Today, we announced a new customer relationship with the Texas Advanced Computing Center (TACC), one of the world’s leading supercomputing centers, to enhance its AI capabilities for use in scientific research applications.
The deal calls for TACC to deploy SambaNova Suite, which combines our DataScale hardware, SambaStudio software, and our Composition of Experts (CoE) architecture for running multiple AI models, as well as for AI inference. At its heart runs the SN40L Reconfigurable Dataflow Unit (RDU) chip, the world’s most flexible and energy-efficient silicon for AI and an alternative to GPUs.
“SambaNova will be our platform for inference on scientific applications,” said Dan Stanzione, Associate Vice President For Research Executive Director at TACC. “We will use SambaNova Suite to host the models we’ve trained on traditional supercomputers to integrate AI inference into the science workflow.”
Based at the University of Texas at Austin and founded in 2001, TACC is home to Frontera, the most powerful supercomputer running on a university campus in the U.S. Funded by the National Science Foundation, its user base of scientists includes four Nobel Prize winners.
TACC is also home to the NSF Leadership-Class Computing Facility (LCCF), which will begin operation in 2026. LCCF will form a key plank of the National Artificial Intelligence Research Resource (NAIRR), an NSF-led initiative coordinated across several federal agencies and other organizations. It’s intended to accelerate the research and development environment for AI in the U.S.
TACC’s supercomputers are used every year for thousands of different simulation, modeling, and data analysis projects. However, as its scientists began using AI, they found that while traditional supercomputers are good at AI training, they weren’t as good for running AI inference.
“We’ll use SambaNova to host the models we’ve trained on traditional supercomputers to integrate AI inference into the science workflow,” Stanzione said. “It will be our dedicated inference platform.”
TACC will run models that stay resident, meaning they’re always on and available for researchers to use. Some are specialized, others are chatbots trained on scientific data. “Our scientists want a model that they can talk to all the time,” Stanzione said. “We need a platform where those models can stay resident and do fast inferencing as part of other scientific workflows.”
Another feature TACC will use is SambaNova’s Composition of Expert (CoE) architecture, which allows the use of many AI models at once, both directly from the SambaNova interface and via its API. Eventually, the platform will host specialized scientific models directly with the ability to conduct inference as part of the process.
“It’s the next step in the evolution of what our users expect from our computing services,” Stanzione said. “Our mission has always been to provide the full computing ecosystem that scientists want and need. And that is evolving into something more interactive and reliant on AI. We think it could be a fundamental shift in how the scientific community interacts with computing.”