Yonatan Geifman

Deci AI developed neural architecture search and model optimisation technology that automated the process of compressing and restructuring deep learning models for faster inference on target hardware, enabling AI teams to deploy models on edge devices, GPUs, and CPUs with significantly lower latency and compute cost than the original training architecture. The Tel Aviv company built AutoNAC, an automated neural architecture construction tool that generates hardware-aware model architectures optimised for specific deployment targets.\n\nThe company raised approximately $80 million in venture funding including a $55 million Series B from investors including Square Peg, Insight Partners, and Jibe Ventures before being acquired by NVIDIA in April 2024. The acquisition gave NVIDIA a model optimisation software layer to complement its GPU hardware, enabling NVIDIA to offer end-to-end AI deployment tools that span hardware, software runtime, and model architecture optimisation. Deci team integrated into NVIDIA developer tools organisation.\n\nDeci competed in the AI model optimisation and neural architecture search market alongside Neural Magic, Hailo, and model compression tools from major framework providers including TensorFlow Model Optimization Toolkit and PyTorch. The acquisition by NVIDIA is consistent with the chip company strategy of acquiring software and tooling companies that deepen customer dependency on the NVIDIA ecosystem. Deci inference acceleration technology provides NVIDIA customers measurable throughput and cost improvements on existing hardware, making it a natural product addition to the CUDA and TensorRT software stack.