Y Combinator’s latest Demo Day featured a cohort of startups that leaned more heavily into deep technology than previous batches. Early-stage venture capitalists identified several companies as the most promising, highlighting innovations that address pressing challenges in computing, robotics, and energy.
One standout is Atomarine, which aims to deploy data centers on barges powered initially by gas and eventually by floating nuclear reactors. This approach leverages seawater for cooling and addresses power shortages and community resistance to new land-based data centers. The company has reportedly secured over $4 billion in potential customer interest.
Dipole Labs is developing optical networking hardware designed to improve data transfer efficiency within AI data centers by eliminating the energy-intensive conversion between light and electricity, potentially enhancing GPU cluster performance.
Isengard focuses on producing jet-powered strike and counter-drones domestically within allied countries at a lower cost than traditional defense contractors. The startup has already generated $10 million in revenue and is noted for its high valuation.
Lamb Labs is creating custom AI inference chips with model weights hardcoded into silicon, aiming to reduce energy consumption and overcome memory bandwidth limitations common in current AI hardware.
Another company is collecting extensive real-world video data of human work activities to train robots, assisting businesses in determining which tasks can be automated effectively.
Nori Robotics has introduced an affordable humanoid robot capable of household chores like cleaning and folding clothes, priced significantly lower than comparable products, signaling progress toward practical home robotics.
Ramp Robotics is developing autonomous machines capable of heavy lifting and construction tasks, with contracts to install solar panels and ambitions to support Mars colonization efforts.
Parasma is exploring the use of human brain cells as a novel, energy-efficient computing substrate for AI workloads, addressing the high power demands of current AI models.
Waddle Labs offers an API that uses large language model agents to generate robot control code from natural language commands, simplifying robot programming and setup.
This cohort’s focus on deep tech innovations reflects a growing investor interest in startups that combine scientific research with scalable commercial applications, potentially shaping the future of computing, robotics, and energy infrastructure.