NIO Defines Four Core Standards for Advanced AI Chips
💡Learn how NIO is redefining automotive AI chip metrics to combat industry-wide performance inflation.
⚡ 30-Second TL;DR
What Changed
Criticized industry 'paper performance' and marketing inflation
Why It Matters
This sets a new benchmark for automotive AI hardware, pressuring competitors to move beyond theoretical TOPS toward real-world performance metrics.
What To Do Next
Evaluate your AI inference hardware based on memory bandwidth and real-world latency rather than just peak TOPS.
Key Points
- •Criticized industry 'paper performance' and marketing inflation
- •Shenji NX9031 features 546GB/s bandwidth and 5nm process
- •Mass production has already exceeded 250,000 units
- •Focus on real-world scheduling and multi-chip coordination
🧠 Deep Insight
Web-grounded analysis with 27 cited sources.
🔑 Enhanced Key Takeaways
- •The Shenji NX9031, first unveiled at NIO Day 2023 and deployed in the ET9 sedan in March 2025, features a 32-core CPU architecture with over 50 billion transistors.
- •NIO's chip subsidiary, Anhui Shenji Technology, secured over RMB 2.2 billion (approximately $320 million USD) in its first equity financing round in February 2026, valuing the unit at nearly RMB 10 billion (approximately $1.4 billion USD) post-investment.
- •NIO has begun licensing its Shenji NX9031 technology to other automotive chip companies, marking its first external commercialization and aiming to reduce its annual spend on external chips, which was up to $300 million on NVIDIA in 2024.
- •Beyond the NX9031, NIO also developed the Yangjian LiDAR control chip, with cumulative production of all in-house chips (Yangjian and NX9031) exceeding 550,000 units by March 2026.
- •The chip is designed to process data from 25 high-definition cameras simultaneously with a processing latency of less than 5 milliseconds, supporting advanced Image Signal Processing (ISP) for superior image clarity in low-light conditions.
📊 Competitor Analysis▸ Show
| Feature/Chip | NIO Shenji NX9031 | NVIDIA Drive Orin-X | NVIDIA Drive Thor-U | Xpeng Turing | Li Auto Mach 100 (Dual) | Tesla HW4 |
|---|---|---|---|---|---|---|
| Process Node | 5nm | 7nm (Orin) | 4nm (Thor) | 5nm | 5nm | N/A (4/5nm rumored) |
| Transistors | >50 billion | 17 billion (Orin X) | N/A | N/A | N/A | N/A |
| CPU Cores | 32 (Cortex-A78AE & A65AE big.LITTLE) | 12 (Cortex-A78AE) | N/A | 40 (Cortex-A78) | N/A | N/A |
| Memory Bandwidth | 546 GB/s (LPDDR5X, 512-bit) | 205 GB/s (LPDDR5) | 273 GB/s (LPDDR5X) | ~273 GB/s (LPDDR5X, 256-bit rumored) | N/A | ~448 GB/s (GDDR6) |
| Computing Power | ~1,000 TOPS (sparse, single chip) | 254 TOPS (Orin X) | ~700 TOPS (dense, single chip) | ~750 TOPS (single chip) | 2,560 TOPS (total, dual chips) | ~720 TOPS (dense) |
| Safety Standard | ISO 26262 ASIL-D | ISO 26262 ASIL-D | ISO 26262 ASIL-D | Cortex-R52 safety cores | N/A | N/A |
| First Deployment | ET9 (March 2025) | Nio NT2.0 models (prior to NX9031) | Li Auto L Series Smart Refresh Edition (May 2025) | G7 Ultra (July 2025) | Li Auto L9 Livis (Q2 2026) | N/A |
| Cost Savings | ~10,000 RMB/vehicle | N/A (high gross margin for NVIDIA) | N/A | ~1,400 USD/vehicle (BOM optimization) | N/A | N/A |
🛠️ Technical Deep Dive
- Manufacturing Process: Built on a 5nm process technology, with Samsung Foundry being a likely manufacturing partner due to capacity and cost considerations.
- Transistor Count: Integrates over 50 billion transistors, significantly more than NVIDIA's Orin X.
- CPU Architecture: Features a 32-core CPU configuration, utilizing ARM Cortex-A78AE and Cortex-A65AE cores in a big.LITTLE layout for balanced performance and power efficiency.
- Memory Subsystem: Employs LPDDR5X memory operating at 8.5 GHz (8,533 Mbps) with a high memory bandwidth of 546 GB/s, achieved through a 512-bit LPDDR5X interface.
- Computing Performance: A single Shenji NX9031 chip delivers approximately 615 kDMIPS of CPU performance and around 1,000 TOPS of sparse computing power.
- Image Processing: Includes a high-performance Image Signal Processor (ISP) capable of processing 6.5 billion pixels per second (6.5 G Pixel/s) with a processing delay of less than 5 milliseconds, enabling superior image clarity, especially in low-light conditions. It supports data input from up to 25 high-definition cameras.
- Safety Features: Conforms to the ISO 26262 ASIL-D safety standard, the highest level for automotive safety, and incorporates Cortex-R52 safety cores along with redundant safety circuits for real-time monitoring, automatic error correction, and fault isolation.
- Sensor Fusion: Designed to process data from various sensors, including LiDAR, alongside cameras and radar.
- System Integration: Works in conjunction with NIO's in-house developed SkyOS full-domain vehicle operating system, which integrates smart driving, smart cabin, and chassis control.
- Multi-chip Coordination: The flagship ET9 sedan utilizes two NX9031 chips to achieve approximately 2,000 TOPS of sparse computing power and millisecond-level hot redundancy. Other models, like the new "5566" series with the Cedar-S system, use a single Shenji chip combined with a Qualcomm SA8295P on a shared domain controller for around 1,000 TOPS.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (27)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
- Google Search Source
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 36氪 ↗