Top 9 Hottest Startups Unveiled at Y Combinator’s Latest Demo Day, as Ranked by Venture Capitalists

On Thursday, Y Combinator held its latest Demo Day, showcasing a batch of startups that focused heavily on deep technology, a shift from previous years.

As is customary each quarter, TechCrunch reached out to early-stage venture capitalists to identify the standout startups from this cohort—both their favorites and those generating buzz. The following list includes startups that received notable mentions from multiple investors.

Commenting on the technology featured, one investor remarked that it felt “like science fiction.” Nevertheless, there was a general agreement that the valuations for these startups were more realistic compared to past batches.

Here are the highlighted startups listed alphabetically:

Automarine

What it’s building: Nuclear-powered data centers located offshore.

Why it’s a favorite: With increasing opposition to new data centers, Automarine aims to address the data shortage by placing data centers on barges at sea, utilizing seawater for cooling. Co-founded by experts from MIT, they plan to pilot a gas-powered model by 2028 and transition to floating nuclear versions by 2032. They claim to have already garnered over $4 billion in letters of intent from potential customers, contributing to a high valuation.

Dipole Labs

What it’s building: Efficient optical networking hardware for AI data centers.

Why it’s a favorite: Dipole Labs has developed an optical switch that allows data to remain in light form, avoiding the energy-intensive conversion to electricity during transmission. This innovation is timely given the high costs of GPUs and the need for efficient data management in computing.

Isengard Industries

What it’s building: Jet-powered drones designed for attack and defense.

Why it’s a favorite: Isengard aims to manufacture these drones within allied nations, significantly reducing costs compared to U.S. contractors. Co-founded by a former Australian Army officer and a seasoned defense entrepreneur, Isengard is already showing revenue of $10 million and attracting significant interest from investors.

Lamb Labs

What it’s building: Specialized inference chips with pre-loaded AI model weights.

Why it’s a favorite: Lamb Labs focuses on enhancing energy efficiency in AI chips by embedding model weights in silicon. Known as “Model Processing Units” (MPUs), these chips tackle memory bottlenecks commonly seen in traditional hardware.

Praxis AI

What it’s building: A system for gathering real-world data to train robots.

Why it’s a favorite: This company collaborates with businesses to collect video data of human tasks, using this material for robot training. With contributions from publicly traded enterprises, Praxis AI has documented data from over 150 environments, positioning itself uniquely in an evolving market.

Nori

What it’s building: Economical robots for daily household chores.

Why it’s a favorite: In its first six weeks, Nori has achieved nearly $500,000 in sales with a humanoid robot designed to assist with laundry. Priced at about $1,600, it offers an alternative to more expensive models. The viability of affordable at-home robots for mundane tasks remains a significant issue in robotics.

Cosmic Robotics

What it’s building: Autonomous robots capable of lifting heavy loads.

Why it’s a favorite: Cosmic Robotics is aiming to develop robotic solutions that facilitate tasks needed for Martian colonization. They are already utilizing their technology for installing solar panels in the U.S. and have secured contracts worth $25 million through 2027, aligning their ambitions with SpaceX’s plans for Mars.

Parasma

What it’s building: Uses human brain cells for computing power.

Why it’s a favorite: With an eye on improving energy efficiency in computing, Parasma investigates leveraging human brain cells as a viable alternative to conventional AI hardware.

Waddle Labs

What it’s building: An API layer for robot control code generation.

Why it’s a favorite: Waddle Labs aims to create a revolutionary approach to robotics programming by using large language models to write executable code from natural language commands. Founded by Harvard alumni, the startup envisions simplifying robot programming for developers.