2026-06-18 👁️ — views

Physical AI in China — Baidu, WeRide, Pony.ai Are Running Robotaxis Now

China runs a parallel robotaxi ecosystem today. Baidu Apollo, WeRide, and Pony.ai operate driverless fleets — the US-China race is closer than most realize.

Article 97 in the Physical AI Benchmark Series — Physical AI in China: Baidu Apollo, WeRide, and Pony.ai Are Running Commercial Robotaxis Today, and the US-China Race Is Closer Than Most Americans Realize

While Tesla and Waymo dominate US Physical AI headlines, China has built a parallel robotaxi ecosystem that is commercially operational, heavily supported by the state, and expanding rapidly. Baidu Apollo operates the largest driverless robotaxi fleet in China under the Luobo Kuaipao brand; WeRide and Pony.ai have both received driverless permits in multiple Chinese cities and are expanding internationally, with Nasdaq listings completed in 2024. US semiconductor export controls have created a training compute constraint — but Chinese companies are aggressively pursuing domestic chip alternatives and compensating with different road data and cost structures.

This article maps the Chinese Physical AI ramp as a global benchmark dimension: who is operating, at what scale, with what technology constraints, and what it means for the global competitive picture across the US-China Physical AI race.

Section 1 — China’s Leading Robotaxi Operators (Est. Mid-2026)

China has five significant robotaxi programs at various stages of commercial operation. Three have achieved substantial driverless commercial scale.

The top three operators — Baidu Apollo, WeRide, and Pony.ai — account for the vast majority of China’s commercial driverless ride volume (est.). All three have received regulatory approval to operate driverless (no safety driver) in at least some zones of multiple Chinese cities. Two of the three have completed international Nasdaq listings, signaling a level of institutional validation and capital access that earlier-stage Chinese tech companies did not achieve.

The comparison to the US landscape is instructive: Waymo operates commercially in four US cities (San Francisco, Phoenix, Los Angeles, Austin). The Chinese operators collectively operate across more cities by count, though the operational density and ride volume per city vary significantly (est.).

Section 2 — Baidu Apollo: The Ramp Leader

Baidu’s robotaxi brand Luobo Kuaipao (萝卜快跑, literally “Radish Run”) has achieved commercial scale that makes it arguably the largest single robotaxi operator globally by deployed vehicle count (est.). Baidu has invested heavily in the Apollo autonomous driving platform for over a decade and is now harvesting that investment in commercial operations across more than ten Chinese cities (est.).

The state-support dimension is the most structurally significant difference between the US and Chinese robotaxi ramps. Chinese municipal governments have designated specific urban zones as AV commercial operation zones and provided regulatory approval pipelines that move faster than US counterparts. National policy frameworks explicitly designate autonomous vehicles as a strategic technology priority — meaning capital, permits, and infrastructure support flow to AV operators as part of deliberate industrial policy rather than through purely market-driven approval processes.

Baidu’s pricing below taxi rates (est.) is a deliberate consumer adoption strategy. By offering rides at a price below human-driven alternatives, Baidu creates consumer familiarity with the technology at scale — building the behavioral and social acceptance that the US autonomous vehicle sector has struggled to generate outside of specific Waymo cities.

Section 3 — The Semiconductor Constraint: US Export Controls and Their Effect

US export controls (BIS Entity List restrictions, October 2022 and subsequent tightening) have significantly constrained Chinese AI companies’ access to Nvidia A100/H100 GPUs — the chips that power AI model training at scale. For autonomous vehicle companies, this creates a specific and quantifiable training compute constraint.

The training compute constraint is real but not fatal to China’s AV ramp. Autonomous driving at the commercial level China’s operators are currently achieving does not require frontier H100-scale training in real-time; it requires sufficient compute to train and iterate on the models that run on the vehicle. The constraint is on the rate of model improvement, not on the ability to operate current-generation models commercially.

The more significant near-term effect may be on the frontier: the next generation of foundation models for autonomous driving that US companies like Waymo and Tesla are training. Chinese companies cannot train those models at the same compute scale, which may create a widening performance gap in model-driven driving quality over a 3–5 year horizon (est.). Whether that gap matters more than China’s advantages in cost, regulatory access, and domestic market scale is the core strategic uncertainty.

Section 4 — Tesla in China: The FSD Approval Puzzle

Tesla is the world’s largest foreign automaker in China by production (Shanghai Gigafactory, est. 750,000 vehicles per year capacity). FSD approval in China is strategically critical to Tesla’s global data flywheel — and it remains pending.

The FSD approval puzzle illustrates the geopolitical dimension of Physical AI. Tesla’s FSD is technically capable — the same system that is operating in Austin and receiving OTA improvements globally. The constraint is not technical; it is regulatory and geopolitical. China has tools to use technology approvals as leverage in broader US-China tech trade negotiations, and autonomous driving software from the world’s largest foreign automaker is a particularly sensitive domain.

If FSD approval comes through, the scale effect is transformative: Tesla sells hundreds of thousands of vehicles annually in China. Adding those vehicles to the supervised FSD fleet — and eventually to a shadow mode data collection flywheel — would be one of the largest single additions to any autonomous driving training dataset in history (est.).

Section 5 — Global Competitive Benchmark: US vs China Physical AI

The honest benchmark conclusion is that the US-China Physical AI race is not a rout in either direction. The US holds meaningful advantages in training compute access and in the scale of Tesla’s consumer vehicle data flywheel. China holds meaningful advantages in regulatory fast-tracking, domestic supply chain cost, and the sheer number of Chinese cities where commercial driverless operations are already live.

The gap that matters most over a 5-year horizon is the training compute gap. If US export controls succeed in materially constraining China’s ability to train next-generation AV foundation models, that could translate into a compounding performance gap in driving model quality. If China’s domestic chip industry closes the gap — Huawei Ascend, Horizon Robotics, and state-backed chip programs are all advancing — the training compute constraint diminishes and China’s cost and scale advantages dominate.

Section 6 — What This Benchmark Means for Physical AI Investors

The China dimension of Physical AI is systematically underweighted in Western investor analysis. Most Western Physical AI investment frameworks treat the market as US-centric — Waymo vs Tesla, with EU AV programs as a distant third. The commercial reality is that China has built a parallel AV ecosystem that is operating commercially at comparable fleet scale (est.) to the entire US robotaxi industry combined.

The investment implications:

Valuation comparisons. WeRide (WRD) and Pony.ai (PONY) are publicly traded on Nasdaq at valuations substantially below Waymo (est. $50B+). Whether that discount reflects genuine technology/market risk or Western market unfamiliarity with Chinese AV operations is an open question that this benchmark maps but does not resolve.

Technology export asymmetry. US semiconductor export controls create asymmetric risk: US AV companies have no corresponding constraint on their access to compute, while Chinese competitors face a meaningful training-compute headwind. That headwind may widen over time if controls tighten, or narrow if domestic Chinese chips improve faster than expected.

Domestic market scale. China is the world’s largest auto market by unit volume. An AV operator that achieves commercial-scale deployment across China’s major cities has access to a domestic ride-hail market of extraordinary scale — one that does not require the difficult international regulatory navigation that US AV companies face when expanding.

The Tesla wildcard. FSD approval in China would be among the largest single events in the global Physical AI competitive landscape. It would add Tesla’s China production volumes to the supervised FSD fleet and eventually to the data flywheel — directly competing with Baidu Apollo on price and data scale in the world’s largest auto market.

Section 7 — About This Series

This is article 97 in the Physical AI Benchmark Series. Previous articles have covered the ramp index, the humanoid race, unit economics, global competition, HD mapping, software and OTA updates, consumer demand, competitive moats, safety data, Waymo Gen 6, Optimus manufacturing, scorecard snapshots, 2030 forecast scenarios, the investor framework, city expansion pipelines, Tesla FSD state approval maps, AV weather and climate constraints, regulatory calendars, robotaxi fare pricing, humanoid deployment trackers, supply chain analysis, consumer adoption demand index, valuation and IPO analysis, the Physical AI 2026 mid-year roundup, AV unit economics cost-per-mile breakdown, the AV data flywheel comparison, the Physical AI supply chain, AV fleet operations, the full lifecycle environmental cost, the accessibility layer, the mapping architecture comparison, the China AV race, simulation and synthetic data training, AV urban planning and city impact, autonomous trucking freight economics, the European AV competitive landscape, the AV sensor technology debate, AV safety metrics, the AV talent war, the global AV regulatory map, AV financial sustainability burn rates, the Tesla Cybercab versus Waymo Gen 6 head-to-head (article 84), AV cybersecurity attack surfaces (article 85), the humanoid robots commercial deployment landscape (article 86), AV fleet electrification and the charging race (article 87), AV data as a business (article 88), AV insurance and liability (article 89), the driverless cabin and passenger experience (article 90), the Physical AI investment landscape (article 91), AV safety vs human drivers statistics (article 92), AV accessibility for elderly and disabled populations (article 93), Waymo’s city expansion playbook (article 94), Tesla’s FSD data flywheel (article 95), and the Tesla Cybercab unit economics (article 96).

This article adds the China Physical AI dimension: who is operating commercially, at what scale, under what semiconductor constraints, and how the US-China robotaxi race looks when mapped as a global benchmark.

Note: All fleet size estimates, ride volume figures, valuation estimates, and competitive assessments in this article are directional estimates based on company public disclosures, Nasdaq filings, analyst research, press coverage, and publicly available regulatory filings as of mid-2026. Where data is uncertain or estimated, figures are labeled “(est.)” and should be treated as directional rather than confirmed definitive figures. This article does not constitute investment advice.

Sources

• Baidu Apollo Luobo Kuaipao operations — Baidu IR ↗
• WeRide Nasdaq listing and international expansion — WeRide ↗
• Pony.ai Nasdaq IPO and operations — Pony.ai ↗
• US semiconductor export controls — BIS Commerce ↗
• Tesla China FSD regulatory status — Tesla IR ↗

Tags