Naveen Rao is an American entrepreneur and engineer known for his work in Artificial intelligence and Machine learning, particularly in the field of Large language model training. He co-founded and served as CEO of MosaicML, a startup focused on making AI model training more accessible and cost-effective, which was acquired by Databricks in July 2023 for approximately $1.3 billion. Prior to MosaicML, Rao held leadership roles at Intel and Nervana Systems, contributing to the development of specialized hardware for Deep learning.
Rao's career spans both hardware and software aspects of AI, reflecting a deep understanding of the computational challenges underlying modern Neural networks. His work has been instrumental in advancing the practical deployment of Generative AI systems, particularly through techniques like model compression and efficient distributed training.
Early Career and Education
Rao earned his Bachelor of Science in Electrical Engineering from Duke University and later completed a PhD in Neuroscience at Brown University. His academic background combined engineering with a focus on biological neural systems, which informed his later approach to Machine learning. Before entering the AI industry, he worked as a chip architect at Broadcom and Qualcomm, where he gained experience in designing high-performance processors.
In 2014, Rao co-founded Nervana Systems with Amir Khosrowshahi and Arjun Bansal. The company developed a custom AI chip called the Nervana Engine, designed specifically for Deep learning workloads. Nervana was acquired by Intel in August 2016 for a reported $408 million, after which Rao became the general manager of Intel's AI products group.
Leadership at Intel
At Intel, Rao oversaw the development of the Nervana Neural Network Processor (NNP) family, including the NNP-T for training and NNP-I for inference. These chips were intended to compete with NVIDIA's GPUs in the AI accelerator market. Under his leadership, Intel also acquired Vertex.AI in 2018, a startup focused on AI model optimization, to strengthen its software stack.
Despite these efforts, Intel's AI hardware struggled to gain significant market traction against established competitors. Rao left Intel in 2020, and Intel later discontinued the Nervana product line in 2021, shifting focus to other AI initiatives. His time at Intel provided valuable insights into the limitations of hardware-centric approaches, which influenced his subsequent work at MosaicML.
Founding MosaicML
In 2021, Rao co-founded MosaicML with Hanlin Tang and Jonathan Frankle. The company aimed to democratize AI by providing tools and platforms for training Large language models efficiently. MosaicML's flagship product, the MosaicML Platform, offered a managed service for training and fine-tuning models, along with the MPT (Mosaic Pretrained Transformer) family of open-source models.
One of MosaicML's key innovations was the use of Transformer (architecture) architectures with optimized training techniques, such as FlashAttention and low-precision training. The company demonstrated that it could train models with billions of parameters at a fraction of the cost typically associated with such efforts. In March 2023, MosaicML released MPT-7B, a 7-billion-parameter model that was notable for its competitive performance and open-source availability.
Acquisition by Databricks and Later Work
Databricks, a data analytics company, acquired MosaicML in July 2023. The acquisition was part of Databricks' strategy to integrate AI model training capabilities into its platform, competing with offerings from OpenAI, Anthropic, and Google Cloud. Following the acquisition, Rao became the vice president of AI at Databricks, leading the company's efforts to develop and deploy Generative AI solutions.
Under Rao's leadership, Databricks released the DBRX model in March 2024, a 132-billion-parameter Large language model with a mixture-of-experts architecture. DBRX was positioned as an open-source alternative to proprietary models, emphasizing efficiency and performance. Rao has also been a vocal advocate for open-source AI, arguing that transparency and community collaboration are essential for responsible AI development.
Contributions and Recognition
Rao has authored several papers on Deep learning and Machine learning, and he holds multiple patents related to AI hardware and software. He has been a frequent speaker at industry conferences, including the Conference on Neural Information Processing Systems (NeurIPS) and the International Conference on Machine Learning (ICML). In 2023, he was named to the Forbes 50 Over 50 list for his entrepreneurial achievements.
His work has influenced the broader AI ecosystem, particularly in the area of cost-efficient model training. By reducing the computational resources required for Large language model development, Rao has helped enable smaller organizations and researchers to participate in AI research and application, challenging the dominance of large tech companies in the field.
Personal Life and Philosophy
Rao is known for his pragmatic approach to AI, emphasizing practical outcomes over theoretical elegance. He has spoken about the importance of aligning AI development with human values and has supported initiatives for AI safety and ethics. In interviews, he has cited his neuroscience background as a source of inspiration, noting that understanding biological intelligence can inform the design of artificial systems.
He resides in the San Francisco Bay Area and continues to be active in the AI startup community, mentoring early-stage companies and advising on technology strategy. His career trajectory from chip design to AI software illustrates the interdisciplinary nature of modern Artificial intelligence research and development.