"Inferring What's Possible"
Israeli deep learning optimization startup acquired by NVIDIA in May 2024 for ~$300M; known for AutoNAC neural architecture search technology.
Deci AI was founded in 2019 in Tel Aviv (Ramat Gan), Israel, by Dr. Yonatan Geifman, Prof. Ran El-Yaniv (Technion), and Jonathan Elial. The company developed an end-to-end deep learning platform powered by Automated Neural Architecture Construction (AutoNAC), enabling AI developers to build, optimize, and deploy faster and more accurate models for cloud, edge, and mobile hardware. Deci's technology delivered 3x–15x inference speedup while preserving accuracy and reducing compute costs by up to 80%.
Key products included DeciLM-7B (most accurate 7B base model with 4.4x throughput vs Mistral-7B), DeciCoder 1B and 6B (code generation), and DeciDiffusion 2.0 (text-to-image, 3x faster than Stable Diffusion). The company raised $55M and set a record 11.8x inference speedup on Intel CPUs at MLPerf. NVIDIA acquired Deci in April–May 2024 for approximately $300 million as part of a broader Israeli AI acquisition strategy (alongside Run:ai).
The company's technology has been integrated into NVIDIA's AI stack.
Primary type & automation activities this supplier delivers:
Open-source PyTorch training library for SOTA computer-vision models (detection, classification, segmentation); home of YOLO-NAS.
Inference runtime engine delivering low-latency, high-throughput model execution with optimal hardware utilization.
Hardware-aware Automated Neural Architecture Construction that redesigns models to boost throughput and cut latency while preserving accuracy.
Contact Deci AI (Acquired by NVIDIA)
WEBSITE
https://deci.aiPHONE
+972 3 375 1969HEADQUARTERS
Israel
Company Facts
Founded
2019
Primary Role
Software/Algorithm
Company Size
-
Primary Region
Middle East
Annual Sales
-
Funding Stage
-
Funding Total
-