commercial-rollout/
– Key takeaways: Confirmed the new unit’s operating scope, tied the move to XPeng’s second-half 2026 passenger-carrying demo operations, and reiterated the 2027 target for operation without safety operators. -
Jiemian (Independent Chinese business media) — XPeng establishes an independent Robotaxi business unit and will start passenger-carrying demonstration operations in the second half of the year (2026-03-23)
– https://www.jiemian.com/article/14149761.html
– Key takeaways: Added VLA 2.0 technical details, Guangzhou road-testing permission for the GX Robotaxi vehicle, three 2026 robotaxi models, and the SDK/ecosystem angle. -
CnEVPost (Background source) — Xpeng Q4 2025 earnings call: Live updates (2026-03-20)
– https://cnevpost.com/2026/03/20/xpeng-q4-2025-earnings-call-live-updates/
– Key takeaways: Provided the prior management timeline for second-half 2026 passenger-carrying demo operations, the 2027 no-safety-operator target, and supporting context on Turing AI chips and XPeng’s broader physical-AI push.
Editorial caveats: The strongest March 23 details were reported by media outlets citing company materials or people familiar with the matter, rather than a standalone English corporate press release reviewed in this workflow. Treat second-half 2026 passenger-carrying operations as a demonstration target, not full commercial scale. Treat the 2027 no-safety-operator milestone as a company goal, not an achieved outcome.
