After 70% of new cars are equipped with L2 autonomous driving capabilities, the real toughest hurdle for intelligent vehicles has arrived.

If a technology is only featured in 10% of new vehicles, it serves as a selling point.

If 30% of new vehicles are equipped with it, it will likely become a key configuration.

However, when over 70% of new vehicles start to carry this technology, the competition logic will completely change.

The latest industry data shows that the penetration rate of L2-level passenger vehicles in China is expected to exceed 70% in 2026, and may even surpass 90% by 2030.

This means that automatic parking, highway navigation, and urban assisted driving, which have been repeatedly emphasized by automakers in the past few years, are rapidly evolving from "premium configurations exclusive to a small number of car models" to the basic capabilities of intelligent vehicles.

A new problem has emerged accordingly:

When all players can "develop smarter driving capabilities", what exactly will the next round of competition focus on?

The answer is probably not adding a few more functions.

Instead, it refers to who can identify problems earlier, fix problems faster, and have the ability to "manage" hundreds of thousands or even millions of vehicles simultaneously when system defects occur.

In other words, the real challenging stage for intelligent vehicles has shifted from "whether the vehicle can drive autonomously" to "whether the enterprise can properly conduct recalls".

When L2 penetration exceeds 70%, the functions themselves start to depreciate in value

Much of the competition in the automotive industry over the past decade has followed a similar rule.

When a new function first emerges, it is extremely valuable.

The earliest large screens, voice control, panoramic imaging, automatic parking, and even highway assisted driving were once important selling points of high-end car models.

However, the most prominent feature of the automotive industry is its extremely fast scaling speed.

Once the technology matures, the supply chain becomes stable, and the cost drops, the same function will soon be deployed from vehicles priced at hundreds of thousands of yuan to models priced at 200,000 to 300,000 yuan, or even cheaper vehicles.

L2 is currently going through this exact process.

The so-called L2 does not mean that the vehicle can drive completely on its own.

The driver still needs to take full driving responsibility, observe the road conditions and take over the vehicle as required by the system.

But from the perspective of consumer experience, a growing number of vehicles can already perform lane keeping, adaptive cruise control, automatic parking, and assisted driving in some highway and urban scenarios.

When these capabilities are only exclusive to a small number of car models, consumers will ask when purchasing vehicles:

"Does this vehicle have this function?"

But after over 70% of new vehicles are equipped with it, the question will gradually turn into:

"Is this function actually user-friendly?"

Later on, it may evolve into a more critical question:

"What should we do if a problem occurs?"

This is one of the most critical changes as the intelligent vehicle industry enters the next stage.

The more popular a function is, the less the function list itself can create real competitive gaps.

Instead, the gap lies in the engineering capabilities hidden behind the functions.

Defects of traditional vehicles are visible; problems of intelligent vehicles may be hidden in the code

In the past, when a vehicle recall was launched, it was relatively easy for consumers to understand.

For example, there were hidden risks in the braking system.

A certain component might break.

The airbag had defects.

The fuel pipeline might leak.

Most of these are typical mechanical and hardware problems.

Locate the faulty component.

Confirm the affected batch.

Notify the vehicle owners.

Ask the owners to go to the store for replacement.

The entire recall logic is relatively clear.

However, as vehicles become increasingly intelligent, the nature of problems has changed.

Essentially, a modern intelligent vehicle is no longer just composed of engines, batteries, motors, tires and steel.

It is also a mobile computing device equipped with a large number of sensors, chips, operating systems and algorithms.

A misjudgment by the software may also affect the driving experience.

A recognition deviation of a sensor may also bring safety risks.

An algorithm that works normally in 99.9% of scenarios does not mean that the remaining 0.1% of scenarios are unimportant.

This is exactly the biggest difference between vehicles and mobile phones.

If a bug appears in a mobile app, the worst result may be a crash.

If a bug appears in the vehicle system, it will face real roads, real vehicles and real pedestrians.

As a result, a problem that was rarely encountered in the automotive industry in the past has emerged:

The launch of a vehicle does not mean the end of software development.

On the contrary, it may just be the beginning.

After vehicles are put on real roads, they are faced with different weather conditions, different road conditions and different driving habits, so the system is facing new tests every day.

This is why the intelligent vehicle industry will pay more and more attention to a term that used to sound less appealing in the future:

Defect management.

It does not sound as advanced as terms like "end-to-end", "large model" or "mapless intelligent driving".

But it may determine how solid the real technical foundation of an automaker is.

Behind the 121 million recalled vehicles, the real change is not that "vehicles are getting worse"

By the end of 2025, China had implemented 3,265 cumulative automotive recalls, involving 121 million vehicles.

In 2025 alone, there were 105 new energy vehicle recalls, involving 2.652 million vehicles.

Seeing such figures, people may easily come to an intuitive conclusion:

Are vehicles getting more and more prone to problems?

In fact, this is not a simple equivalent relationship.

The increase in the number of recalls may correspond to several changes at the same time.

First, the number of vehicle ownership and product types is growing larger and larger.

The more vehicles there are on the market and the more complex the car models are, the more products that theoretically need to be included in defect management.

Second, automotive technology is becoming more and more complex.

In the past, the core functions of a vehicle were mainly determined by the mechanical system, but today a large number of electrical and electronic architectures, software systems, chips, sensors and algorithms are added.

The improvement of system complexity itself means that the difficulty of defect management rises.

Third, and more importantly:

The capability to identify problems is constantly improving.

In the past, some problems might not be noticed until a large number of vehicles failed.

Today, automakers can collect operation data through connected vehicles.

After an abnormal situation occurs, they can quickly judge whether there is a common problem in a certain type of car model, a certain software version or a certain batch of components.

Some software problems do not even require the owner to drive the vehicle back to the 4S store.

One OTA upgrade can complete the repair.

Therefore, we cannot simply assume that "the more recalls there are, the worse the quality is".

In the era of intelligent vehicles, what we should really pay attention to is another set of questions:

How long does it take for the enterprise to identify the defect?

How long does it take to confirm the scope of impact?

How long does it take to notify consumers?

Can the problem be solved through OTA?

If OTA cannot solve the problem, how long does it take to complete the offline maintenance?

More importantly, will the same problem happen again?

These are the indicators that truly determine the level of the quality system.

In the future, automakers may need to own a "million-level error correction system"

When the L2 penetration rate is only 10%, an assisted driving problem affects only a small number of vehicles.

When the penetration rate reaches 70% or even 90%, the logic is completely different.

Assume that an automaker sells 1 million intelligent vehicles a year.

A certain software version has an extremely low probability problem.

Even if the occurrence probability is only 1 in 10,000, it may mean that 100 vehicles will encounter the same situation among 1 million vehicles.

If this problem involves safety, the enterprise cannot simply regard it as an "occasional bug".

At this moment, what is really tested is the entire system.

Can vehicle data be uploaded quickly?

Can different accidents and abnormal situations be automatically clustered?

Can engineers judge whether they are caused by the same reason?

Is the number of affected vehicles 10,000 or 1 million?

If an upgrade is needed, can the server support large-scale OTA?

If offline maintenance is necessary, can the national service network undertake the task?

This means that a new capability that used to receive less attention is emerging in the competition of intelligent vehicles:

Large-scale error correction capability.

In the past, one of the most important capabilities of an automotive company was to stably assemble 1 million parts into a vehicle.

In the future, it must also have another capability:

To continuously manage 1 million vehicles that have been sold and are running on the road.

These two capabilities are completely different.

Manufacturing ensures the quality "before delivery".

The real difficulty of intelligent vehicles is that the enterprise still needs to take full responsibility "after delivery".

OTA makes recalls faster, and also makes the responsibility boundary more complex

Intelligent vehicles have another very special feature:

Many problems can be solved through OTA.

This seems to be a huge progress.

In the past, if 100,000 vehicles needed to be recalled due to software problems, all owners might have to go to the store for maintenance.

Now, pushing a new version overnight can theoretically cover a large number of vehicles.

The cost is lower.

The efficiency is higher.

It is also more convenient for consumers.

But OTA also brings a new problem.

If a safety problem can be fixed remotely, should it be regarded as an ordinary software upgrade, or a recall?

If the enterprise keeps fixing functional problems through updates, which ones belong to normal product iteration, and which ones belong to defect rectification?

Do consumers have the right to know?

This is also an increasingly important part of intelligent vehicle governance.

Because OTA cannot become a kind of "invisible maintenance".

For important safety-related defects, consumers need to know:

Where the problem is.

Why the problem occurred.

Which vehicles are affected.

What the upgrade solves.

What to do if the upgrade fails.

The real sign of a mature intelligent vehicle enterprise in the future may not be how frequently the software is updated.

Instead, it can clearly explain why every update is launched.

Consumers should also adopt a new set of indicators when purchasing intelligent vehicles

In the past, when consumers bought vehicles, they usually paid attention to several factors.

Price.

Power performance.

Fuel consumption or cruising range.

Space.

Brand.

Configuration.

After intelligent vehicles become popular, a new list of indicators should be added.

First, check the capability boundary of assisted driving.

What exactly the vehicle can do, and what it cannot do.

In which scenarios the driver must take over the vehicle.

Marketing rhetoric is one thing, and the actual responsibility boundary is another.

Second, check the OTA capability.

Whether the software of an automaker can be continuously updated not only affects the functional experience, but also directly affects the efficiency of defect repair.

Third, check the historical recalls and disposal situations.

It is not simply to compare who has more recalls.

What really matters is whether the enterprise responds fast enough and handles the problem transparently enough after encountering it.

Fourth, check the after-sales service network.

Because not all problems can be solved through OTA.

Hardware problems such as batteries, chassis, braking systems, steering systems and sensors still rely on the offline maintenance system in the end.

The last indicator, which is also the most easily ignored one:

Whether the enterprise is willing to admit the problem.

There is no automotive enterprise that never encounters problems.

Zero defect is almost impossible for a complex industrial system.

What really matters is whether the enterprise takes the initiative to identify problems, disclose problems and fix problems after finding them, or takes action only when the problem cannot be avoided.

These two different quality cultures may lead to huge differences in the long run.

The second half of the intelligent vehicle competition focuses on "what to do after making mistakes"

In the past few years, the most widely spread stories in the intelligent vehicle industry almost all revolve around "what new capabilities the vehicle has mastered".

It can perform automatic parking.

It can support highway navigation.

It can provide urban assisted driving.

It supports voice interaction.

It can automatically find charging piles.

These capabilities are still very important.

But as L2 gradually becomes the mainstream configuration, the real challenging problems of the industry are emerging.

The more functions there are, the more complex the system will be.

The more complex the system is, the more attention should be paid to the probability of edge problems.

The larger the number of vehicles, the more people a small-probability defect may affect.

As a result, the capability boundary of automotive companies is being redefined.

A truly mature intelligent vehicle enterprise should not only prove that:

"How smart my vehicle is."

It should also prove that:

"If the vehicle makes a mistake, how fast can I find it."

"After finding the problem, how fast can I fix it."

"After the repair, can I ensure that similar problems will not happen again."

When the L2 penetration rate exceeds 70%, intelligent vehicles have gone through the stage of "whether the function is available".

The next stage is about reliability.

It is about responsibility.

It is about after-sales service.

It is also about recall management.

In the past, the competition in the automotive industry focused on who could manufacture a vehicle well enough.

In the era of intelligent vehicles, there is a more difficult exam:

After millions of vehicles are already running on public roads, who can continue to manage them properly.

This may be the most noteworthy competitiveness after intelligent vehicles truly enter the mass market.

This article is from the WeChat official account "BT Business & Tech", written by Zhi Yu, and published with authorization from 36Kr.