China EVs: Aion N60 Win, Tesla FSD Wait, GL8 Push

China’s EV and smart-car market delivered three telling signals in late August: GAC Aion’s budget-priced N60 won D1EV’s Guiyang intelligent driving competition with a record 117.81/120 score, Tesla’s Full Self-Driving (FSD) rollout in China remained stuck in regulatory and competitive limbo, and Buick accelerated its electrified GL8 MPV offensive after reaching 2.2 million cumulative GL8 deliveries in Shanghai. Taken together, these developments show how the Chinese auto market is moving beyond simple hardware bragging rights toward software efficiency, localized compliance, and platform-led cost control.

Aion N60’s Win Challenges the “More TOPS = Better ADAS” Narrative

On August 15, at the second D1EV Intelligent Driving Competition in Guiyang, the GAC Aion N60 took the preliminary title with:

• Zero human takeovers
• 117.81 points out of 120
• A new record across 30 competition stops

It then went on to win the final as well.

What makes the result remarkable is the price point. The N60’s top trim sits in the RMB 120,000 class, far below many premium Chinese EVs and smart cars that market significantly more expensive autonomous driving hardware.

N60 hardware: modest on paper, impressive on the road

According to D1EV, the N60 uses:

• Qualcomm 8650 domain controller
• Roughly 200 TOPS of compute
• 1 x 192-line LiDAR
• 2 x 4D millimeter-wave radars

In a market where consumers have been trained to equate:

• bigger models,
• higher TOPS,
• more powerful chips,
• and denser LiDAR

with stronger driver assistance, the N60’s performance is a direct challenge to that logic.

Why a 200-TOPS Car Beat More Expensive Rivals

The key explanation from GAC and WeRide executives is not that hardware no longer matters, but that algorithm architecture and compute utilization matter more than headline specs suggest.

D1EV’s interviews with Xu Wei, head of intelligent driving at GAC Platform Technology Research Institute, and Liu Zhenya, WeRide vice president of technology, point to a central shift: one-stage end-to-end models are changing how assisted-driving systems use resources.

One-stage end-to-end vs. traditional modular stacks

Traditional advanced driver-assistance systems often rely on a layered “model + rules” architecture:

• Perception module
• Prediction module
• Planning/control logic
• Rule-based safety constraints

That approach typically scales by adding more computing power to more separate functions.

By contrast, a one-stage end-to-end model integrates perception, prediction, and planning into a single neural network that outputs the vehicle trajectory directly from sensor inputs.

This brings two important advantages:

1. Better software efficiency
2. Stronger scalability across hardware tiers

WeRide says its strongest model runs on L4 robotaxi hardware using dual Nvidia Thor-X chips, with each chip offering around 5x the compute of the Qualcomm 8650. The N60 uses a distilled version of that same model, adapted to about 200 TOPS.

That distinction matters. A smaller distilled model does not retain 100% of the original capability, but it can preserve a large share of the performance while dramatically lowering hardware cost.

Specs vs. real-world efficiency

The N60’s result suggests that nominal compute is only part of the story. Real-world performance also depends on:

• Model compression quality
• Operator support
• Memory access efficiency
• Heterogeneous scheduling across NPU/CPU/GPU
• Software deployment optimization
• Training data quality

This is why the often-quoted TOPS number is becoming a weaker standalone indicator of autonomous driving capability.

The “Bare-Shell House” Problem: Qualcomm’s Flexibility Comes With Work

One of the more revealing metaphors in the D1EV report compares chip platforms to housing.

• Nvidia: like a furnished apartment, easier to move into
• Qualcomm: like a bare-shell house, offering flexibility but requiring more work

GAC reportedly debated the platform choice internally in mid-2025. WeRide preferred Nvidia, but GAC eventually chose Qualcomm 8650, which meant GAC, WeRide, and Qualcomm had to jointly optimize deployment to fully exploit the chip’s NPU, CPU, and GPU resources.

This is an important insight for the broader Chinese EV industry. If more automakers and software partners learn to maximize heterogeneous compute on lower-cost hardware, the market value of a chip may shift away from raw capability and toward:

• Cost-performance ratio
• Supply stability
• Software ecosystem support
• Engineering adaptability

In other words, the companies that can “decorate the bare-shell house” efficiently may have a major cost advantage in the next phase of ADAS competition.

Data Quality, Not Just Data Volume, Is Becoming the New Battleground

Another major theme from the N60 story is that consistent data can matter more than sheer fleet size.

WeRide openly acknowledges that it does not have the biggest mass-production vehicle fleet compared with players such as:

• Huawei-backed ecosystems
• BYD
• Momenta

But once end-to-end driving enters imitation learning at scale, the challenge is no longer just collecting footage. The challenge is ensuring the model learns the right behavior.

Why robotaxi data may be especially valuable

WeRide argues its robotaxi fleet provides unusually consistent training data because:

• The same system handles similar scenarios in similar ways
• Behavioral variance is lower than in human driving data
• Post-processing costs can be lower

Human driving data often contains conflicting reactions to the same situation. For example, when approaching a slow vehicle, different drivers may brake early, change lanes aggressively, or wait patiently. A model trained on inconsistent behavior needs heavy filtering and scoring to identify the preferred action.

WeRide also highlighted its Genesis cloud world model, which uses edge-case robotaxi data and simulation to increase the density of low-frequency scenarios in training.

That matters because road environments are effectively infinite. Rather than memorizing specific intersections, a capable system should learn reusable action patterns such as:

• reversing to create space,
• adjusting path geometry,
• then completing a multi-point turn.

That reportedly helped the N60 handle Guiyang’s three-point-turn challenge in qualifying.

Platform Strategy Is What Makes Smart Driving Affordable

Even if the software works on lower-cost compute, there is still the question of how intelligent driving drops into the RMB 100,000-120,000 segment.

GAC’s answer is platformization.

Instead of customizing sensors, domain controllers, and software architecture for every model, automakers can standardize hardware across multiple vehicles and spread:

• R&D cost
• Validation and testing cost
• Software adaptation cost
• OTA maintenance cost
• Procurement cost

across much larger volumes.

This is one of the biggest structural trends in the Chinese EV market today. It mirrors what leaders such as BYD have done well with vertical integration and top-level architecture decisions.

Smart driving cost reduction: where the savings come from

The larger takeaway is that China’s intelligent driving race is no longer just about who has the biggest chip. It is increasingly about who can industrialize advanced software cheaply enough to scale into the mainstream market.

Tesla FSD in China: The Real Story Is Delay, Not Retreat

If Aion N60 shows how fast local players are improving, Tesla’s China FSD situation shows how hard it is for a foreign leader to enter an already maturing market.

In recent days, Chinese social media was flooded with claims that:

• Tesla had removed China from FSD’s available-region list
• Its Shanghai data center team had withdrawn
• FSD China approval had effectively failed

According to D1EV, these claims are not supported by the available evidence.

What actually happened

The confusion appears to have originated from a Tesla North America website list updated on August 21, which included only 12 regions where the supervised FSD subscription is available.

China mainland currently offers a one-time purchase option of RMB 64,000, not a subscription, so it would not be expected to appear in that specific list.

Tesla China has also reportedly denied claims that the Shanghai data-center team had been evacuated, calling them false.

This fits a broader pattern: limited official materials are repeatedly being interpreted as major policy signals.

Why FSD China Approval Is Still So Difficult

Tesla has already made meaningful compliance progress in China, including:

• Local data storage via a Shanghai data center
• A Lingang AI training center reportedly operational since February 2026
• Cooperation with Baidu on compliant high-definition mapping
• Passing multiple automotive data-security standards

Yet three major constraints remain.

1. Administrative approval

China requires regulatory procedures for OTA upgrades involving autonomous driving functions under rules tied to the Ministry of Industry and Information Technology and the State Administration for Market Regulation.

Tesla briefly opened a limited FSD trial in March 2025, but paused it after roughly a week. That suggests technical preparation alone is not enough; final permission is still externally determined.

2. Local training compute

Because large-scale data exports are restricted, the China version of FSD needs substantial local training and adaptation. That raises the question of high-end AI chip access at a time when advanced semiconductors sit at the center of US-China technology controls.

Tesla has said current compute is sufficient for present needs, but scaling China-specific training could become a bottleneck.

3. Commercial timing

Even if FSD is approved, Tesla would be entering a market that has changed dramatically. China’s urban NOA and intelligent driving stacks are no longer niche premium features.

Competition now includes:

• Huawei ecosystem players
• Momenta
• Horizon Robotics
• WeRide
• DeepRoute.ai
• OEM self-developed solutions

And perhaps most importantly, Chinese consumers are generally less willing than US buyers to pay separately for expensive ADAS software. Many automakers simply bundle advanced assisted-driving functions into the vehicle price.

Tesla’s China FSD Timeline: Expectations Kept Slipping

One reason the rumor cycle remains intense is that Tesla’s China FSD timeline has moved repeatedly.

Reported timeline drift

D1EV’s key point is that these statements often reflected Tesla’s own expectations, not confirmed regulatory decisions.

That distinction is critical for investors and consumers alike.

Buick GL8 Shows the Other Side of China’s NEV Market

While headlines often focus on flashy autonomous driving battles, Buick’s latest GL8 move is a reminder that China’s auto market is also being reshaped by practical electrification in high-volume family segments.

On August 26 in Shanghai, Buick held a ceremony marking delivery of its 2.2 millionth MPV, with Chinese sprint star Su Bingtian becoming the symbolic 2.2 millionth GL8 owner.

Key GL8 milestones

• First Buick GL8 rolled off the line in 1999
• 2 millionth vehicle rolled out in September 2024
• 2.2 million cumulative users reached in less than a year after that milestone
• More than 80% of current GL8-family users are families, according to Buick

That pace is notable. It suggests the GL8 remains highly resilient even as China’s market shifts toward electrified and intelligent vehicles.

Buick’s New GL8 Energy Strategy

Buick also launched its GL8 “Super Refresh Season” and laid out a broader new-energy product matrix spanning the RMB 200,000, 300,000, and 400,000 price bands.

Buick GL8 lineup and pricing

Buick says discounts in the current campaign reach as much as RMB 55,000 relative to official guide prices, including trade-in subsidies and promotional incentives.

Durability and battery messaging

Buick is also leaning heavily on engineering credibility:

• Every GL8 reportedly completes two winters and two summers of validation
• Testing covers 120+ operating conditions
• Total durability testing reaches 6.5 million km
• Ultium-platform battery fleet data has accumulated 5.6 billion km
• Buick claims a zero self-ignition record for those batteries

That messaging targets a different but equally important Chinese market reality: buyers increasingly want electrification, but they also demand proven durability and family-oriented reliability.

Why This Matters

These three stories are connected more closely than they first appear.

1. China’s smart-driving market is entering an efficiency phase

Aion N60’s victory suggests the next battleground is not just compute scale, but:

• software-hardware co-optimization,
• model distillation,
• data consistency,
• and platform economics.

That is a major shift from the older narrative of spec-sheet supremacy.

2. Tesla no longer has the luxury of time

Even if Tesla FSD wins approval in China, the company may arrive in a market where local competitors already offer capable urban NOA-like functions at far lower effective user cost.

The longer the delay, the smaller the disruptive effect.

3. Electrification in China is broadening beyond sedans and SUVs

Buick’s GL8 push shows how quickly new-energy competition is moving into MPVs, a category increasingly important for families and premium multi-use buyers. Range, battery safety, and product segmentation are now as important in MPVs as they are in mainstream EV crossovers.

Global Implications

For global automakers and suppliers, China is again acting as the industry’s stress test.

• For ADAS developers, it shows that elegant software can offset lower-cost hardware better than many expected.
• For Tesla, it underlines that localization is about regulation, compute, pricing, and competitive fit—not just technical readiness.
• For legacy global brands, Buick’s GL8 strategy shows that strong nameplates can still be revitalized through electrification and sharper price-band positioning.

China’s market is effectively defining three future rules for the global auto business:

1. Affordable intelligence beats expensive excess
2. Local compliance is a product feature, not a back-office task
3. Platform scale is the only way to make advanced technology mainstream

What Comes Next

The next few months will be worth watching on three fronts.

First, the industry will want to see whether the Aion N60’s competition performance translates into broader consumer recognition and whether more OEMs adopt similar distilled end-to-end ADAS architectures on lower-cost Qualcomm or other mass-market compute platforms.

Second, Tesla’s China story will likely hinge less on rumor-driven website interpretations and more on actual regulatory signals. Any movement after key diplomatic or policy milestones could reset sentiment, but the company’s window for maximum impact is narrowing.

Third, Buick’s GL8 family will serve as a useful barometer for whether electrified MPVs can combine volume, pricing power, and family adoption at scale in China’s increasingly crowded new-energy market.

In short, China’s auto industry is showing that the future of smart mobility will not be won by the biggest hardware stack alone. It will be won by the companies that can turn software, supply chain discipline, and platform scale into products ordinary buyers can actually afford.