The battle for advanced autonomous driving has yet to begin.Author | Wang RuiEditor | Xi Zi
“We have achieved and even surpassed the most advanced chips made with 4nm technology using a 7nm process.”
At the China Electric Vehicle 100 Forum (2025), Shan Jizhang, founder and CEO of Black Sesame Intelligence Technology Co., Ltd., presented a PPT comparing model efficiency, ResNet at 50 FPS, NVIDIA Orin at 56 FPS, and the newly optimized Thor with 770T computing power, which is expected to reach 300 FPS, while the Black Sesame A2000 achieved 328 FPS.
In a subsequent interview, Black Sesame’s CMO Yang Yuxin added, “Based on our latest generation of NPU architecture for ASICs, the A2000 not only matches the international giants in absolute numbers but also provides higher efficiency.”
How can it achieve performance comparable to NVIDIA? What is the technical gap between Chinese chip manufacturers and NVIDIA? When will mass production and deployment break the dominance?
At the forum, Shan Jizhang and Yang Yuxin provided answers one by one.
A2000: A Single Chip Achieving Full-Scene UrbanNOA+
During the presentation, Shan Jizhang mentioned a concept— “Generalized Intelligent Driving”, which can be simply understood as using a computational brain to deploy various multimodal large models.
In Black Sesame’s design, the A2000 is this computational brain.
“By 2026, generalized intelligent driving large models will achieve large-scale mass production and deployment.” Shan Jizhang introduced that by then, vehicles will automatically park from home parking spaces, exit parking lots, enter urban roads via ramps, navigate through various complex scenarios with mixed traffic, and finally park in company parking lots. This complete set of full-scene intelligent driving functions can be achieved with just this one chip.
One cannot help but marvel, has Chinese chip technology reached this level?! So how does the A2000 achieve this?
In terms of computing power, for different levels of autonomous driving needs, the A2000’s computing power can be expanded from 256 TOPS to 1000 TOPS. In comparison, the NVIDIA Orin-X chip, which has the highest market share for advanced autonomous driving scenarios, has a single computing power of 254 TOPS.
The A2000 is based on Black Sesame’s self-developed “Jiushao” NPU architecture, which can execute complex neural network models more efficiently, enhancing data throughput and computational efficiency, achieving the best balance between performance, bandwidth, and cost.
The “Jiushao” NPU architecture supports data flow optimization under different computational combinations, avoiding computational bottlenecks, and can better adapt to the rapid evolution of end-to-end intelligent driving algorithms.
At the same time, this is also the industry’s highest safety-rated NPU, capable of avoiding random errors and failures during model inference, supporting consistency in training and deployment, ensuring high safety and determinism for autonomous driving systems.
The “Jiushao” NPU also supports mixed precision computing, including INT8/FP8/FP16, which can better adapt to mainstream models such as Transformer and MoBA, improving the efficiency and performance of neural network processing, achieving full-stack capabilities for urban NOA+ data closure with a single chip.
Shan Jizhang introduced that besides intelligent driving, this chip can also be used for various general computing applications, including edge applications, embodied robots, and inference cards. With its 96-channel video parsing capability, the A2000 can also empower smart transportation, security monitoring, and other scenarios.
Yang Yuxin stated, “Our entire technical logic is quite similar to Tesla. Tesla integrates the entire SIT system into the Optimus robot, which is why we believe that robots, especially humanoid robots, will grow within the automotive industry chain due to their high reusability.”
The battle for advanced autonomous driving has yet to begin
Looking back over the past few years, Black Sesame Intelligence has achieved remarkable results in intelligent driving, especially in the field of advanced autonomous driving.
As Shan Jizhang said, “We can proudly say that in traditional Chinese independent passenger car brands, we rank first in both the production volume of car manufacturers and the number of models produced.”
It is reported that its Huashan series A1000 chips have been deployed in numerous domestic models, covering over 40 manufacturers including BYD, Geely, and Dongfeng, and are still continuously being adopted in designated models.
However, looking at the entire market, in the past two years, compared to suppliers like NVIDIA and Horizon, Black Sesame still has some gaps in shipment volume and market share.
But in his view, it is still too early to discuss winners and losers because “the battle for advanced autonomous driving has not yet begun”.
“We see this market divided into several stages. Three years ago, customers did not ask about technology, only price; this year, entering the high-speed NOA era, customers do not look at technology, only cost, because high-speed NOA products have already been validated; the technical solutions for urban NOA are still maturing, and I believe this process will continue for another two to three years, after which urban NOA will begin to be widely adopted, and that will be the start of this battle.”
In Black Sesame’s view, 2025 will be a significant watershed for intelligent driving. Reports indicate that by the end of this year, the penetration rate of urban NOA in passenger vehicles will reach 20%, and by 2030, almost all vehicles will have urban NOA capabilities. “Intelligent driving will become a standard feature like air conditioning and seating.”
With the arrival of the “Universal Intelligent Driving” era, as a leading company in the Chinese chip industry, Black Sesame is very optimistic about market prospects. Especially in competition with foreign giants, the competitive advantages of Chinese companies are becoming increasingly evident.
Yang Yuxin stated that from a technical perspective, global AI-related chips will lag NVIDIA by half a beat to a beat, but in the field of intelligent driving, our maturity in product performance is gradually catching up.
“We believe that in the inference field, ASIC chips have advantages over GPUs in terms of cost and power consumption. Once the scenarios begin to solidify, the advantages become more apparent.” The intelligent driving chips launched by Black Sesame belong to ASICs.
At the same time, in terms of user experience and price, Chinese car manufacturers have a very good competitive advantage, which lays a solid foundation for the entire automotive industry chain. “In recent years, China has not been idle; the capabilities developed have made us very competitive globally.”
However, Yang Yuxin also stated that at the highest level of competition, it is not just about product strength and commercialization capabilities. “Currently, we are facing global tariff encirclement, which is not something that those of us in technology or product development can decide. But we believe in the strength of China, and we hope to rise alongside the rise of the Chinese automotive industry.”
Self-Developed Chips by Car Manufacturers May Be a Temporary Attempt
Since last year, NIO and Xpeng have been increasing their investment in self-developed intelligent driving chips. The main reasons are twofold: cost reduction and differentiation.
Self-developed chips can save costs for car manufacturers and also significantly reduce software investment. Because developing software on a general platform requires more investment to achieve differentiation, achieving some differentiation through a combination of hardware and software may lead to some savings in software investment.
Regarding whether car manufacturers should develop their own chips or procure third-party chips, Yang Yuxin believes that according to the previous development patterns of various industries, third-party chip suppliers are definitely needed and will be needed in the long term. Most manufacturers still hope to have more professional partners to provide chip support, as there are not many manufacturers with the resources and determination to develop chips.
Just like in the mobile phone era, Qualcomm and MTK are the undisputed leaders, while other OEMs developing their own chips, such as Huawei and Samsung, have been successful in supporting their differentiation. Google’s current phone also uses self-developed chips, but much of the design is still outsourced to Samsung.
Yang Yuxin believes that a more reasonable approach is for car manufacturers to highlight their differentiation while outsourcing the design of general-purpose chips to more professional chip partners.
“Everyone can look at Tesla, which is considered the best practice so far. To achieve differentiation in intelligent driving, they chose to develop all chips in HW3.0, including front-end design, back-end design, tape-out, and NPU design, as well as algorithms and toolchains. They have indeed made significant progress. However, returning to HW4.0, from a commercial rationality perspective, Tesla retained its chip architecture design, NPU, algorithms, and software systems while outsourcing chip design to Broadcom, which is more reasonable. Because in terms of team maturity, IP procurement prices, and foundry prices, Broadcom is better than Tesla.”
“The upstream of the chip industry is very realistic; it only looks at volume, regardless of the size of your company name.”
Therefore, Black Sesame Intelligence believes that in the future, car manufacturers will still consider finding the optimal cooperation model from a commercial perspective. According to Yang Yuxin, Black Sesame is also exploring deeper cooperation models with car manufacturers.
Star mark and follow the “Intelligent Vehicle Planet” public account
The most sincere reporting on smart cars
