On the afternoon of October 19, the RISC-V China Summit kicked off in Shenzhen.
At this event, Alibaba DAMO Academy unveiled the XuanTie E908A, an automotive-grade enhanced processor, which is the first RISC-V CPU IP to simultaneously achieve ASIL-B and ASIL-D compliance.
Many people immediately asked Vehicle: Does this mean Alibaba is going to make automotive chips?
This article will not discuss how popular RISC-V is. Vehicle wrote about "New Opportunities for RISC-V in Automotive Electronics" a few years ago. However, it is only now that some major players are making moves. Therefore, this article deeply explores what exactly Alibaba has released with this core, who in the automotive industry will be affected, and what to focus on next.
1. First, Clarify What Was Released: It's Design Blueprints, Still Three to Five Years Away from Hitting the Road
Let's first sort out the automotive-related releases from Alibaba this time:
•XuanTie E908A: An automotive-grade enhanced processor covering both ASIL-B and ASIL-D, with a Single-Point Fault Metric (SPFM) exceeding 90%, targeting safety islands, domain controllers, and vehicle gateways.
•C950A: The next-generation product in the newly unveiled autonomous driving and embodied intelligence roadmap, targeting a performance 5 times that of existing RISC-V automotive-grade products.
•C950: The server flagship released in March. This time, the focus is on its progress in agentic workloads and large model inference, which also serves as the foundation for the C950A.
There is a premise here that is easily overlooked.
DAMO Academy sells CPU IP, which is the processor design, licensing it to chip companies for use. It does not produce chips itself.
To use an analogy, IP is like a set of architectural blueprints. Chip companies buy the blueprints to build the building, which is tape-out; then they hire people for inspection, which is automotive-grade certification; finally, automakers take delivery, which involves design wins and mass production. We also shared this in our previous article "Automotive Chip Industry Chain 101".
How long does this entire process take?
Li Weili, a processor architect at DAMO Academy himself, stated at an automotive chip conference this year: It typically takes three to five years for an automotive-grade chip to complete functional safety and other certifications.
Therefore, my judgment is that the E908A will not appear in any vehicle in the short term.
What it will change first is the core selection list on the project initiation table of chip companies next year.
2. Where Will It Go in the Vehicle? Safety Islands, Zonal Controllers, and Gateways Are Multiplying
Alibaba's official documents list three scenarios for the E908A: safety islands, domain controllers, and vehicle gateways.
These three positions happen to be exactly where the number of chips is increasing during the generational shift in electrical and electronic (E/E) architectures.
01 Safety Island: The Duty Officer in Autonomous Driving Chips
In current autonomous driving chips, the main computing power handles perception, planning, and running end-to-end models. It is fast and complex, making it impossible to guarantee 100% error-free operation.
Therefore, a small core is placed separately within the chip, with independent power supply and independent clock, specifically monitoring the main computing power. If the heartbeat of the main computing power stops or the output is incorrect, it takes over, causing the vehicle to degrade, pull over, or stop.
This is the safety island. It does not require high computing power but has the highest functional safety requirements, typically ASIL-D.
In the past, this position was basically the territory of Arm real-time cores. Now RISC-V has entered: Mobileye's EyeQ Ultra integrates 12 RISC-V cores on-chip, and Dipingxian Journey 6 series also partially uses RISC-V.
02 Zonal Controllers and Gateways: Multiplying in Number
Another change comes from the architecture.
An architecture executive at Shangqi stated at a chip conference this year that 2026 to 2028 will be the three years when domestic automakers collectively transition their E/E architectures. It is expected that by 2027, the penetration rate of vehicle models adopting a central computing plus zonal control architecture will exceed 30%.
The new architecture typically consolidates the dozens of distributed controllers of the past into one central computing platform plus several zonal controllers. Each zonal controller must manage lights, doors, seats, and chassis actuators, requiring high real-time and medium-to-high safety-level control cores.
I think DAMO Academy has chosen a very accurate position this time.
The E908A is not rushing to compete for the most challenging autonomous driving main computing power; instead, it is first guarding those positions that are multiplying and must be sufficiently safe.
03. Who Will Be Affected? Mid-Tier Chip Companies and Automakers Developing In-House Chips Benefit the Most
Putting the E908A back into the context of the entire industry, the impact needs to be discussed across several categories of players.
Beneficiary 01: Mid-Tier Automotive-Grade MCU and SoC Companies
About 70% of the high-end automotive-grade MCU market share has long been held by NXP, Infineon, and Renesas. For domestic chip companies wanting to make ASIL-D chips, the core options are either licensing Arm, writing it from scratch, or using domestic RISC-V cores.
Developing an in-house core requires maintaining a large team. Nuclei System Technology developed the core for Dongfeng DF30, expanding the team from about 20 to around 500 people, and conducted over 600 tests for the chip.
For companies that cannot afford an in-house core team but want to avoid Arm licensing constraints, having an additional safety core produced by a major player is a solid option.
Beneficiary 02: Automakers Developing Their Own Autonomous Driving Chips
Lixiang, Xiaopeng, and Weilai have all been developing in-house autonomous driving chips in recent years, and such chips typically require a safety island.
The E908A provides these teams with an additional ready-made solution to replace Arm. In a context where export controls and licensing terms could change, having a backup is valuable in itself.
Under Pressure: Arm and Domestic RISC-V Peers
Arm's position in safety islands and automotive-grade real-time cores will lose some market share.
On the domestic front, Nuclei System Technology has already tied up with Dongfeng, with the first batch of mass-produced vehicles rolling off the line by the end of this year; Guoxin's RISC-V multi-core ASIL-D MCU has already been sampled, and its automotive chip revenue grew 108% year-on-year in the first half of this year. XuanTie is entering the market a step later than them.
Its trump card is scale: XuanTie processors have shipped over 6 billion units cumulatively, it leads the Wujian Alliance, and it collaborates with SGS on certification services.
3. International Giants Are Still on the Way
Overseas, Infineon announced its RISC-V automotive-grade MCU family in March 2025, integrating it into the AURIX brand. However, a chip executive at Chang'an Automobile said it will be mass-produced no earlier than 2028.
Quintauris, a joint venture of NXP, Bosch, Infineon, Qualcomm, and Nordic, is still developing a unified automotive RISC-V reference platform.
Therefore, domestic RISC-V automotive-grade chips have a first-mover window of about two to three years.
Of course, automakers choosing RISC-V are not primarily looking to save on licensing fees. The licensing fee for CPU IP in a single MCU accounts for a very small proportion. The real reasons are supply chain controllability, the ability to add custom instructions, and better energy efficiency and area.
Caliber: Data referenced by Dongfeng during selection shows that under the same computing power, RISC-V power consumption can be reduced by about 70%, and chip area can be reduced by up to 60% to 70%. Specifically for each chip, it varies by design.
The real bottleneck is software: AUTOSAR driver adaptation, certified compilers, and whether automakers are willing to provide mass production opportunities.
4. C950A Is the Bigger Game: 5x Sounds Impressive, but the Base Is Actually Not High
In this release, I think the roadmap actually carries more weight.
The E908A guards control and safety, while the C950A targets computing: the main CPU for autonomous driving domain controllers and embodied intelligent robots.
Its foundation is the C950 released in March this year: with a maximum clock frequency of 3.2GHz, SPECint2006 single-core score exceeding 70 for the first time, and an out-of-order window of over 1,000 instructions.
At this summit, DAMO Academy also mentioned that it can run Qianwen3.8-Flash-Next at 43 tokens/second.
It is compared against existing RISC-V automotive-grade products. Existing automotive-grade RISC-V cores are mostly MCU-level control cores, so the baseline itself is not high.
However, this roadmap did not provide a timeline, process node, or clock frequency, nor did it specify the functional safety level achieved.
In the position of the main CPU for autonomous driving, Arm's automotive application processor cores (AE series) have already entered mass production projects. The C950A will be competing against this tier of opponents.
My judgment is that the C950A currently looks more like a letter of intent. Its highlight is that Alibaba has the opportunity to integrate "CPU core + Qianwen model + cloud," allowing smart cockpit large models and robots to reuse the same set of CPU and AI extensions.
Whether this path can be successfully navigated depends on whether any automakers or Tier 1 suppliers are willing to use mass production projects for validation. This is not yet visible.
Control first, then compute—this sequence is correct; when the second step will be taken remains to be seen.
5. In Conclusion: From "Usable" to "Reliable," Several Proofs Are Still Missing
In his keynote speech, Meng Jianyi said the goal is to push high-performance RISC-V from "usable" to "reliable".
Applied to the automotive sector, this statement is very apt. The E908A shows that DAMO Academy wants to enter the automotive-grade market and has found the right position; however, "reliability" must be proven by certifications, customers, and mass-produced vehicles, none of which Alibaba currently has.
Therefore, Alibaba's RISC-V entering vehicles still has a long way to go.
That concludes today's sharing. If you find it useful, please help by liking, sharing, and clicking "Wow," and welcome to discuss your views in the comment section.
*Reproduction and excerpting are strictly prohibited without permission. To obtain the reference materials for this article: join our Knowledge Planet to download a massive amount of reference materials from our official account, including the above references.