Ambarella is showcasing its CV7 edge artificial intelligence vision system-on-chip at CES 2026, positioning it for a broad mix of high-end imaging and automation workloads. The SoC is designed for artificial intelligence based 8K consumer devices, multi-imager enterprise security cameras, robotics, industrial automation, and high-performance video conferencing systems, where low-latency visual intelligence needs to run at the edge rather than in the cloud. It is engineered to support multi-stream automotive designs that execute convolutional neural networks and transformer-based models in distributed environments, including artificial intelligence vision gateways and hubs used in fleet video telematics, 360-degree surround-view and recording, and passive driver assistance systems.
The CV7 supports simultaneous processing of multiple video streams up to 8Kp60 in order to handle dense multi-camera configurations common in vehicles, smart buildings, and factories. Relating to the prior generation, the CV7 consumes 20 percent less power due to Samsung’s 4nm process technology, which marks Ambarella’s first use of this node and is intended to extend deployment into thermally constrained designs. The chip follows Ambarella’s algorithm-first design philosophy so that its heterogeneous compute blocks can manage imaging, encoding, and inference tasks concurrently without overloading any single subsystem.
The SoC integrates Ambarella’s proprietary artificial intelligence accelerator, an image signal processor, and video encoding engines alongside Arm cores, I/Os, and supporting logic to create a tightly coupled edge platform. Artificial intelligence performance leverages Ambarella’s third-generation CVflow accelerator, which delivers more than 2.5x artificial intelligence performance over the previous-generation CV5 SoC, enabling more complex transformer-based networks and emerging vision-language models to run locally. Image signal processing enhancements include high dynamic range, dewarping for fisheye optics, and 3D motion-compensated temporal filtering, which together enable image quality in low light, down to 0.01 Lux, and improved HDR for both video and stills. The SoC has a max video encode of a single 4Kp240 stream, or dual 8Kp30, and for security cameras, the solution supports over 4x 4Kp30 with multiple streams alongside the latest transformer-based artificial intelligence networks.
General-purpose compute on the CV7 was upgraded to a quad-core Arm Cortex-A73 complex that offers 2x higher CPU performance over the previous SoC, giving developers more headroom for application logic, networking, and custom analytics around the dedicated accelerators. Additionally, its 64-bit DRAM interface provides a significant improvement in available DRAM bandwidth compared to the CV5 to sustain multi-stream, high-resolution processing. Ambarella states that its portfolio of edge artificial intelligence SoCs has shipped more than 39 million units to date, and positions the CV7 as a way for consumer and enterprise camera makers to integrate advanced imaging and higher edge artificial intelligence performance into next-generation designs. According to the company, CV7 SoC samples are available now and are being demonstrated at an invitation-only exhibition during CES 2026 for customers interested in evaluating the platform.
