Gasgoo Munich- The Robotaxi race is heating up fast in 2026.
An overview of the market reveals that over the past period, everyone from OEMs like Geely and XPENG to core tech providers like Yihang Intelligence, SenseAuto, Lenovo, and Desay SV has been rushing into the Robotaxi sector. Even leaders in mass-production ADAS, such as Momenta and DeepRoute.ai, are actively pivoting toward the track.
Yet, amidst the buzz, a critical question remains: Why have so many companies chosen this year to place their heaviest bets on Robotaxi?
New and Old Players Double Down
Recently, the first mass-produced XPENG Robotaxi rolled off the assembly line in Guangzhou. Notably, this marks the first time a Chinese OEM has completed the mass production of a Robotaxi using a full-stack, in-house approach.
An executive from XPENG's Robotaxi division revealed that the vehicle is expected to begin pilot operations in the second half of this year.
Built on XPENG's flagship GX model, this full-stack, in-house Robotaxi is equipped with four self-developed Turing AI chips, delivering 3,000 TOPS of computing power. It also features XPENG's second-generation VLA model and boasts Level 4 autonomous driving capabilities.
Image Source: XPENG
XPENG established its Robotaxi division in March to accelerate commercialization. The launch of this pre-installed mass-production model marks a critical step for the automaker, moving from technical verification to actual product deployment.
The XPENG Robotaxi is the second prototype unveiled by a Chinese automaker recently. At the Auto China 2026, Geely debuted Eva Cab, a natively developed Robotaxi prototype, with plans to launch a custom version for Cao Cao Mobility in 2027.
According to released information, Eva Cab will feature a "Quantum-level AI EEA 4.0" architecture, integrating Nvidia's SuperChip and ThorU, along with a Qualcomm Snapdragon 8797. Total computing power also exceeds 3,000 TOPS. Additionally, the vehicle is equipped with the world's first "2,160-line digital LiDAR," with a maximum detection range of 600 meters.
Beyond the upgraded specs, Eva Cab removes the steering wheel, accelerator, and brake pedals—fully embodying a "born for Robotaxi" design philosophy.
Historically, Robotaxi development followed an "additive" approach: retrofitting mass-produced vehicles with autonomous systems. Natively developed Robotaxis, however, take a "subtractive" approach. By innovating the vehicle architecture at the ground level, autonomous kits are treated as core systems rather than afterthoughts. This allows for the removal of redundant components meant for human drivers, cutting costs while boosting performance.
But native design comes at a price. A significant challenge is the massive upfront capital required to build dedicated platforms and production lines, creating a high cost barrier. If a project falls short of expectations, the sunk costs for these vehicles far exceed those of retrofitted models.
Beyond the high-profile entry of automakers, a wave of companies previously quiet in this sector—including Yihang Intelligence, Lenovo, SenseAuto, Desay SV, and ECARX—are making their move.
Yihang Intelligence showcased its first Robotaxi prototype at the Auto China 2026. Based on a mass-produced model equipped with Yihang's ADAS system, the prototype achieves higher-level functions through technical upgrades and enhanced hardware and software redundancy. Previously focused on mass-market ADAS, Yihang's debut signals its official entry into the Robotaxi race.
Chen Yuhang, founder and CEO of Yihang Intelligence, believes that Robotaxis have reached a tipping point in both technical maturity and business viability. That's why the company chose Robotaxis as a key landing spot for AI in the physical world.
Image Source: WeRide
Lenovo and SenseAuto have chosen to partner with WeRide and T3 Go, respectively.
While these companies have different backgrounds and resources, their varied entry strategies point to a shared consensus: across the Robotaxi supply chain, players believe the market is ready for a bet.
Even more intriguing are the moves by Momenta and DeepRoute.ai.
These aren't newcomers to the field. Momenta partnered with SAIC Mobility in 2021 to launch China's first L4 operation platform backed by an automaker. That same year, DeepRoute.ai launched a pilot Robotaxi service in Shenzhen.
In the years that followed, as the L4 sector cooled, both companies shifted their focus to mass-production ADAS, supplying L2 and L3 solutions to automakers, accumulating data, and waiting for the right moment.
Image Source: Desay SV
Now, in 2026, they are converging on the Robotaxi track once again.
It's worth noting that both Momenta and DeepRoute.ai are now front-runners in the mass-production ADAS space. Their collective return to Robotaxis sends a clear signal: if L4 prospects were still uncertain, companies with mass-production projects and cash flow wouldn't easily shift their focus back. Their return is, in itself, a vote of confidence in the industry's maturity.
Of course, returning comes with challenges. These intelligent driving Tier 1 suppliers must deliver ADAS solutions to automakers while deploying Robotaxis. The resources, talent, and organizational capabilities required for these two tasks are fundamentally different.
A dual-track strategy offers stability, but it risks preventing either track from reaching its full potential.
Why 2026?
Because years of refinement have brought Robotaxi technology, cost curves, business validation, and policy environments to a tipping point—shifting from "quantitative change" to "qualitative change."
On the technology front, the most significant shift is that the "progressive" route from L2 to L4 has proven viable.
Early on, the "leapfrog" approach—aiming straight for L4 and Robotaxis—was dominant. But it proved technically difficult and slow to deploy.
Today, intelligent driving Tier 1 suppliers like Yihang Intelligence, QCraft, and Momenta have forged a more pragmatic path. They first mass-produce L2 driver-assistance systems to gather massive amounts of real-world road data before "scaling up" to L4. This dual-track strategy offers a more efficient solution to the "long-tail problems" that once plagued the industry.
Chen Yuhang, founder and CEO of Yihang Intelligence, points out that after a decade of development, mass-market ADAS technology has converged—moving from basic L2 to highway NOA, then city NOA, and now end-to-end driving. While nationwide driverless operation requires long-term exploration due to extreme complexity, Robotaxis focus on localized operations in specific cities. This allows existing autonomous driving capabilities to form a commercial closed loop in a controlled area, making technical deployment feasible.
Yu Qian, CEO of QCraft, echoes this logic. The company's mass-production ADAS and L4 businesses share the same underlying technical architecture. "Whether L2 or L4, they are essentially developed based on the same technical route," Yu Qian explains. "The difference lies only in product form: L4 has its specific product logic, and L2 has its own scenario-adapted logic."
Image Source: Momenta
Momenta's underlying logic relies on a unified autonomous driving model to achieve universal application across scenarios. This reduces multi-scenario R&D costs while allowing data from various fields to feed back into the model, creating a platform advantage.
It is for this reason that Momenta CEO Cao Xudong notes that the company's L4 efforts won't stop at Robotaxis; they will also include Robovans, with Robotrucks coming next year.
On the business front, the math for Robotaxis is finally starting to work.
For years, the biggest hurdle to scaling Robotaxis wasn't technology—it was cost. A retrofitted vehicle could cost over 1 million yuan, making large-scale deployment a distant dream with no short-term return.
But now, that cost curve is bending downward.
"From the perspective of intelligent driving costs, the widespread adoption of city smart driving functions in mass-produced vehicles has significantly lowered the cost of core sensors and hardware like LiDAR and computing power," says Chen Yuhang. "This makes the cost structure of Robotaxi operations preliminarily profitable."
Consider Pony.ai's seventh-generation autonomous driving kit: its BOM cost dropped 70% from the previous generation. Consequently, the company's 2027 Robotaxi model based on this system will see total vehicle costs fall below 230,000 yuan—cheaper than a base-model Tesla Model 3. The hardware cost for WeRide's Robotaxi GXR is currently around $40,000 and is expected to drop another 15%. Baidu Apollo Go's sixth-generation vehicle has seen unit costs drop to just over 200,000 yuan.
How were these costs cut? Domestic supply chains are the key driver.
Consider LiDAR: a few years ago, a single unit cost over 10,000 yuan; now, it's in the thousands. For instance, RoboSense's ADAS LiDAR averaged about 2,600 yuan in 2024 but has fallen to 1,800 yuan in 2025. Prices for RoboSense's M platform and Hesai's ATX series have already dipped into the $200 range.
Beyond supply chain savings, Pony.ai CEO Peng Jun attributes the sustained cost decline to economies of scale, architectural innovations in vehicle design, and close collaboration with OEMs.
Image Source: Pony.ai
As costs fall, revenue is showing the first glimmers of profit.
In 2025, revenue from Pony.ai and WeRide's Robotaxi businesses reached 115 million yuan and 148 million yuan, respectively—year-on-year increases of 128.6% and 209.6%. More tellingly, Pony.ai has achieved positive unit economics in Guangzhou and Shenzhen, while WeRide reached break-even on unit economics in Abu Dhabi. Baidu Apollo Go has also turned profitable in several core Chinese cities.
Of course, profitability in individual cities doesn't guarantee overall profit, but it proves one thing: Robotaxi commercialization isn't a fantasy. Over a decade of technological investment is finally bearing tangible commercial fruit.
The next challenge is replicating this experience in broader markets quickly.
Notably, breakthroughs in policy are creating the conditions for this expansion.
Multiple cities in China have now approved fully unmanned commercial operations. The significance of these moves lies not just in "permission," but in providing companies with a predictable compliance path.
Overall, Chen Yuhang believes that falling costs, the industry's entry into the "embodied intelligence" phase, and the ability to form closed-loop technology in local urban scenarios—combined with domestic and global commercial trends—mean Robotaxis have arrived at their best moment for deployment.
The Starting Gun Has Fired, But Don't Celebrate Yet
In 2026, Robotaxis stand at a historic juncture, shifting from "storytelling" to "calculating the economics."
Recently, Ni Licheng, CEO of SAIC Mobility, stated that while Robotaxi deployment scenarios have existed for years, true large-scale scaling hasn't arrived yet—"though that day is coming soon." His prediction: the next year or two will mark the starting point for large-scale Robotaxi development.
Jin Jun, PwC China's Automotive Sector Leader, also believes that as numerous autonomous taxi companies launch pilots in China and the U.S., more vehicles and fleets will see broader testing and promotion from the second half of this year through the first half of next.
Goldman Sachs predicts that China's Robotaxi market will accelerate, driven by lower costs, expanded fleet sizes and service coverage, software improvements, and higher public acceptance. The bank expects China's Robotaxi fleet to reach 14,000 units in 2026, up from a previous forecast of 13,000.
But the other side of the coin must also be faced.
On one hand, China's mobility market is no longer growing. Warnings about ride-hailing saturation have been issued for over a year. No one can give a definitive answer on how much market value large-scale Robotaxi deployment can actually create.
On the other hand, while top players have achieved positive unit economics, the gap between "single-vehicle profitability" and "company profitability" is as wide as the distance between a single store breaking even and an entire chain turning a profit. Moreover, expanding into a new city requires significant upfront fixed costs.
Image Source: Screenshot from East Money
The stock performance of Pony.ai and WeRide illustrates the problem—revenue is rising, but share prices are falling.
There is also uncertainty regarding technology and regulations.
On the tech front, a massive system failure by Baidu Apollo Go in Wuhan in late March served as a wake-up call. At the time, multiple Robotaxis stopped running en route, leaving some passengers stranded inside the vehicles for extended periods. Baidu Apollo Go suspended services across Wuhan as a result, and there is still no news on a restart.
This incident shows that despite multiple rounds of iteration and broad verification, the system stability and robustness of Robotaxis in large-scale operational scenarios still fall short of the requirements for true commercial deployment.
In terms of regulations, rules on liability and insurance for Robotaxi accidents remain unclear in some countries and regions, directly impacting expansion rhythms. The potential impact of Robotaxis on social employment structures has also sparked widespread concern. Technology moving fast doesn't automatically solve every problem.
So, when everyone shouts "Year One," it's precisely when the accounts need to be clearest.
From "storytelling" to "calculating the economics," the Robotaxi sector has spent more than a decade getting here. The excitement of 2026 isn't the finish line—it's the start of the real test.
