XHEART New Core Voyage Releases X7 Ultra and Discloses X9 Goal of 3,000 TOPS per Single Chip

Urban assisted driving is moving downmarket into models priced around 100,000 yuan, yet smart-driving vendors are still competing over higher compute ceilings. On one side, lower costs are being used to expand coverage; on the other, capability is being reserved for larger on-vehicle models. These two demands run in parallel, making this competition hard to settle on a single compute figure alone.

On October 29, XHEART New Core Voyage released the X7 Ultra. By the company's own account, its equivalent large-model compute for smart-driving scenarios reaches 700 TOPS; it also disclosed the R&D goals for the next-generation X9: single-chip large-model compute of 3000 TOPS and memory bandwidth of 700GB/s, with a planned launch in 2027.

The shift in compute demand first shows up in the tasks the models take on. When XPeng introduced the new version of its second-generation VLA in August this year, it said the on-device model's parameter count had expanded 3.5 times and introduced memory of information from the past 30 seconds and scenario prediction for the next 6 seconds. The model needs to call on longer historical information and simulate possible future changes.

XHEART CEO Li Zonglin laid out the company's core technology route on October 29: large models define chips. As intelligent driving moves from multiple small models to end-to-end large models, the way chips are designed also needs to change accordingly.

Momenta CEO Cao Xudong said at the launch event that the next stage of high-level autonomous driving will be large models defining chips. To achieve scaled L4 autonomous driving, the vehicle side will need to carry large models with tens of billions or even hundreds of billions of parameters, which will place higher demands on chip compute, energy efficiency, bandwidth and cost.

These tasks must be completed in a timely manner while driving. Larger parameter counts, longer context and more simulation all increase computing and storage pressure. Beyond compute, bandwidth has also become key: if model weights and intermediate data cannot be delivered to the compute units in time, adding compute may not shorten inference time proportionally. The X9 raising both compute and bandwidth targets responds precisely to this kind of demand.

For suppliers that need to lock in architectures in advance, this is also a bet on next-generation models. Reserving too little may limit model upgrades; reserving too much adds hardware, power and cooling burdens. As a result, model companies are participating earlier in hardware design so that architectures are optimized around actual tasks.

XHEART chose to work with Momenta. Under Momenta's plan, the R7 Ultra solution built on the X7 Ultra will expand model parameter scale to 5 times the original. The two sides are putting models and hardware into the same upgrade plan in an attempt to reduce adaptation losses. But competitors are already positioning themselves along this route.

NVIDIA offers DriveOS, CUDA, TensorRT and other tools around DRIVE AGX Thor. When automakers choose a platform, they are also choosing the toolchain on which subsequent model development and deployment will depend. New entrants seeking customers need to take on far more work than delivering hardware.

Domestic suppliers already have production scale. Horizon Robotics' planned Journey 7 targets L3 and L4, with native optimization for larger-parameter HSD smart-driving models and cockpit models, and is expected to complete tape-out in early Q2 2027.

This means XHEART faces mature platforms that already hold share on one side, and on the other must compete with next-generation products that rivals are developing. Software-hardware co-design is becoming a common choice for multiple suppliers.

Automakers' in-house development adds another competitive route. XPeng disclosed in October that its Youyou Robotaxi, which obtained Guangzhou's L4 demonstration operation qualification, is equipped with four self-developed Turing AI chips, achieves 3000 TOPS of effective vehicle-level compute, and runs the second-generation VLA. Its XPENG AI ENGINEERING cooperation system has also brought smart driving, chips and toolchains into its external cooperation scope. Some potential buyers are becoming solution providers.

A distinction is needed here: XPeng's 3000 TOPS is the combined vehicle total across multiple chips, while XHEART New Core Voyage's X9 is a single-chip target. Different vendors also use different computing precisions and equivalent-compute definitions, so actual driving capability cannot be ranked directly on this basis.

But the race for high-end platforms cannot replace the immediate need for popularization. Horizon Robotics disclosed in September that the 128 TOPS Journey 6M is expanding into the mainstream market around 100,000 yuan; as of the end of August, seven automaker brands had already achieved platform-based urban assisted driving deployment based on the product; a single Journey 6M solution can also use air cooling, adapting to the power supply and cooling constraints of fuel vehicles.

For these models, procurement must calculate the cost of the entire system: how much experience improvement the extra compute can bring, whether more complex cooling is needed, and whether the corresponding development investment can be borne. The 3000 TOPS aimed at next-generation models and the current popularization of urban assisted driving correspond to different product budgets and delivery requirements.

XHEART has already built up a certain production base. According to company disclosures, the X7 series shipped more than 200,000 units cumulatively in 8 months of mass production, is installed on 10 mass-production models, and has won orders for more than 40 models.

3000 TOPS shows the direction in which smart-driving vendors are investing for next-generation models, but the market will not pay only for a compute ceiling. What XHEART must win is automakers' belief that, among existing suppliers, in-house automakers and low-cost solutions, the same hardware and development budget will yield more deliverable smart-driving capability if they choose it. Risk disclosure and disclaimer

Markets carry risk, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation or needs of individual users. Users should consider whether any opinions, views or conclusions in this article fit their particular circumstances. They invest at their own risk.