Jensen Huang announced that the inflection point of Robotaxi has arrived, and joined hands with Uber to form an ecological alliance of 100,000 L4 vehicles.
"Both human and robot technologies are still evolving. Meanwhile, there is a type of robot that is clearly at a turning point in the era. It is the Robotaxi!"
Recently, Jensen Huang, the founder, president, and CEO of NVIDIA, announced at the GTC conference that the turning point for Robotaxis is approaching. There are currently about 50 million taxis globally, and they will be supplemented by a large number of driverless taxis. This will be a huge market.
To this end, NVIDIA has launched "an important product": the brand - new NVIDIA DRIVE AGX Hyperion 10 platform.
Under this platform, NVIDIA officially announced a series of new partnerships. The most eye - catching one is the cooperation with Uber. The two parties plan to deploy approximately 100,000 Robotaxis on a large scale starting from 2027.
Meanwhile, NVIDIA is also collaborating with automakers such as Stellantis, Lucid, and Mercedes - Benz to build L4 - level autonomous passenger cars on the DRIVE AGX Hyperion 10 platform. It will also expand L4 - level autonomous driving technology to the long - haul freight transport sector with Aurora, Volvo Autonomous Solutions, and Waabi.
Jensen Huang, the founder of NVIDIA
"We built this architecture to enable global automakers, whether they produce commercial vehicles, passenger cars, or taxis specifically for autonomous driving, to build vehicles with autonomous driving capabilities." In short, Jensen Huang wants to use the Hyperion 10 platform to connect autonomous driving to various global networks.
As the Robotaxi track heats up again, NVIDIA's strategic focus in the automotive sector seems to be further tilting towards the L4 field.
01
Hyperion 10 Debuts: Soaring Computing Power Aims at L4
"This is a brand - new computing platform, and I think it will be very successful." Jensen Huang is very confident about the Hyperion 10 platform.
Judging from the parameters, it is indeed good.
The NVIDIA DRIVE AGX Hyperion 10 is equipped with two in - vehicle DRIVE AGX Thor platforms based on the Blackwell architecture, with a total computing power of over 2000 TOPS, nearly 8 times that of the previous - generation Orin chip. It can fuse diverse input data from all - around sensors and is optimized for Transformer, Vision - Language - Action (VLA) models, and generative AI workloads.
NVIDIA demonstrates autonomous driving technology
In terms of sensors, it includes 14 high - definition cameras, 9 millimeter - wave radars, 1 lidar, and 12 ultrasonic sensors. Compared with Hyperion 9, it has 2 fewer lidars and 8 fewer ultrasonic sensors, which is basically the same as Hyperion 8.
NVIDIA introduced that one of the biggest features of Hyperion 10 is its modular architecture and support for customization. Manufacturers and assisted - driving developers can flexibly configure it according to their own needs, which can not only shorten the development cycle but also reduce costs.
Another major feature is its scalability and compatibility with existing assisted - driving software. Enterprises can use Over - the - Air (OTA) technology to seamlessly integrate and deploy models from this platform into driverless taxis and autonomous vehicle fleets.
The DRIVE platform is NVIDIA's full - stack solution for autonomous vehicle development.
As early as 2015, NVIDIA launched the NVIDIA Drive series of platforms to empower the autonomous driving ecosystem. At CES 2015, it launched the first - generation platforms based on NVIDIA's Maxwell GPU architecture: DRIVE CX, which is equipped with 1 Tegra X1 and mainly targets the digital cockpit; and DRIVE PX, which is equipped with 2 Tegra X1 and mainly targets autonomous driving.
NVIDIA DRIVE platform
Since then, NVIDIA has updated the Drive platform one or two times almost every year, released a vehicle - grade SoC chip every two years, and continuously increased the computing power.
At the NVIDIA GTC conference in 2021, NVIDIA released the DRIVE Hyperion 8, a reference platform for autonomous driving software and hardware development, including core computing, middleware, and in - vehicle AI functions. In terms of hardware for this computing platform, it is equipped with two Orin chips, each with a computing power of 254 TOPS, supporting 12 cameras, 9 millimeter - wave radars, 12 ultrasonic radars, and 1 lidar.
In 2022, NVIDIA released the next - generation autonomous driving platform, NVIDIA DRIVE Hyperion 9, at the GTC conference that year. The reference design of this platform includes 2 Atlan chips. In terms of sensors, it includes 14 cameras, 9 millimeter - wave radars, 3 lidars, and 20 ultrasonic sensors.
According to the original plan, Hyperion 9 was supposed to be installed in vehicles starting from 2026. However, with the release of Thor, Atlan was self - eliminated before it was even launched, and Hyperion 9 was actually not mass - produced. Hyperion 10 emerged along with Thor.
With the launch of the NVIDIA DRIVE AGX Hyperion 10 platform, NVIDIA is also shifting more of its focus to L4 - level autonomous driving.
02
L4 Ecosystem: Form Alliances and Move towards Commercialization
Currently, the most concerning thing is undoubtedly the cooperation between NVIDIA and Uber.
After all, the two parties have considerable "ambitions". That is, to build a global L4 network through the Hyperion platform and operate in collaboration with Uber, with the goal of achieving a market scale of $750 billion (approximately RMB 5.35 trillion) by 2030.
After all, expanding to 100,000 vehicles in 2027 is a significant volume.
On October 28th local time, NVIDIA and Uber announced that they will jointly expand the global L4 - level autonomous mobility network. Relying on Uber's new - generation driverless taxis and autonomous delivery fleets and using the Hyperion 10 platform, it will help Uber gradually expand the scale of its global autonomous vehicle fleet, with a plan to gradually expand to 100,000 vehicles starting from 2027.
NVIDIA and Uber officially announce their cooperation
These vehicles will be jointly developed by Uber, NVIDIA, and other ecosystem partners and will use NVIDIA DRIVE technology. In addition, the two parties will also jointly build a data factory based on the NVIDIA Cosmos world - base model development platform to sort out and process the data required for the R & D of autonomous vehicles.
Actually, in the past period, Uber has officially announced a series of partnerships based on Robotaxis. Currently, this cooperation with NVIDIA is an extension of the previous ones. The partners include Avride, May Mobility, Momenta, Nuro, Pony.ai, Wayve, and WeRide.
In the statement released by Uber, it also said: "These functions together form a powerful data - processing system, covering data collection, annotation, scenario mining, synthetic data generation, and large - scale training, aiming to accelerate the deployment process of autonomous driving from pilot testing to commercial profitability."
This means that the two parties have both technological cooperation and commercial discussions.
Not only for mobility platforms, NVIDIA also hopes to help automakers build autonomous vehicles.
So, on that day, Jensen Huang also announced that NVIDIA has established in - depth cooperation with multiple automakers through the Hyperion 10 platform to further expand the L4 - level autonomous driving ecosystem, including Stellantis, Lucid, and Mercedes - Benz.
NVIDIA's autonomous driving ecosystem
Stellantis is developing its autonomous driving platform, which will integrate NVIDIA's full - stack AI technology. Specifically optimized, it aims to support L4 - level autonomous driving functions and meet the needs of Robotaxis and will be connected to the Uber ecosystem.
Stellantis plans to provide at least 5000 Robotaxi pilot models to Uber starting from 2026. The first - batch models may come from Stellantis' AV - Ready platform, especially the K0 mid - size van and the STLA Small platform. Pilot operations will start in some cities in North America and Europe in 2026 to prepare for large - scale deployment.
Lucid is promoting the implementation of L4 - level autonomous driving capabilities for its new - generation passenger cars and will use NVIDIA's full - stack assisted - driving software based on the DRIVE Hyperion platform in its upcoming US models.
Mercedes - Benz is promoting the global implementation of L4 - level autonomous driving based on its self - developed operating system MB.OS and the DRIVE AGX Hyperion platform. It is reported that the L4 - level autonomous driving fleet jointly developed by NVIDIA and Mercedes - Benz will be launched in 2025.
In the trucking field, Aurora, Volvo Autonomous Solutions, and Waabi are jointly developing L4 - level autonomous trucks based on the NVIDIA DRIVE platform.
As Jensen Huang said, he hopes that the Hyperion 10 platform can enable commercial vehicles, passenger cars, and Robotaxis to have autonomous driving capabilities.
03
From the "Fiercely Competitive" Auto Market to Ecosystem Expansion: A New Narrative for the Automotive Business
Judging from Jensen Huang's description at this GTC conference, NVIDIA seems to be placing more emphasis on L4 - level autonomous driving in the automotive field, which is also NVIDIA's strength.
As mentioned above, NVIDIA has been involved in the automotive field for a long time.
At the GTC conference in 2015, Jensen Huang, the founder of NVIDIA, revealed his ideas about autonomous vehicles: "Deep - learning neural networks can enable vehicles to learn autonomous driving." The next year, NVIDIA launched the first - generation autonomous vehicle computing platform, Drive PX, which was favored by Tesla under Elon Musk's leadership and became the computing chip for Tesla's HW2.0 hardware to develop the AutoPilot advanced intelligent assisted - driving function.
Since then, NVIDIA's automotive business has officially started, mainly focusing on the autonomous driving chip business.
At the "NIO Day" in 2021, the NIO ET7 was officially unveiled and announced as the first mass - produced vehicle of the Orin series.
The NIO ET7 is the first mass - produced vehicle of the Orin series
Subsequently, including the NIO ET7, IM Motors and WM Motor's M7 both said they would be equipped with four Orin chips, with a total computing power of over 1000 TOPS.
At that time, the single - chip computing power of its competitors, such as Mobileye, Huawei, and Horizon, which could be mass - produced, was basically in the dozens of TOPS.
Therefore, at that time, the industry believed that the delivery of NVIDIA's Orin chip would be a milestone event for electric vehicles. On the one hand, the computing power of autonomous driving chips would replace the horsepower index of traditional fuel vehicles and become a new competitive point in the automotive industry. On the other hand, NVIDIA would start to "dominate" the automotive circle.
Obviously, the result is not entirely the case.
Although NVIDIA has a relatively high market share in the global high - computing - power autonomous driving chip market and has received a series of orders from the automotive circle due to its high computing power and strong ecosystem.
For a long time, some of NVIDIA's automotive partners needed to develop L3 - level and above intelligent driving, and there were not many chip options on the market. Moreover, some of these automakers were also developing their own chips. Another part regarded NVIDIA as one of the chip suppliers and installed the chips in non - high - volume models.
This has led to NVIDIA's automotive business accounting for a relatively small proportion of its revenue. The financial report for the second quarter of 2025 shows that NVIDIA's automotive and robotics business revenue was $586 million, a year - on - year increase of 69%, but it accounted for only 1.25%.
According to NVIDIA's prediction, the vertical automotive revenue this year will reach $5 billion, exceeding the total of the past five years. However, it still seems to be a long way off.
If NVIDIA wants to take its automotive business to the next level, the fiercely competitive auto market is obviously not its strength. The broader autonomous driving track has more room for imagination.
This article is from the WeChat official account "Cyber - car" (ID: Cyber - car). Author: Zhang Lianyi, Editor: Qiu Kaijun. Republished by 36Kr with permission.