2026-06-17 👁️ — views

International Physical AI Race — China, Europe, and the Global Competitive Map (Mid-2026)

Baidu operates 1,000+ robotaxis in 10 Chinese cities, Wayve raises $1.05B, and Europe sets strict AV rules — the global physical AI race is accelerating.

Why a global map matters

The US-centric narrative around Waymo and Tesla misses a crucial fact: China’s AV industry is already operating at commercial scale in more cities than any US player, and Europe has produced the world’s first Type-approved Level 3 highway system. The physical AI race is global — and the competitive dynamics are shaped as much by regulation and chip access as by engineering. This is Article 8 in the physical AI benchmark series.

Global Physical AI Landscape

Threat level reflects risk to US-led (Waymo / Tesla) narrative and market share, not geopolitical risk.

Section 1 — China’s structural advantages

China has built three compounding advantages in physical AI that are difficult to replicate on a short timeline.

Government mandate as a growth lever

Chinese municipalities operate designated AV zones with rapid permit pathways. The central government has made autonomous driving a strategic priority under its New Infrastructure initiative. Unlike the US state-by-state patchwork, China’s system allows a single policy shift to unlock commercial operations across dozens of cities simultaneously.

Baidu Wuhan was the first city globally to offer fully driverless commercial rides at meaningful density — predating Waymo’s equivalent density in Phoenix by a measurable period. Wuhan’s “Apollo Park” zone grew from a test area to a live fare-paying service without a safety driver, a threshold that took US regulators years to approve elsewhere.

BYD: the hidden training-data threat

BYD surpassed Tesla in global EV volume in 2023 and shipped over 3 million EVs annually by 2026. BYD is developing a humanoid robot platform (“BYD Robot”) alongside in-car AI systems for its vehicle lineup.

The strategic implication: if BYD deploys fleet-learning autonomy across even a fraction of its production vehicles — mirroring Tesla’s FSD data-collection model — it would instantly rival Tesla’s fleet scale in the Chinese market. BYD has a structural manufacturing cost advantage (vertical integration of battery, motor, and chassis) that makes any hardware-cost argument in robotics or AVs difficult for external entrants to match.

China’s humanoid cluster

China’s state-directed industrial policy has funded more than 20 humanoid startups since 2023. Published government targets for 2025 included 10,000 humanoid units deployed in factory settings. While the execution rate against that target is uncertain (label: estimate), the policy infrastructure — preferential land allocation, subsidized compute access, pilot factory designations — is real and accelerates both capital formation and talent concentration.

Section 2 — Europe’s regulatory approach

Europe’s path is neither China’s mandate-speed nor the US’s state-by-state experiment. It is slower, more prescriptive, and potentially more durable.

EU AI Act and UN ECE WP.29

The EU AI Act classifies certain autonomous vehicle systems as high-risk AI, triggering conformity assessments, explainability requirements, and human-oversight mandates before deployment at scale. UN ECE WP.29 is the international framework governing type-approval of automated driving systems — both L3 and L4 — across 54 signatory countries including the EU, Japan, and South Korea.

This regulatory architecture slows initial deployment but creates a defensible moat for early approvals. A system that clears WP.29 type-approval can, in principle, scale across all signatory markets simultaneously.

Mercedes-Benz Drive Pilot: the world’s first Type-approved L3

In 2021–2022, Mercedes-Benz became the first manufacturer to receive Type-approval for a Level 3 conditional automated system under UN ECE Regulation 157. Drive Pilot operates at up to 60 km/h on German Autobahn sections, allows the driver to fully disengage from monitoring (hands off, eyes off), and assumes legal liability for the drive during activation.

The limitation — 60 km/h, specific road sections, specific weather — means Drive Pilot is commercially modest today. But it is the first proof point that a production vehicle has satisfied the legal and technical bar for L3 in a major regulatory regime. It establishes the certification pathway for future L3 and L4 systems.

Wayve’s differentiated approach

UK-based Wayve raised a $1.05B Series C in 2024 — the largest single raise in European AV history — with strategic backing from SoftBank, NVIDIA, and Microsoft. Unlike every other European AV program, Wayve does not operate a robotaxi fleet. Instead, it trains its embodied AI driving policy on consumer vehicle data, a direct parallel to Tesla’s FSD data-collection model.

This matters because it sidesteps the permit bottleneck: Wayve’s system accumulates training data from ordinary consumer vehicles on ordinary roads, not from a licensed robotaxi zone. If Wayve’s approach generalizes — and the NVIDIA and Microsoft backing suggests belief that it will — it could scale rapidly without requiring city-by-city regulatory approval for a commercial fleet. The acquisition risk is real: with Microsoft and NVIDIA already inside, Wayve is a natural absorption target for a larger AV stack.

Section 3 — US vs. China AV ramp comparison

Ride counts for Baidu labeled estimate; Waymo figure from public reporting.

Section 4 — Implications for Tesla and Waymo

China’s speed creates narrative risk

The commercial-scale benchmark matters for investor and regulatory confidence, not just operational efficiency. If Baidu reaches 500,000 weekly rides before Waymo expands to its fifth major city, the narrative around “who is winning autonomous vehicles” shifts — regardless of safety performance or technology sophistication differences.

The competitive risk is not that Baidu enters the US market. It is that the global technology framing changes from “US leads physical AI” to “China already operates at scale,” which affects capital flows, talent allocation, and policy priorities.

Export restrictions: near-term pain, long-term uncertainty

The NVIDIA H100 and A100 export bans to China, active since October 2022, forced Baidu, WeRide, and their peers to accelerate investment in domestic alternatives. Horizon Robotics (Journey 6 chip) and Huawei (Ascend 910B) are the primary domestic alternatives. Both are measurably behind NVIDIA’s current-generation training throughput — creating a real compute disadvantage for Chinese players in training large-scale AV foundation models.

The uncertainty: the export restrictions accelerated China’s domestic chip industry by creating both demand and government subsidy. A 3–5 year lag in training compute today may close faster than the lag in, say, semiconductor fabrication nodes. The restriction is a headwind, not a wall.

Wayve as acquisition target

Microsoft’s and NVIDIA’s participation in Wayve’s Series C is strategic, not passive. Both have strong incentives to integrate a mature AV foundation model into their broader AI infrastructure stacks. Wayve’s consumer-fleet training approach — if it generalizes — would be more valuable inside a cloud (Microsoft Azure) or chip ecosystem (NVIDIA DRIVE) than as a standalone company. An acquisition would accelerate Wayve’s scale while giving either acquirer a credible foothold in European AV.

What to watch in H2 2026

1. Baidu weekly rides trajectory — Does Baidu hit 200K+ weekly rides before Waymo hits 200K? The crossing point is the narrative inflection.
2. Horizon Robotics Journey 6 training benchmarks — Public performance disclosures will reveal how close domestic Chinese chips are to NVIDIA H100 equivalents.
3. WeRide’s Middle East expansion — Abu Dhabi driverless permits give WeRide a non-China proving ground with English-language visibility.
4. Wayve deployment partnership — Which OEM or fleet operator announces a Wayve licensing agreement? A volume OEM deal would validate the consumer-fleet model.
5. EU L4 type-approval timeline — When does the first L4 system receive EU type-approval under the emerging WP.29 L4 framework?
6. BYD in-car AI announcement — If BYD announces fleet-learning autonomy tied to its production vehicles, recalibrate the training-data threat from Medium to High.

The global physical AI race is not US vs. one challenger — it is a multi-front competition where regulatory regimes, chip access, fleet scale, and humanoid ambition each create different advantages for different players. The US leads in technology sophistication and commercial driverless operations today. The lead is real but not irreversible.

Sources

• Baidu Apollo robotaxi operations — Baidu Q1 2026 earnings ↗
• WeRide NYSE listing and global operations — WeRide investor relations ↗
• Wayve $1.05B Series C — Wayve press room ↗
• Mercedes-Benz Drive Pilot L3 approval — Mercedes press ↗
• Pony.ai NYSE listing — Pony.ai investor relations ↗

Tags