The AI boom has a dirty secret: data centers are hemorrhaging power before it ever reaches the GPUs doing the actual work. Indian semiconductor startup C2i just landed $15 million from Peak XV Partners and TDK Ventures to fix that bottleneck, testing a grid-to-GPU approach that could reshape how hyperscalers power their AI infrastructure. As training runs push facilities past their electrical limits, C2i's timing couldn't be sharper.
C2i Semiconductors just closed a $15 million round led by Peak XV Partners (formerly Sequoia India) with participation from TDK Ventures, betting that the next AI infrastructure battle won't be fought over chips - it'll be about the power flowing into them.
The Indian startup is tackling a problem that's become painfully obvious to anyone running large language models at scale: conventional data centers waste massive amounts of electricity converting high-voltage grid power down to the low-voltage direct current that GPUs actually need. Those conversion losses add up fast when you're running tens of thousands of accelerators, and they're pushing facilities straight into their power capacity walls.
C2i's approach bypasses traditional power distribution architectures entirely. Instead of stepping voltage down through multiple conversion stages - each one bleeding efficiency - the company's semiconductor technology handles the grid-to-GPU transformation more directly. It's the kind of unglamorous infrastructure innovation that doesn't make flashy demo videos but could determine which hyperscalers can actually afford to keep scaling their AI ambitions.
The timing makes sense. Microsoft, Google, and Amazon are all scrambling to secure power contracts for new data center builds, with some projects stalled entirely because local grids simply can't deliver enough juice. Meta has reportedly walked away from potential sites after hitting electrical capacity constraints. Making better use of available power isn't just an optimization - it's becoming existential for AI infrastructure players.
