Editor’s comment: “In this guest column, US attorney Eric Postow, of Holon Law Partners, brings a holistic perspective to self-driving, considering the multiple intricate, interconnected relationships of automated vehicles operating within broader emerging ecosystems of data, regulatory architecture, and urban design.”
Nothing moves in isolation: AI in the auto industry and the society that surrounds it
In April 2026, hundreds of journalists gathered at the Beijing International Automotive Exhibition to watch NIO, Xpeng, Xiaomi, Huawei, and Dongfeng unveil the next generation of intelligent vehicles.
NIO’s flagship ES9 debuted with over 40 industry-first technologies. Huawei announced a substantial multi-year investment to expand computing power for smart driving. The message across the floor was direct: software is now the product.
As Nissan Motor’s China chief, Stephen Ma, put it: “There is no longer a distinction between a technology company and a car company.” The line between vehicle and urban system is blurring. That convergence is the subject of this article.
The argument is this: a fully realized automated driving economy is not purely a product problem – it is also a societal infrastructure problem. Vehicles are increasingly nested within broader systems of data, computing power, regulatory architecture, and urban design.
The country that grasps this is not simply the one building the best car, it is the one building the most capable environment for that car to operate in.
The SAE automation ladder and what it demands
The SAE International levels of driving automation – Level 0 (no automation) through Level 5 (full autonomy, no human intervention) – are a useful shorthand, but they conceal as much as they reveal.
It is worth noting where the industry actually stands. Most commercially deployed systems today operate at Level 2 or Level 2+, meaning the driver must remain engaged and ready to intervene.
True Level 3 deployment, where the vehicle manages all driving tasks under defined conditions, remains limited. Level 4 systems – capable of full autonomy within a geofenced area – exist in controlled commercial pilots, notably robotaxi services in specific cities. Full Level 5 autonomy remains ahead.
The progression from Level 2 to Level 4 is not primarily a sensor or software challenge. It is an ecosystem challenge. Higher automation levels require rapid decisions using real-time environmental data: traffic conditions, road surface changes, pedestrian behaviour, and emergency vehicle movements.
Core driving decisions are processed onboard – that is where most real-time computation happens. But reliable performance at scale also depends on map accuracy, infrastructure signals, V2X (vehicle-to-everything) communications, and data ecosystems that no single automaker controls.
Level 5 deserves its own note. In theory, a fully autonomous vehicle could operate entirely on onboard processing, without external infrastructure. In practice, deploying Level 5 at scale – across varied road conditions, geographies, and weather – will almost certainly require dense supporting infrastructure: updated digital maps, connected traffic management, and standardized data exchange. The vehicle may be theoretically self-sufficient; a viable deployment at scale is not.
What separates demonstration projects from mass deployment is the depth of integration between vehicle and environment. China has come to understand this, and is building accordingly.
China: A case study in nested intelligence
The most consequential developments in Chinese automotive AI are not happening inside the vehicles. They are happening in the cities those vehicles drive through. Shenzhen has been transforming its urban fabric into an AI testbed. Humanoid robots assist with subway security. The city’s intermediate court has piloted AI-assisted case handling.
Autonomous delivery vehicles operate along commercial routes. Robotaxi company Pony.ai has reported progress toward per-vehicle unit economics in Shenzhen, though the broader path to fleet-level profitability remains ongoing.
The city committed ten billion yuan to an AI and robotics industry fund, generated approximately 220 billion yuan from its core AI sector in 2025, and is targeting one trillion yuan in smart terminal output this year.
This is not coincidence. It is policy. The national “AI Plus” blueprint embeds AI into healthcare, manufacturing, and urban infrastructure simultaneously. When NIO’s CEO William Li speaks of developing automotive-grade chips and a whole-vehicle operating system, he is describing a strategy that only makes sense inside a country already building the digital infrastructure those systems depend on.
NIO is developing in-house silicon through its chip unit, Shenji, specifically to tailor chip design around its AI applications – including advanced driver assistance – in ways that external suppliers cannot easily replicate. That investment is credible because the surrounding environment is evolving fast enough to use it.
China’s National Integrated Computing Power Network, launched in 2021, pools capacity from data centers, supercomputer clusters, and intelligent computing centers into something closer to a public utility.
Peng Cheng Lab describes the vision plainly: “In a few years, computing power will be like today’s electricity. Where it is does not matter; if you want to make calculations and receive computing power online, you can just pay and use it.”
Core driving decisions are processed onboard, but the broader ecosystem – map updates, traffic modeling, fleet coordination – depends on distributed computing at scale. China’s “whole-of-nation” approach – mobilizing state labs, tasking leading firms with building dedicated AI ecosystems, pooling talent and resources across government and industry – means the car is treated as one node in a larger system.
Over 30 Chinese cities were already building or planning intelligent computing centers as of early 2024. What does this feel like on the ground? A contact who recently returned from a family trip to China described a detail that stayed with them. A family member had gone to bed. Overnight, their NIO vehicle navigated itself to a nearby battery swap station, exchanged its depleted pack for a charged one, and returned home. By morning the car was ready.
This is NIO’s Battery as a Service model in practice – 20 swap stations supporting the Hexi Corridor section from Xi’an to Dunhuang, with the full Silk Road route of over 30 stations projected to connect by September 2026. When vehicle, infrastructure, and service network work together seamlessly, the experience stops feeling like technology. It feels like a utility.
None of this is without complication. China’s approach raises legitimate concerns around cybersecurity, data governance, and the concentration of sensitive infrastructure data in state-adjacent institutions.
As autonomous systems become more deeply embedded in urban life, questions about who controls the data those systems generate – and under what legal framework – become central to any serious evaluation.
The UK and USA: Strength in parts, gaps in wholes
The United States and United Kingdom are not starting from nothing. The US is home to Tesla’s driver assistance stack, Waymo’s operational robotaxi service, and some of the world’s leading AI research institutions.
The Brookings Institution’s 2025 analysis identified San Francisco and San Jose as dominant across all three AI readiness pillars: talent, innovation, and adoption. American venture capital leads global AI funding.
But the same analysis reveals a structural problem. AI activity is heavily concentrated in a short list of coastal hubs, with broad hinterlands that lag across all three pillars. A Level 4 vehicle cannot serve only San Francisco – and a technology that works only in elite metros is not a mass-market product.
The UK brings strong AI research, a pragmatic regulatory culture, and genuine depth in automotive software and design. But neither country has demonstrated the kind of city-scale AI infrastructure deployment that China’s major urban centers are now executing.
US autonomous vehicle programs have faced real setbacks: regulatory friction, liability uncertainty, and the difficulty of handling edge cases in cities not built to support dense autonomous operation.
Of the two, the UK has done more of the right thinking. It has assembled several interlocking pieces that no US state has matched. The Automated Vehicles Act 2024 establishes primary legislation covering safety, liability, and operator responsibilities – with commercial pilots running from Spring 2026.
The Department for Transport’s Transport AI Action Plan, published in June 2025, explicitly connects autonomous vehicles, infrastructure AI, data sharing, rail, and urban mobility into a single government strategy – described by techUK as a first-of-its-kind attempt to map what full AI integration in transport actually looks like.
A companion Transport Data Action Plan addresses data standards, APIs, and cross-organization data governance directly. And the Centre for Connected and Autonomous Vehicles, active since 2015, is already running 5G infrastructure pilots using public assets – streetlights, bus shelters – to support AV connectivity. This is a policy architecture that the US has not built at any level of government.
But architecture is not execution. The same techUK assessment that praised the plan called for more detail on delivery. A March 2025 industry report found that IoT traffic signals and congestion management systems in UK cities remain “modally siloed”, limiting integration between road and rail networks.
The digital infrastructure for vehicle-to-everything communications is still patchy. Full implementation of the AV Act has already slipped from 2026 to late 2027. The UK has the right framework. It has not yet built the system that framework describes.
The deeper difference is framing. In China, autonomous vehicles are a component of urban AI infrastructure. In the US and UK, they are primarily consumer products or corporate technology investments, with infrastructure expected to follow. Both approaches can work – but likely at different speeds.
There are signs that US leadership understands the stakes. At a recent speech in Washington DC, a senior US intelligence official described the global AI competition as a two-nation contest between the US and China, and was clear that the United States does not intend to lose.
The framing was not incremental, it was transformative. On the other side of this era, the official suggested, we will not look like we do today. That seriousness is meaningful. But recognizing the competition is different from structuring a society to win it. The harder question is whether US governance – fragmented, federalized, weighted toward private-sector primacy – can translate urgency into coherent infrastructure investment.
The gap is not only in hardware and deployment speed. Interoperability is equally critical. For autonomous vehicles to function across city boundaries and state lines, data formats, cybersecurity protocols, and communications standards must be harmonized. This is slow, unglamorous work. It is also foundational – without it, capable vehicles are constrained by the patchwork systems around them.
Choosing a route
The Chinese model carries real risks. The whole-of-nation approach creates tension between state coordination and market-led innovation. Regional fragmentation threatens duplicated investment. And the concentration of mobility and urban data in state-adjacent institutions raises questions that democratic societies will and should weigh carefully – on surveillance, civil liberties, and data sovereignty.
The US and UK should not replicate the Chinese model. But they cannot treat autonomous vehicles as purely a product competition. The lesson is structural: the car is not the unit of analysis, the city is – the regulatory and standards ecosystem, and the data-sharing agreements between automakers, municipalities, and infrastructure operators.
Remaining competitive means making decisions now about AI-integrated urban environments, not just AI-enabled vehicles. That means investment in computing infrastructure, regulatory frameworks that enable real-world deployment, data treated as shared infrastructure rather than only a corporate asset, and cybersecurity and interoperability standards elevated to first-order policy priorities.
Disclaimer: Opinions expressed in this column are solely those of the individual contributors.
