XPeng Robotaxi Named "YOYO": 2,000 Rides at 1 Yuan Each — What's Still Missing Before It Turns a Profit?

The XPeng Robotaxi fleet conducts road tests in Guangzhou, moving toward open commercial operation for users.

On October 8, XPeng officially gave its Robotaxi a Chinese name — "XPeng YOYO." That same week, its internal test rides in Guangzhou had just passed 2,000 orders, each charged at 1 yuan.

2,000 orders, 1 yuan each — those two numbers sit right there, but wedged between them is a problem every Robotaxi company has yet to solve: whether the cost per ride can actually be pushed below the 1 yuan the passenger pays.

An XPeng YOYO Robotaxi driving on the road (no safety driver in the main seat).

But one misconception needs clearing up: this is not a test vehicle hastily converted from an existing car.

It is China's first Robotaxi developed from scratch with autonomous driving capability built in at the factory, based on the flagship GX model. The car is equipped with 4 of XPeng's self-developed Turing AI chips, delivering 3,000 TOPS of onboard compute — which XPeng says is the highest of any car in the world today. It runs a second-generation VLA large model, relying neither on high-precision maps nor lidar.

The cost savings are hidden in those words "relying neither on." A single lidar unit costs several hundred to over a thousand US dollars, and high-precision maps require continuous spending on updates and maintenance. Removing these cuts away a large chunk of the "hardware is too expensive" criticism that Robotaxis most often face. Combined with the fact that it is factory-installed and mass-produced on the same production line as regular cars for sale, the per-unit cost is closer to a car you could normally buy rather than a purpose-built device. Not relying on high-precision maps has another practical benefit: when entering a new city in the future, there is no need to first spend large amounts of time and money mapping the local area — the model can find its own way, making the marginal cost of expansion much lower.

But being cheap does not mean being profitable. What XPeng is really betting on is a different path: using the scale and cash flow accumulated from selling cars to amortize the investment in Robotaxi. If the bet pays off, it will see its books turn positive earlier than companies that focus purely on Robotaxi. If it doesn't, for XPeng it is just another expense to burn through on top of its main business.

XPeng's Turing AI chip and physical-world large model support the Robotaxi's 3,000 TOPS onboard compute.

So what do those 2,000 orders actually prove

In August of this year, XPeng completed over 2,000 internal test orders in Guangzhou, running through the entire chain from "order on phone — car picks up passenger autonomously — drop off at destination." It also obtained Guangzhou's remote testing qualification, allowing it to conduct tests without a safety driver in the main seat on designated roads. The 1 yuan charged per internal test ride was never about making money — it was about getting the payment and service processes running smoothly first. Once the mini-program launched, the pool of people who could try it expanded from company employees to external users.

2,000-plus orders — not a lot, but not a little either. The truly hard part is crossing from "the car can drive itself" to "it can carry ordinary passengers," and XPeng has just climbed over the toughest stretch. It does not prove profitability; it only proves that this paid system can actually run — there is still a long way to go before "revenue per ride covers cost."

View from inside the test vehicle: the full experience from mini-program order to autonomous pickup.

XPeng is taking the road in reverse

Other companies doing Robotaxi typically spend money first to get autonomous driving working, then apply the technology in a downgraded form to mass-produced cars. XPeng does the opposite: it first relies on selling L2 assisted-driving cars to support the team and accumulate data, then uses Robotaxi to validate L4.

This approach has a practical advantage: it does not have to wait for L4 to mature before starting to recoup costs. L2 cars keep selling, data keeps flowing, and Robotaxi is more like the ultimate exam at the top of this data chain rather than an isolated money-burning project. He Xiaopeng has also put the investment on the table: the smart-driving team will expand to over 3,500 people, with annual R&D spending of no less than 4.5 billion yuan. As for how long it will take for that money to pay back on Robotaxi, he did not say.

But that is only the visible number — there is another layer of logic underneath.

XPeng as a whole is still losing money. It posted a net loss of 1.34 billion yuan in the second quarter, with R&D spending up about 32% year-over-year, while money is also being poured into robotics and overseas expansion. Robotaxi is just one piece of it, and for now it is still a money spender, not a money maker.

Looking further ahead, He Xiaopeng himself estimates that L4 will truly land in policy terms between 2028 and 2032, and it will vary by country and by scenario. Removing the safety driver from the main seat does not mean no one is involved throughout; a car being able to drive and a car being permitted to charge ordinary passengers for commercial operation are two different things.