The wave of smart electric vehicles has been sweeping the market for several years, but 2024 has marked a turning point and differentiation.

On the operational level, some manufacturers have merged brands to cut costs and improve efficiency; however, other companies have taken the opposite approach, further expanding their market presence. Technologically, the dominance of rule-based algorithmic autonomous driving is gradually waning, making way for end-to-end large models. This shift involves not only significant departmental restructuring but also raises higher demands on the strategic vision of leaders.

"End-to-End" Sees Massive Growth

Event Overview: Since the beginning of 2024, the technological direction of smart driving has shifted from "rule-exhaustive approaches" to "end-to-end solutions." The latest addition to the end-to-end movement is Zeekr, which announced its debut of the end-to-end Plus architecture at the Guangzhou Auto Show, incorporating Zeekr’s digital precognitive network into the end-to-end large model. As such, Tesla, "Wei Xiaoli," Xiaomi, Zeekr, Mercedes-Benz, and even joint venture automaker GAC Toyota have joined the end-to-end camp.

Commentary: Before the emergence of "end-to-end" solutions, autonomous driving development faced an insurmountable challenge. Modulated autonomous driving based on rule algorithms made system integration very complex due to the separate development and optimization of perception, planning, and decision-making modules. Writing rules for autonomous driving required addressing all corner cases. For instance, one algorithm is needed for a person suddenly appearing at the roadside, and another for a plastic bag blowing into the roadway, but it’s impossible to exhaustively cover all road scenarios. Thus, such modular autonomous driving solutions can only come closer to covering all scenarios but will never reach 100%.

End-to-end solutions are different; they do not rely on code for machines to execute rigidly but rather "educate" the model. For example, an end-to-end model is like teaching a child that jumping from a one-meter high step could be harmful, and in turn, the child will apply this knowledge. When faced with a step of 90 centimeters, the child will consider whether to jump, and may decide to jump from a height of 50 centimeters because they understand the world. However, a rule-based child, told that one meter is dangerous, would jump without hesitation from a height of 99 centimeters, as the rule dictates "no jumping above 1 meter."

However, end-to-end is not a cure-all; it has a "high ceiling but low floor." Using the child analogy again, once the child understands the world, they may perceive that one meter isn't that high, and they might jump from heights exceeding one meter. A rule-based child, on the other hand, would never behave this way.

In this context, the intelligent driving sector is witnessing significant changes: engineers who previously had to write numerous algorithms and exhaust scenarios no longer need to do so, as the volume of code dramatically decreases. The C++ code for Tesla's FSD beta V12 version contains only 2,000 lines, while version V11 had 300,000 lines.

On the flip side, end-to-end is a "black box." Although it can be educated and will output results based on its understanding of the world, nobody knows how its neural network understands this world, nor can anyone control it. Therefore, algorithm engineers are still needed to write the foundational rule codes to act as a safety net against end-to-end models demonstrating unexpected behaviors like "jumping from a height of 1.1 meters."

Currently, many automakers are undergoing significant adjustments in their intelligent driving engineering teams. Autonomous driving algorithm engineers who can transition to end-to-end approaches are rapidly adapting, while others risk being optimized out. Even when these engineers transition, their roles are primarily focused on setting baseline rules for end-to-end models rather than being central positions.

The wheels of time continue to roll forward. Who could have imagined that just two years ago, algorithm engineers earning an average annual salary of one million are now facing job insecurity?

NIO's Third Brand Firefly Launched

Event Overview: On the evening of November 20, NIO officially named its third brand "Firefly" and announced that its first product would carry the same name. Firefly targets the premium compact car market, further enriching NIO's product lineup and efficiently contributing to sales through existing distribution channels. Within NIO's structure, Firefly is positioned similarly to BMW's MINI brand.

Commentary: In the fiercely competitive Chinese automotive market, many enterprises have begun to "retract," with examples like Feifan being absorbed back into Roewe and Lynk & Co merging with Zeekr, driven by a desire to eliminate internal strife, integrate resources, and reduce costs. In contrast, NIO is expanding outward, launching not only a second brand, Elle, but also promptly announcing a third brand, Firefly. At first glance, this seems counter to the prevailing trend, but it actually aligns with NIO's strategic considerations.

From a market positioning perspective, NIO covers the segment above 300,000 yuan, Elle targets around 200,000 yuan, while Firefly focuses on premium compact cars priced in the range of over 100,000 yuan. The overlapping demographics between these three groups are minimal, making it ineffective to cover different audiences under one brand. In this era where "the market needs to know who you are," if a brand's image becomes blurred, it risks being lost in the crowd. From this viewpoint, creating three distinct brands makes sense for NIO.

From a powertrain type angle, both NIO and Elle are nearly synonymous with pure electric, as Li Bin has painted a grand vision for a widespread battery swap network. These two brands can be seen as NIO's optimal vehicles for promoting its battery swap philosophy. Meanwhile, Firefly is reportedly set to offer an extended-range version, laying the groundwork for NIO's further growth in volume.

Public data shows that NIO sold 20,976 vehicles in October, maintaining a solid performance. Nonetheless, a glance at the sales rankings reveals that crossing the 20,000-unit mark has become a new threshold for mainstream new forces, as brands like Zeekr, Deep Blue, Xiaopeng, and Xiaomi have all surpassed this figure, while Leapmotor, HarmonyOS, and Li Auto consistently hover between 40,000 and 50,000 units.

However, the "gold content" of NIO's 20,000 sales is notably high, primarily due to its relatively elevated transaction prices; moreover, it exclusively sells pure electric vehicles, managing to sustain this volume even amidst a slowdown in pure electric growth and a surge in extended-range models, indicating a stable market position.

Going forward, the third brand, Firefly, might become the main force driving NIO's sales to aspire for 30,000, 40,000, or even 50,000 units per month.

However, the segment for vehicles priced in the tens of thousands is currently the most fiercely contested subset of the market, with competitors like Volkswagen's ID.3 and Xiaopeng's MONA M03 being strong players here, one promoting big-brand quality and the other emphasizing high-cost-performance intelligent vehicles. How NIO shapes the image of the Firefly brand will heavily test its marketing expertise. Another question arises: will NIO's adept user service system be maintained in the lower-tier Firefly brand as well?