To set new reliability standards for data centres worldwide, a semiconductor startup is redefining AI power delivery, optimising voltage conversion and efficiency at unprecedented scales.
A Bengaluru-based start-up, C2i, is developing semiconductor solutions for artificial intelligence (AI) infrastructure, targeting one of the sector’s toughest challenges: efficient power delivery for hyperscale data centres and graphics processing unit (GPU) clusters, an area that remains far less mature than technologies in electric vehicles (EVs), industrial automation, and drones.
In an exclusive conversation with Preetam Tadeparthy, Founder and Chief Technology Officer (CTO) of C2i, he explained that the first generation of products, scheduled for silicon tape-out in 2026, focuses on voltage regulation and control for GPUs.
The first generation of products, scheduled for silicon tape out in 2026, focuses on voltage regulation and control for GPUs. The design uses a software-defined approach, with proprietary algorithms managing switching behaviour, power density, and response times. “This allows a more adaptable way to regulate power compared to traditional voltage systems,” Preetam added. Operating as a fabless startup, the team uses industry-standard electronic design automation (EDA) tools such as Cadence, Synopsys, and Siemens, while fabrication and packaging are handled through external foundries and ecosystem partners.
“Our proprietary intellectual property (IP) is a key differentiator, contributing to a reported 10-15% system-level improvement in GPU performance per watt. This comes from innovations in fast-switching control, power-conversion stages, and how GPUs handle dynamic sleep and performance transitions. The result is not only higher performance but also extended server lifespan, potentially doubling the usable life of AI infrastructure,” said Preetam.
The startup claims to follow a multi-stage power-conversion approach for now, focusing on the final stage closest to the processor, while future designs aim to reduce intermediate stages and enable direct 800-volt-to-sub-1-volt conversion. This will involve a hybrid approach using silicon for low voltages and partnering with gallium nitride (GaN) providers for high-voltage stages. C2i also emphasises software-defined voltage regulation, making its solution scalable across a wide range of power requirements, from hundreds of watts to several kilowatts per server or rack.
C2i is also working on improving the uptime and reliability of AI servers. Its modular solutions aim to be plug-and-play, self-tuning, and capable of operating continuously without downtime. Combined with rigorous pre-silicon validation and robust design for thermal and electrical performance, the company is targeting next-generation AI clusters with scalable efficiency.
