XPeng uses 3,000 TOPS Turing chips

- XPeng said on May 18 it began mass production of its first robotaxi in Guangzhou, using an in-house Level 4, pure-vision autonomous stack. - Four self-developed Turing AI chips provide 3,000 TOPS of on-board compute, which XPeng said powers perception without LiDAR or high-definition maps. - XPeng said pilot robotaxi operations start in H2 2026, with fully driverless service targeted by early 2027.

XPeng said on May 18 that its first mass-produced robotaxi had rolled off the production line in Guangzhou, giving the Chinese electric-vehicle maker a concrete launch point for a service it says will begin pilot operations in the second half of 2026. The vehicle is built on the company’s GX platform and is engineered to Level 4 autonomous-driving standards, according to XPeng’s release. Reuters reported the company is targeting fully driverless operations by early 2027. ### Why does the chip number matter? XPeng said the robotaxi uses four self-developed Turing AI chips with a combined 3,000 TOPS, or tera operations per second, of effective on-board computing power. In plain terms, that is the processing budget XPeng says it needs to run the vehicle’s perception and decision stack inside the car rather than relying on a simpler assisted-driving system. (prnewswire.com) The company described that 3,000-TOPS figure as “industry-leading” in its announcement. CnEVPost, citing the same company release, said XPeng plans to launch three robotaxi models in 2026 and said the Turing chips are central to that hardware strategy. ### What is XPeng actually saying about the driving system? (prnewswire.com) XPeng said the robotaxi runs a pure-vision Level 4 system built on its in-house full-stack advanced driver-assistance technology and Turing system-on-chip architecture. The company said the vehicle does not use LiDAR or high-definition maps, a notable design choice in a robotaxi market where some rivals still rely on more sensor-heavy stacks. (prnewswire.com) The GX-based robotaxi is also tied to XPeng’s VLA 2.0 end-to-end large model architecture, according to company materials and follow-on coverage. XPeng said the model underpins perception and driving decisions for the vehicle’s vision-only setup. ### How far along is this beyond a concept car? Guangzhou is where XPeng said the first mass-produced unit rolled off the line, and the company framed the event as China’s first automaker-led mass production of a robotaxi through full-stack in-house development. (prnewswire.com) Reuters separately reported that XPeng had begun mass production at its Guangzhou headquarters. (xpeng.com) Electrek reported the rollout makes XPeng the first automaker in China to mass-produce a fully in-house Level 4 robotaxi. That does not mean a public commercial network is already operating at scale, but it does mean XPeng is now presenting the vehicle as a production program rather than only a test fleet. ### What happens next? (prnewswire.com) XPeng said it plans to start pilot robotaxi operations in the second half of 2026 to test technical performance, user acceptance and the business model. The company said it aims to remove the on-site safety officer by early 2027, which is the milestone it is using for fully autonomous operations. Amap is set to become XPeng’s first global ecosystem partner as the company opens its robotaxi software development kit, according to XPeng’s website. (electrek.co) That gives the next phase a named outside participant as XPeng moves from factory rollout to pilot service deployment. (xpeng.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.