DEEPX has announced the launch of its DX-H1 vision neural processing unit (V-NPU), a dedicated video intelligence chipset designed to process hundreds of AI video channels at just 30 W. According to DEEPX, the DX-H1 combines video decoding, AI inference, and encoding on a single chip, targeting data center video processing workloads and offering a structural alternative to traditional general-purpose graphics processing units (GPUs).
The DX-H1 V-NPU features an all-in-one architecture with a multi-channel engine for decoding, encoding, and transcoding, supported by a dedicated neural processing unit core. DEEPX claims that this approach allows the DX-H1 to streamline operations by handling stream input, preprocessing, AI inference, and re-encoding directly on a single device. This is intended to eliminate the need for separate GPU servers and hardware codecs in video AI pipelines.
DEEPX reports that, based on its internal testing, integrating these functions on the DX-H1 reduces hardware costs by approximately 80 percent and power consumption by 85 percent compared to GPU-based solutions with similar channel densities, while supporting continuous real-time inference. These specifications make the DX-H1 particularly suited for data center applications in smart cities and industrial surveillance environments, where power constraints and high-density video workloads are common operational challenges.
The company states that the new V-NPU builds on experience from prior DX-H1 deployments, which have added AI capabilities to legacy systems such as closed-circuit television and network video recorders. The updated variant is intended as a core platform for new server deployments in smart cities, traffic control centers, and large-scale industrial complexes, shifting the data center server architecture focus from GPUs to V-NPUs.
The DX-H1 V-NPU has received a CES 2026 Innovation Award, with DEEPX set to officially unveil it at CES 2026 in Las Vegas.
“Large-scale video AI can no longer be a secondary task borrowing spare resources from general-purpose GPUs; it must evolve into a dedicated industry running on specialized chipsets,” said Lokwon Kim, CEO of DEEPX. “The DX-H1 V-NPU is not merely a low-cost alternative but a fundamental redesign of video intelligence infrastructure, optimizing memory hierarchy and computation scheduling for environments where video streams pour in by the second.”
Source: DEEPX
