Recently, automotive chips have once again moved to the center of industry debate.
On one side, BYD released its self-developed 4nm intelligent driving chip, Xuanji A3. NIO, XPeng, Li Auto and other technology-driven carmakers are also pushing forward with self-developed chips or in-house AI computing architectures.
On the other side, memory chip prices are rising sharply, becoming a very real pain point for automakers and suppliers.
AI data centers are consuming huge amounts of HBM, DRAM, NAND and advanced packaging capacity. The limited number of global memory chip leaders are shifting more production capacity, capital expenditure and customer priority toward AI computing.
At the same time, Qualcomm just held its 2026 Automotive Technology and Cooperation Summit, where the focus has clearly moved beyond cockpit chips to cockpit-driving integration, AI agents, Physical AI and in-vehicle intelligence.
Horizon Robotics' recent stock volatility has also made the market rethink the business model of third-party intelligent driving chip and solution providers.
And if Tesla's supervised FSD further enters and improves in China, the pressure on Chinese OEMs and intelligent driving suppliers will become even stronger.
These may look like separate news stories. But together, they point to one major trend:
The automotive chip battle is shifting from "who has higher computing power" to "who controls the future smart car."
In the past, chips were mainly a supply chain issue. Today, chips are becoming a core part of automakers' intelligence capability, cost structure, supply chain security and architectural control.
So the real question is: what exactly is this new chip battle about?
From Computing Power to Control
The first structural change is the shift in vehicle E/E architecture.
In the past, a vehicle could have more than 100 distributed ECUs. Cockpit, intelligent driving, body control and chassis systems were largely separated.
But smart vehicles are moving toward cockpit-driving integration, zonal control and central computing.
This means the future is not about having more chips in the car. It is about whether the vehicle can support more complex software and AI capabilities with higher integration, lower latency and safer isolation.
Take cockpit-driving integration as an example.
Image source: Qualcomm
It is not simply putting the cockpit and intelligent driving functions onto one chip.
The real challenge is how to run tasks with different safety levels and real-time requirements on the same computing platform.
An infotainment system may freeze or restart. But perception, decision-making and control chains related to intelligent driving cannot be affected. Safety-related functions must remain real-time, reliable and isolated.
That is why automotive chips cannot simply follow the logic of consumer electronics.
Smartphone chips focus on performance, power consumption and user experience. Automotive chips must also answer questions of safety, reliability, redundancy, lifecycle and vehicle-grade certification.
So the automotive chip competition is no longer a simple TOPS race.
What really matters is effective computing power, model adaptation, memory bandwidth, power efficiency, functional safety and engineering capability.
In other words, chips are no longer just hardware components. They are part of the vehicle's intelligent architecture.
Whoever defines the chip gets closer to defining the capability boundary of the future smart car.
AI Models Are Changing Chip Expectations
The second major change is the arrival of large AI models in vehicles.
In the past few years, when automakers talked about intelligent driving chips, they often focused on computing power: how many TOPS, whether the chip could support NOA, or whether it could enable urban assisted driving.
But once large AI models enter the vehicle, expectations change.
End-to-end intelligent driving requires deeper integration of perception, prediction and planning.
Multimodal interaction requires the vehicle to process voice, images, scenarios and user intent at the same time.
In-car AI agents need to understand, reason and respond inside the vehicle.
Future Physical AI may even require the vehicle to understand the real world and continuously interact with its environment.
These tasks do not only require more computing power. They also require more efficient data flow and model deployment.
A chip may look impressive on paper with a high TOPS number. But if memory bandwidth is limited, on-chip cache is insufficient, model deployment is inefficient, or the algorithm does not match the chip architecture, the real user experience will not come out.
This is why the industry is now talking about large models in vehicles, in-car AI, cockpit-driving integration, AI agents and Physical AI.
Behind all these concepts is one fundamental change:
The car is moving from electrification and digitalization into the AI era.
The in-vehicle chip must evolve from a traditional computing platform into an AI inference platform.
In the future, buying a high-computing-power chip will not be enough. The real competition is whether the automaker can make chips, algorithms, models and vehicle architecture work together as one system.
Memory Chips Become the Real Pain Point
This brings us to the most immediate pain point: memory chip price increases.
In 2021, when people talked about automotive chip shortages, they often thought of MCUs, power semiconductors and analog chips.
But this round is different.
The tightest area may not be the intelligent driving SoC that everyone talks about. It may be memory.
Why?
Because high-end smart vehicles need larger and faster memory for multi-screen cockpits, lidar systems, large-model NOA, end-to-end driving models and in-vehicle AI.
This increases demand for DRAM, LPDDR, eMMC, UFS and NAND.
But AI servers are also competing aggressively for memory.
HBM, DRAM, NAND and advanced packaging resources are being locked up by AI data centers. This means automakers may face weaker bargaining power and lower supply priority in some memory categories.
Will memory price increases lead to broader automotive chip supply pressure?
My view is: yes, but not all chips will be in shortage at the same time.
The direct pressure is on automotive memory chips. But the impact will not stop there.
AI servers are absorbing memory, packaging, wafer and materials resources. At the same time, once automakers see rising prices and longer delivery times, they tend to lock orders earlier and build inventories earlier.
This supply chain psychology may also spread to MCUs, power devices, analog chips and power management chips.
So the 2026 automotive chip issue is not a repeat of the 2021 full-scale shortage.
It is more likely to be structural tension: higher costs, configuration adjustments, delivery fluctuations, production prioritization and supply chain reshuffling.
The key lesson is clear: Smart vehicles need not only computing power. They also need memory, packaging, capacity and supply certainty.
Tesla FSD and OEM Chip Development
Another major variable is Tesla FSD.
If Tesla's supervised FSD further enters and improves in China, the competition will become more intense.
Tesla brings not just another driver assistance feature, but a system supported by global fleet data, end-to-end models and large-scale software iteration.
Image source: Tesla
Chinese automakers have made very fast progress in urban NOA, end-to-end driving and AI models. In many Chinese road scenarios, they already have strong local advantages.
But if Tesla FSD really runs at scale in China, the competition will move from "who has intelligent driving features" to "who has better stability, generalization, iteration speed and cost efficiency."
That is why automakers are paying much more attention to chip definition.
BYD's Xuanji A3 is important not simply because BYD has released another chip. It signals that leading automakers are no longer satisfied with only buying chips from outside.
NIO, XPeng and Li Auto are also moving toward self-developed chips or deeper AI computing architectures.
But not every automaker needs to fully develop its own chip. Chip development requires huge investment, long cycles, engineering validation and enough vehicle volume to dilute cost.
For million-unit-scale automakers, however, chip self-development or deep chip definition may become a tool for controlling cost, performance and architecture.
This is why BYD's move matters.
If BYD can combine self-developed chips, self-developed algorithms, large-scale vehicle deployment and cost control, it may redefine the cost structure of advanced intelligent driving.
Other large Chinese automakers may not all build chips themselves, but they will almost certainly participate more deeply in chip definition, architecture planning, algorithm adaptation and capacity locking.
The essence of OEM chip development is not to prove "we can also make chips."
It is to turn intelligence capability from an external purchase into an internal system capability.
Third-Party Platforms Are Also Being Redefined
So what happens to third-party chip companies?
Qualcomm and Horizon Robotics are both outside automakers, but their roles are different.
Qualcomm is more like a third-party intelligent computing platform company. It is competing for the computing entrance, software ecosystem and developer ecosystem of the future smart car.
Image source: Qualcomm
Horizon Robotics goes deeper into intelligent driving. It provides not only chips, but also algorithms, toolchains, reference solutions and production support.
In advanced intelligent driving, Horizon is no longer just a chip supplier. It is increasingly becoming an intelligent driving solution provider and ecosystem platform.
The challenge is that automakers are also moving deeper into chip definition, and some are starting to develop chips themselves.
So third-party players must prove that they are not just substitutes for in-house development.
They must prove they are still more valuable in cost efficiency, production speed, algorithm adaptation and open ecosystem support.
In the future, the winners will not be determined by a single chip.
Automakers want to regain definition power.
Platform chip companies want to occupy the computing entrance.
Intelligent driving solution providers want to prove they remain the best choice.
And underlying semiconductor resources are being redistributed by AI.
That is the real essence of the escalating automotive chip battle.
Chips are not just computing power.
Chips are part of the control system of the future smart car.
