Y Combinator's latest Demo Day spotlighted a cohort that leaned strongly into deep tech, with investors noting that the ideas felt ambitious yet the pricing discipline was more grounded than in recent batches.
Among the most talked-about startups was Automarine, which is developing floating, nuclear-powered data centers designed to operate at sea. The concept aims to address rising compute demand, power constraints, and cooling challenges by using offshore infrastructure and seawater-based thermal management.
Dipole Labs drew attention with optical networking hardware built for AI data centers. Its approach keeps data in light form longer, reducing energy loss and latency in the movement of information between chips.
In robotics and automation, several companies stood out. Praxis AI is gathering real-world human work data to train robots more effectively, while Nori is targeting everyday household tasks with an affordable humanoid robot priced far below many competitors. Cosmic Robotics is building autonomous machines for heavy lifting and industrial work, with applications already tied to solar installation.
Lamb Labs is pursuing custom inference chips that hardcode model weights into silicon to improve efficiency, and Waddle Labs is creating an API layer that can generate robot control code from natural language instructions. Another notable entrant, Parasma, is exploring brain-cell-based computing as a long-term path toward lower-energy AI systems.
Across the batch, the common thread was clear: founders are pushing toward infrastructure, robotics, and compute systems that could reshape how intelligence is powered and deployed. If these ideas mature, they may help define the next phase of efficient, scalable technology.
