YC Demo Day spotlights deep-tech startups in compute and robotics
Y Combinator’s latest Demo Day highlighted nine startups that venture investors repeatedly identified as the batch’s most discussed companies. The selections span floating data centers, optical networking hardware, autonomous robotics, drones, specialised AI chips and biological computing, marking a notably deeper-tech cohort than many recent YC classes.
Investors told TechCrunch that the technology felt “like science fiction,” but they also saw more grounded valuations than in recent batches. The companies were chosen after being flagged by at least two early-stage investors, rather than through a ranking of their commercial prospects.
Compute constraints shape the batch
Atomarine is developing nuclear-powered data centers on offshore barges. It argues that sea-based facilities could ease pressure on constrained power supplies, avoid local resistance to new data-center construction and use seawater for cooling. The company plans a gas-powered pilot in 2028, followed by floating nuclear power ships in 2032, and says letters of intent represent more than $4 billion in customer interest.
Dipole Labs is addressing another data-center bottleneck: movement of data among GPUs. Its proposed optical switch keeps data in optical form rather than repeatedly converting it to electricity and back, a process the company says creates power use and heat. Lamb Labs is pursuing efficiency at the chip level with Model Processing Units, custom inference chips designed to hardcode AI model weights into silicon and reduce memory-bandwidth constraints.
Parasma also focuses on computing energy requirements, but through research into training human brain cells to potentially power computing. These approaches reflect the range of technical bets emerging around the cost, power availability and throughput limits associated with AI infrastructure.
Robotics, drones and physical deployment
Several of the selected companies are focused on machines operating beyond the data center. Isengard Industries says it will mass-produce jet-powered strike and counter-drones within allied countries. The company is already generating $10 million in revenue, while its founders’ experience includes an Australian Army background and a Ukraine-focused drone business.
Praxis AI gathers real-world video and work data for robot-training customers. It says it works with publicly traded companies and has recorded data in more than 150 environments. Waddle Labs takes a different route: it uses LLM agents to generate, test and deploy robot-control code through an API, with the company claiming developers can configure a robot in about 20 minutes.
The wider YC robotics pipeline, including YC Spring 2026 Demo Day robotics startups, shows how quickly investor attention is moving from software-only products toward systems that must perform reliably in physical settings. Nori has launched a roughly $1,600 humanoid robot aimed at cleaning and folding clothes, while Cosmic Robotics says its heavy-lifting machines are installing solar panels in the US and holds $25 million in contracts through 2027.
What businesses should assess
For businesses evaluating this wave, the practical question is not whether every long-range promise will reach production. It is whether a supplier can demonstrate a deployable system, clear energy or operating benefits, and credible evidence of demand, contracts or revenue before it becomes part of a critical technology plan.