NIO’s Chip Unit GeniTech: Turning Profitable Yet？

A car company unwilling to become a chip company is not a true technology company.

The weight of this statement continues to rise in China’s auto industry today.

At the 2026 World Artificial Intelligence Conference (WAIC), where ticket prices on the secondary market were bid up from RMB 168 ($25) to as high as RMB 1,500 ($221), a familiar name appeared — GeniTech.

GeniTech, NIO’s chip division, made its first appearance at WAIC as an independent brand, bringing its core products including the GeniTech NX9031 series, the RUDO Embodied Intelligence Development Kit (RUDO RDK), and the STAR X08 Domain Controller.

Following its battery-swapping business, NIO’s chip division is also beginning to “walk independently.”

Since late 2024, NIO founder William Li has maintained a positive view toward embodied intelligence, calling it an inevitable direction for automakers. However, he has also said NIO will remain in a wait-and-see mode for an extended period.

The reasons are straightforward: on one hand, NIO’s automotive business has yet to achieve long-term stable profitability; on the other hand, the robotics industry has yet to reach scale.

Rather than entering the market today, NIO is more interested in whose robots will use its chips. GeniTech’s independent debut represents NIO’s current exploration of the embodied intelligence industry.

What possibilities did GeniTech showcase at WAIC? How does its approach differ from automakers such as Li Auto, XPeng, and Geely? Let’s take a closer look.

GeniTech Expands Its Reach

The most important information disclosed by GeniTech is that the high-end NX9031X chip under the GeniTech 9031 series has shipped more than 200K units in total. (However, most media reports cited the figure as 300K units.)

The number may seem insignificant compared with smartphone chips, but in automotive-grade semiconductors, NX9031X has proven it is more than a “One More Thing” product unveiled at a launch event. The chip has entered large-scale validation, making it one of the few self-developed automotive chips in China to achieve commercial-scale deployment.

NIO founder William Li previously revealed in April that when NIO relied on NVIDIA chips in previous years, the company spent up to $300 million annually on chip purchases. Based on current exchange rates, that equals about RMB 2.03 billion ($300 million). With NIO targeting annual sales growth of 30%-40%, chip procurement has become a significant cost pressure.

The arrival of NX9031X could reduce NIO’s cost per vehicle by around RMB 10,000 ($1,500). At the same INT8 precision level, based on NIO’s previous claim that “one GeniTech chip equals four Orin-X chips,” the chip’s theoretical computing power is estimated to exceed 1,000 TOPS. By comparison, NVIDIA Thor-U is rated at 700 TOPS. Its bandwidth reaches 546GB/s, also surpassing Thor-U.

Although Thor-U still holds advantages in software ecosystem, 4nm process technology and floating-point computing performance, NX9031X’s lower procurement cost and computing advantages give NIO more possibilities. More importantly, self-developed chips allow NIO to control its own technology roadmap.

This capability is not only applied to vehicle development cycles, but can also be reused in emerging fields such as embodied intelligence. That has led to the development of the NX9031U, a mid-range chip designed specifically for embodied intelligence, as well as the entry-level general-purpose AI chip NX9031C.

NX9031X provides the foundation for NX9031U to understand the physical world. It also brings large-scale production experience. The two chips share the same 5nm automotive-grade SoC architecture. With more than 300K GeniTech series chips shipped, validated algorithm models, software maturity and adaptability can be reused for robotics applications to the greatest extent.

However, robots have different requirements from vehicles in areas such as latency, thermal management and cost control. As a result, NX9031U lowers its maximum equivalent computing power to 800 TOPS. The chip adopts air cooling, retains a heterogeneous computing architecture, and supports multimodal real-time inference, making it better suited for robotics scenarios.

If NX9031U is the “brain,” then the STAR X08 Domain Controller serves as the “body and central nervous system.” The STAR X08 Domain Controller is powered by the GeniTech NX9031U chip. It features a 28-core ARM CPU, 64GB LPDDR5x memory, and 546GB/s bandwidth, providing up to 800 TOPS equivalent computing power. It also supports 4K@450fps video encoding.

According to GeniTech, a single NX9031U chip can complete the full computing chain covering perception, reasoning, planning and control.

Connecting the “brain” with the “cerebellum” is the RUDO RDK development kit.

According to GeniTech’s description, its self-developed TypeI Hypervisor enables hardware-level isolation of computing resources, dividing the system into real-time control domain and AI inference domain. It supports flexible partitioning of CPU, GPGPU and memory resources, ensuring that high-load large-model inference does not interfere with the real-time motion control channel, thereby safeguarding robotic motion safety.

It supports the Ubuntu 24.04 LTS and ROS2 Jazzy ecosystems, with a built-in dedicated inference engine for large models, compatible with CUDA and commonly used BLAS and Solver libraries. It has already been adapted to mainstream VLA and world models including π0.5, OpenVLA, and RT-2.

For developers accustomed to the NVIDIA CUDA ecosystem, switching to the GeniTech platform requires almost no relearning, “truly achieving interface compatibility and out-of-the-box usability.”

At this point, GeniTech’s basic strategy in embodied intelligence has become clear. NIO’s latest expectation for GeniTech may have evolved from “cost reduction” to “self-sustaining growth.”

NIO’s Ambition

GGeniTech is currently one of the few chip units among China’s new EV makers to become an independent company. The current landscape of in-house automotive chip development is complex, with companies generally falling into four categories.

The first group focuses on internal use only, such as Tesla. The second group works through technology licensing or joint development, such as XPeng and Volkswagen. The third group starts with internal applications before seeking external supply, such as Li Auto’s Xinchang Zhixin. The fourth group consists of third-party suppliers that grew out of the automotive industry, such as ECARX.

Nio’s GeniTech sits between several models. It was spun off from Nio Group as an independent company, accepts market-based financing, and is working toward becoming a supplier.

These companies that export chips or plan to do so are not directly challenging Nvidia or Qualcomm. Instead, they are offering complete technology platforms. XPeng’s VLA system and Turing AI chip have secured Volkswagen’s supplier nomination. Nio is offering a platform covering chips, domain controllers, and RUDO RDK development kits.

Tesla’s approach carries unique advantages. For Chinese companies separating chip businesses from automakers, the logic may follow a classic saying: during a gold rush, the shovel seller often makes the most reliable profits.

In China’s NEV market, the company with consistently strong profitability is CATL. In the broader smart vehicle and AI era, the biggest winner has been Nvidia. Chinese automakers also want to become the “shovel sellers.”

The deployment of autonomous driving chips in production vehicles gives automakers massive real-world data. This data feeds back into chip iteration, creating an advantage that traditional suppliers struggle to replicate.

According to official figures, NX9031X has shipped more than 200K units. The chip is using Nio and Onvo vehicles to absorb early investment costs. The next step is to push this ecosystem into more application scenarios during the AI boom.

NX9031X, NX9031U, and NX9031C represent Nio’s bet on future growth. Nio itself is unlikely to directly enter robot manufacturing in the short term. However, it is positioning itself further upstream, using a more fundamental approach to prepare for the sector.

The mission of GeniTech is to become a company the robotics industry cannot operate without. Even if it cannot achieve the dominance of CATL or Nvidia, GeniTech aims to become one of the key players.

Challenges remain. The chip market has a strong winner-takes-all effect. Nio’s annual sales volume is still far from competing with Nvidia or Qualcomm. Convincing other companies to adopt its solutions will also require overcoming trust concerns.

Capital markets have already shown confidence in GeniTech. Since its establishment in June 2025, it has secured nearly RMB 3B ($414M) in financing. Morgan Stanley noted that GeniTech’s business expansion could become a key factor in reshaping Nio’s valuation. When NX9031 series annual shipments exceed 1M units, the chip business could shift from a “burden” into a profit contributor.

From PPT narratives to broader commercial possibilities, Nio, XPeng, and Li Auto have passed their most challenging phase. But the test has not disappeared. Who will succeed in becoming the “shovel seller”? The answer may depend on the scale-up of the robotics industry and the choices made after L3 regulations are introduced.