The 9 buzziest startups from Y Combinator’s latest Demo Day, according to VCs
Y Combinator’s newest Demo Day showcased nine deep‑tech startups that attracted significant VC interest. The cohort features innovations ranging from floating nuclear data centers to AI‑optimized chips, all with realistic valuations and strong market potential.
Y Combinator’s most recent Demo Day, held on Thursday, drew attention for its unusually deep‑tech focus. While every cohort brings fresh founders, this batch leaned heavily into science‑fiction‑like solutions that promise to reshape industries. Venture capitalists were quick to single out nine startups that stood out as the most compelling, each offering a unique approach to solving pressing problems in data, defense, robotics, and AI.
Floating Nuclear Power for Data Centers
Atomarine, co‑founded by an MIT computer science and naval engineer and an MIT nuclear engineering PhD, proposes a radical solution to the growing demand for data center cooling and power. The company plans to launch a gas‑powered pilot by 2028, with a transition to floating nuclear power ships slated for 2032. By situating data centers on barges at sea, Atomarine can tap seawater for near‑free cooling and reduce the environmental footprint of traditional land‑based facilities. Early letters of intent have already generated over $4 billion in projected revenue, positioning Atomarine as one of the highest‑valued startups in the cohort.
Optical Networking to Cut AI Power Use
Dipole Labs tackles a hidden cost in AI data centers: the energy wasted when data moves between GPUs. Their optical switch eliminates the conversion between light and electricity, allowing data to travel as photons directly to the required chip. This innovation promises to keep GPUs running at full capacity without the heat and power drain associated with traditional electrical interconnects, a critical advantage as GPU costs continue to climb.
Mass‑Produced Drone Defense
Isengard Industries is reimagining drone production by localizing manufacturing within allied countries. Co‑founded by a former Australian Army officer and a defense entrepreneur who previously scaled a Ukraine‑focused drone company to $60 million in revenue, Isengard already generates $10 million annually. The firm’s jet‑powered attack and counter‑drones can be produced at a fraction of U.S. contractor costs, earning it a high valuation and strong VC interest.
AI Chips With Hard‑Coded Models
Lamb Labs, founded by an Imperial College London AI PhD and an Oxford theoretical physicist, is developing custom inference chips that embed AI model weights directly into silicon. These “Model Processing Units” (MPUs) eliminate memory bandwidth bottlenecks, dramatically reducing energy consumption during inference. By hard‑coding models, Lamb Labs aims to make AI deployment more efficient and cost‑effective for enterprises.
Robotics Training From Real‑World Video
Praxis AI collects videos of humans performing work across more than 150 environments and converts them into training data for robotic systems. The company partners with publicly traded firms to provide realistic, diverse datasets that help robots learn tasks in varied settings. As businesses explore which jobs are best suited for automation, Praxis AI’s data could become a cornerstone of next‑generation robotics.
Affordable Home Robots
Nori, launched just six weeks ago, offers a humanoid robot that cleans and folds laundry for around $1,600. The company has already recorded nearly $500,000 in sales, indicating strong consumer demand for low‑cost domestic robots. Nori’s focus on everyday chores addresses a key barrier to widespread home robotics adoption: affordability.
Heavy‑Duty Construction Robots for Mars
Cosmic Robotics is building autonomous robots capable of lifting heavy objects, a technology essential for future Mars colonization. The founders envision a future where robots install solar panels and construct habitats on the Red Planet. With $25 million in contracts through 2027, Cosmic Robotics is racing to align its timeline with SpaceX’s Mars ambitions, targeting an exploratory mission by 2028.
Brain‑Powered Computing
Parasma explores the use of human brain cells as a more energy‑efficient alternative to silicon for running AI models. By training neural tissue to perform computation, the startup seeks to reduce the power demands that currently limit large‑scale AI deployment.
AI‑Generated Robot Control Code
Waddle Labs offers an API that translates natural‑language commands into executable robot control code. Founded by Harvard graduates, the platform uses large language model agents to write, test, and deploy code in under 20 minutes. By abstracting hardware details, Waddle Labs aims to democratize robotics development and accelerate the arrival of a “ChatGPT” moment for robots.
Across these nine startups, a common thread emerges: each company is tackling a specific bottleneck in its field with a bold, technology‑driven solution. The cohort’s realistic valuations and clear customer interest suggest that Y Combinator’s newest batch is poised to deliver tangible impact rather than speculative hype.
Why it matters
These startups illustrate how deep‑tech innovations can address real‑world challenges—from energy‑efficient data centers to affordable domestic robots—potentially reshaping multiple industries and driving economic growth.
Key points
- Atomarine’s floating nuclear data centers could solve cooling and power shortages.
- Dipole Labs’ optical switch reduces AI data center energy use.
- Isengard Industries localizes drone production, cutting costs.
- Lamb Labs’ hard‑coded AI chips improve inference efficiency.
- Praxis AI provides real‑world video data for robotic training.
- Nori offers an affordable home robot for everyday chores.
- Cosmic Robotics builds heavy‑lift robots for Mars construction.
- Parasma explores brain cells for energy‑efficient computing.
