Rodrigo Liang, CEO of SambaNova, explains how their AI chips, built on a data flow architecture optimized for high-performance inferencing, outperform Nvidia’s GPUs by delivering faster, more energy-efficient real-time AI responses, particularly benefiting sovereign cloud initiatives with strict data security needs. Partnering with Intel to scale deployment, SambaNova aims to power the future of personalized AI agents that will transform daily productivity by handling routine tasks efficiently.
In this interview, Rodrigo Liang, CEO and co-founder of SambaNova, discusses how his company is challenging Nvidia in the AI chip market by focusing on high-performance inferencing technology. Unlike Nvidia’s GPUs, which were originally designed for graphics processing and later adapted for AI training, SambaNova’s chips are built on a data flow architecture specifically optimized for AI workloads. This approach allows SambaNova to deliver inference performance that is 5 to 10 times faster than GPUs while consuming only one-tenth of the power, significantly reducing operational costs and energy consumption in data centers.
Rodrigo emphasizes the growing importance of inference in AI, which involves using trained models to generate real-time responses for millions of users, as opposed to training models which is done by a smaller group of researchers over extended periods. He explains that inference requires low latency and high availability to meet user expectations for instant responses, something traditional GPUs struggle with. SambaNova’s technology addresses this need by enabling fast, real-time inference, which is critical as AI applications become more pervasive and interactive.
The conversation also covers SambaNova’s strategic partnership with Intel, which provides the scale, capital, and ecosystem support necessary to compete with Nvidia’s dominance. While SambaNova has developed advanced chip technology over nine years, partnering with Intel helps accelerate deployment and integration into cloud infrastructure. Rodrigo highlights that the future of AI involves heterogeneous systems where different technologies work together, and Intel’s broad portfolio complements SambaNova’s core strength in fast inference.
Rodrigo shares that SambaNova’s chips are gaining traction particularly in sovereign cloud initiatives across countries like Australia, Germany, the UK, France, and Japan. These countries prioritize data security, privacy, and compliance by hosting AI models and data within their borders, often in standard air-cooled data centers that cannot support the high power demands of Nvidia’s GPUs. SambaNova’s energy-efficient chips enable these sovereign clouds to operate effectively, addressing national security concerns and allowing countries to develop AI models tailored to their own legal and cultural contexts.
Looking ahead, Rodrigo envisions a future where every individual relies on personalized AI agents to enhance productivity and daily life. These agents will perform tasks such as composing emails, summarizing information, and making reservations, fundamentally changing how people interact with technology. He predicts that within five years, AI agents will become indispensable tools, freeing humans from routine tasks and enabling greater creativity and connection. This vision underscores SambaNova’s mission to power the next generation of AI applications with fast, efficient inference technology.