Source: XPeng

Xpeng unveiled the much anticipated G7 mid-size SUV on June 11, 2025, with pre-sales commencing immediately and deliveries slated for Q3 2025.

The G7 stands out as the first "AI car" with L3-level computing hardware, powered by the company’s self-developed Turing AI chip. The G7’s starting pre-sale price is 235,800 RMB (approximately 32,800 USD)

Source: XPeng

The name "Turing" is a deliberate homage to Alan Turing, the pioneering British mathematician and computer scientist widely regarded as the father of theoretical computer science and artificial intelligence.

The Turing AI chip, developed entirely in-house by Xpeng, is engineered from the ground up for demanding AI applications.

It boasts a robust 40-core processor and is uniquely capable of locally running large AI models with up to 30 billion parameters, a substantial leap in on-device AI processing.

The chip's architecture integrates two proprietary Neural Processing Units (NPUs) and a specialized design meticulously optimized for neural networks, ensuring formidable computing power for Xpeng's AI vehicles.

A key metric in evaluating AI chip performance is TOPS, which stands for Trillions of Operations Per Second. This unit quantifies the sheer computational power of AI chips, reflecting the number of trillions of operations a processor can execute within a single second.

By this same metric, the Turing chip is touted by Xpeng as a "computing power ceiling," delivering a staggering single-chip computing power of 750 TOPS.

To provide context, this single Turing chip is effectively equivalent to three Nvidia Orin X chips, with each Orin X typically offering around 254 TOPS.

This deep customization of the Turing chip allows for an impressive 100% utilization of its raw computing power, a significant advantage over more general-purpose chips.

The G7's Ultra version notably integrates three of these Turing AI chips simultaneously, culminating in a combined total computing power of 2,250 TOPS.

That amount of computing power even exceeds NVIDIA’s next-gen centralized computer, the DRIVE Thor, which will achieve a performance of up to 2,000 TOPS.

Source: NVIDIA

XPeng claims that the 3x Turing AI chips setup boast processing capability surpassing other offerings in the automotive sector by a factor of 3 to 23 times.

The triple chip setup also features 216 GB of memory (RAM), leading the industry’s competing models by 8 to 10 times.

Source: XPeng

Furthermore, the Turing chip is designed to effectively replace the traditional CPU and GPU architectures for AI workloads within the vehicle.

Most modern high-end vehicles today feature Qualcomm's Snapdragon 8295, a powerful System-on-Chip (SoC) primarily designed for advanced infotainment and cockpit functions, with its AI capabilities typically in the range of tens of TOPS (e.g., around 30 TOPS for its NPU).

The Turing chip, with its 750 TOPS per single chip, represents an order of magnitude leap in dedicated AI processing power, allowing for far more complex and real-time AI computations vital for autonomous driving and advanced cabin functionalities, moving beyond the traditional roles of a general-purpose CPU and graphics-focused GPU.

NIO’s NX9031

NIO's Shenji NX9031, unveiled in December 2023 for the ET9 executive flagship sedan is built on a 5nm process, packs over 50 billion transistors, and features a 32-core CPU.

NIO's founder William Li stated that the Shenji NX9031's computing power is comparable to four Nvidia Drive Orin X chips, implying a figure around 1,016 TOPS, given that NIO's current NT 2.0 platform models already utilize four Orin X chips to achieve that total.

In contrast, Xpeng's Turing AI chip, as showcased in the G7's Ultra version, fundamentally aims to replace the vehicle's traditional CPU and GPU for AI workloads, a crucial differentiator.

Each Turing chip delivers a staggering 750 TOPS, and with three integrated, the G7 boasts a combined 2,250 TOPS.

This significantly higher aggregate computing power, coupled with its dedicated AI architecture and ability to run large models with 30 billion parameters, suggests Xpeng is pursuing a more integrated approach that takes over the core processing functions of the vehicle beyond just autonomous driving.