Pony.ai Gen-7: the robotaxi built on the line
A Toyota bZ4X that comes off GAC Toyota's own production line already autonomous — four lidar, an automotive-grade kit rated to 600,000 km, and a bill of materials 70 per cent below the last generation. Nobody else is building robotaxis this way, and it is the reason Pony.ai can talk about thousands of cars rather than hundreds.
There are three ways to get a robotaxi. You can retrofit a car somebody else built, which is what Waymo did with the Jaguar I-Pace and is still doing across most of its fleet. You can design a purpose-built vehicle and manufacture it yourself, which is the Cybercab and the Zoox pod. Or you can persuade an established carmaker to build the autonomous version alongside the ordinary one, on the same line, as a variant. Pony.ai took the third route, and it is the least discussed and probably the most consequential.
The Gen-7 bZ4X robotaxi entered mass production at GAC Toyota in February 2026. It is not a bZ4X with sensors added afterwards. It is a bZ4X built as a robotaxi, by Toyota's Chinese joint venture, to Toyota's manufacturing standards, with Pony.ai's autonomous driving kit fitted as part of the vehicle rather than bolted to it.
Vehicle platforms running the same Gen-7 kit: the Toyota bZ4X, the BAIC ARCFOX Alpha T5 and the second-generation GAC Aion V. One autonomy stack, three carmakers' production lines.
What is actually on the car
The sensing is deliberately conventional and deliberately generous. Every Gen-7 robotaxi carries four Hesai AT128 lidar units as its primary lidar — 200 metres of range each, a 120-degree field of view, 1.53 million points a second — arranged for full 360-degree coverage, alongside cameras and radar. There is no attempt here to prove a point about doing without.
That is worth holding next to the Tesla Cybercab, which we reviewed last week. Both cars make almost exactly the same power: 219 hp, front-wheel drive, in a vehicle designed to spend its life doing short urban trips. One carries four lidar and the other carries none, and they are the two purest statements of the opposing bets in this industry, built to the same performance envelope for the same job.
The kit is 100 per cent automotive-grade with a design life of 600,000 kilometres, which is the specification that matters most and gets discussed least. A robotaxi that needs its sensors replaced at 150,000 km has an economic problem no software update fixes. Building to 600,000 km means the hardware is meant to last as long as the fleet duty cycle, and it is only achievable if the parts come from an automotive supply chain rather than a robotics one.
The cost collapse
Pony.ai says the Gen-7 autonomous driving kit costs 70 per cent less to build than Gen-6, with the compute down 80 per cent and the solid-state lidar down 68 per cent, and a further 20 per cent cut scheduled for 2026 production. The stated destination is a complete robotaxi — vehicle and kit together — for under RMB 230,000 by 2027, which is roughly $32,000.
Set that against the estimate of about $125,000 for a Waymo Ojai. If Pony.ai gets anywhere near its number, the capital cost of putting a driverless car on the road falls by three quarters, and the question stops being whether robotaxis are affordable and becomes how fast anyone can build them. That is the whole argument for doing it this way: a carmaker's production line is a machine for making things cheaper at volume, and Pony.ai has rented three of them.
"A carmaker's production line is a machine for making things cheaper at volume, and Pony.ai has rented three of them."
Does it work
Yes, and with numbers attached, which is rarer than it should be. Pony.ai holds fully driverless commercial permits in all four Chinese tier-one cities — Beijing, Shanghai, Guangzhou and Shenzhen — which no other operator can claim. Its fleet stood at 1,975 vehicles on 30 June 2026 against a year-end target of more than 3,500. Hesai cites more than 500,000 autonomous hours and a claimed tenfold safety improvement over human drivers, which is the company's figure rather than a regulator's and should be read that way.
More usefully, Pony.ai has published unit economics. It says Gen-7 cars reached breakeven across Guangzhou in November 2025 and Shenzhen in February 2026, at 23 rides a day per vehicle and RMB 338 of net revenue — about forty-seven dollars — on a one-month average. The definition it gives includes vehicle and kit depreciation, charging, maintenance, remote assistance, insurance, ground staff and parking. Counting kit depreciation and remote assistance is what makes the claim worth taking seriously; most operators quietly leave both out.
What it excludes is research, overhead and the capital to buy the fleet, which is why the company still lost $45.4 million in the second quarter of 2026 on revenue of $36.2 million. The car pays for itself. The company does not, yet, and has $1.39 billion of cash with which to keep trying.
What we're watching
- Whether the RMB 230,000 target holds for 2027, and whether it means the whole car or just the kit at the point it is claimed.
- How many bZ4X robotaxis actually get built. Thousands were promised for 2026; the fleet was 1,975 across all platforms at the half year.
- Whether any other autonomy company gets a carmaker to build its vehicle on the main line rather than in a special facility. If the model spreads, retrofitting ends.
- Whether unit economics hold as the fleet triples. Breakeven in the best-served districts of two cities is not the same as breakeven everywhere.
- The Gen-4 robotruck shown at IAA in September 2026, which runs the same architecture. Pony.ai's freight business already earns more than its robotaxi business.
Spec sheet — Pony.ai Gen-7 robotaxi (Toyota bZ4X)
The ai.cars verdict
- The only robotaxi built as a variant on a mainstream carmaker's own production line, rather than retrofitted or purpose-built in a special facility. That is the story.
- Four Hesai AT128 lidar as standard, 200 m each, with full 360-degree coverage — the exact opposite bet to a Cybercab making identical power with none.
- Automotive-grade kit rated to 600,000 km. The least glamorous specification here and the one that decides whether the economics work.
- Kit bill of materials down 70 per cent on Gen-6, targeting a complete car under RMB 230,000 by 2027 — roughly a quarter of what a Waymo Ojai is reckoned to cost.
- Driverless and paid in all four Chinese tier-one cities, with published unit economics that count kit depreciation and remote assistance rather than hiding them.
- The car pays for itself at 23 rides a day. The company lost $45.4m last quarter and has $1.39bn to keep going.
- Verdict: the most credible route to robotaxis at volume that anyone is currently taking, and the one fewest people are watching.
How does it stack up? Compare the Gen-7 with the Tesla Cybercab, or build your own comparison.
