Deeptech startup Zenteiq is targetting high-value industrial use cases such as battery design and semiconductor analysis, as the startup begins training its AI models on compute infrastructure, founder Sashikumar Ganesan said in an interaction with Fe.
Supported by the IndiaAI Mission, Zenteiq is focusing on integrating physics-based modelling with AI systems, targeting applications such as thermal and electromagnetic analysis in sectors including semiconductors, automotive and aerospace. “Today’s generative AI systems are very strong in language tasks—text, audio, code—but when it comes to predictive analytics in domains like thermal or aerospace engineering, they fall short,” Ganesan said adding the firm is trying to bring scientific rigour and predictability into AI models.
Google TPU-based compute expansion
The company recently secured compute support of over 2,000 TPUs through Ishan Technologies. The infrastructure is based on Google’s tensor processing units (TPUs).
Zenteiq has begun early-stage training and benchmarking of its models, with initial releases expected next year. It has completed proof-of-concept (PoC) projects with an automotive original equipment manufacturer (OEM). These projects focus on battery design, including thermal behaviour during fast charging and optimisation of operating conditions.
“By varying parameters like temperature and environmental conditions, we can improve battery efficiency during the design stage itself,” he said.
AI “surrogate models” to cut experiment cycles
The firm’s models complement existing simulation tools and are aimed at reducing the number of physical experiments and computational cycles required in product development. “Instead of running hundreds of simulations or physical experiments, we are building models that can act as a surrogate and reduce the number of cycles needed,” Ganesan said.Separately, Zeteiq has unveiled an AI platform called AhamX, which connects learners, content creators, academic institutions and industry stakeholders.“The platform integrates skilling, training and industry connect into a single ecosystem,” Ganesan said, adding that the commercial rollout of its core AI models is expected to follow.