Nvidia (NASDAQ: NVDA) quietly acquired AI startup OmniML in February this year. OmniML’s technology miniaturizes machine learning models, enabling them to function on edge devices instead of relying on the cloud. Evidence of this acquisition surfaced from a spokesperson and LinkedIn profiles of former OmniML staff, now enlisted under Nvidia.

After raising $10 million in seed funding in March last year, supported by investors such as GGV Capital and Qualcomm Ventures, the startup partnered with Intel to expedite the deployment and evolution of AI applications across a multitude of sectors. Their proprietary software platform, Omnimizer®, has been instrumental in augmenting machine learning model performance, resulting in swifter and more efficient AI.

The acquisition indicates Nvidia’s strategic intent to harness OmniML’s unique AI compression software, capable of creating compact, scalable machine learning models.

Edge devices such as drones and smart cameras often grapple with a lack of computational power to support large AI models. OmniML’s innovation lies in its ability to compress these models, allowing them to function smoothly on edge devices, and thereby accelerating machine learning tasks. Their technology automates model co-design, training, and deployment, targeting GPUs, FPGAs, SoCs, and MCUs.

With the acquisition, Nvidia can enhance its AI chips for autonomous vehicles, drones, and industrial robots, and further their efforts to build smaller and more power-efficient AI models.

In real-world applications, OmniML’s technology has been proven in areas like video surveillance, boosting the situational awareness of smart cameras, and in precision manufacturing, where it has refined quality control detection models. Its successful integration with Amazon Web Services’ AutoML Library and Meta’s PyTorch deep learning framework highlight the potential of OmniML’s technology in shaping the future of AI.

The startup was founded in 2021 by Dr. Di Wu (CEO), Dr. Huizi Mao (CTO), and Dr. Song Han.