Y Combinator’s latest Demo Day took place on Thursday, and this batch of founders stood out for leaning much more heavily into deep technology than recent cohorts.
TechCrunch asked early‑stage venture capitalists to name the most promising companies from the cohort. Startups that at least two investors flagged as buzziest made the list. Several investors described the technologies on display as “like science fiction,” while also noting that valuations appeared more grounded than in recent rounds.
Below are the most‑mentioned startups, listed alphabetically, with what they are building and why they attracted investor attention.
Atomarine
- What they’re building: sea‑based data centers powered by nuclear energy
- Why it’s notable: growing local opposition to land‑based data center buildouts and power constraints have increased interest in alternative deployments. Atomarine’s founders include an MIT graduate with computer science and naval engineering credentials and an MIT PhD student in nuclear engineering. The company plans a gas‑powered pilot by 2028 and aims to transition to floating nuclear power ships in 2032. Atomarine says it has secured more than $4 billion in potential customer interest via letters of intent, a claim that one VC cited as a driver of the startup’s high valuation within the cohort.
Dipole Labs
- What they’re building: energy‑efficient, high‑speed optical networking hardware for AI data centers
- Why it’s notable: GPU clusters lose time waiting for data transfers, and conventional networking converts data between light and electricity, consuming power and producing heat. Dipole Labs says it has developed an optical switch that avoids that conversion so data can remain in light form and be routed directly, reducing latency and energy use — a timely solution given the high cost of GPUs and the pressure to maximize compute utilization.
Isengard Industries
- What they’re building: locally producible, jet‑powered attack and counter‑drone systems
- Why it’s notable: Isengard aims to mass‑produce jet‑powered offensive and counter‑drone platforms inside allied nations at a fraction of the cost charged by prime contractors. The founders include a former Australian Army officer and a defense entrepreneur who previously scaled a Ukraine‑focused drone startup to $60 million in revenue. Isengard itself is reportedly generating $10 million in revenue and, according to two investors, commands one of the loftier valuations in the YC batch.
Lamb Labs
- What they’re building: custom inference chips with hardcoded AI model weights
- Why it’s notable: inference workloads spend significant energy fetching model weights from memory. Lamb Labs, founded by an Imperial College London AI PhD and an Oxford theoretical physicist, proposes ultra‑efficient chips — called Model Processing Units (MPUs) — that hardcode model weights into silicon to eliminate memory‑bandwidth bottlenecks.
Praxis Robotics
- What they’re building: collecting real‑world video and data to train robots
- Why it’s notable: Praxis partners with companies to capture video and data of humans performing work and turns that material into training datasets for robot developers. The startup says it is already working with publicly traded companies and has collected video in more than 150 different environments. That dataset could be valuable as businesses experiment with which tasks are best for humans versus robots.
Nori
- What they’re building: affordable robots for everyday tasks
- Why it’s notable: launched only six weeks ago, Nori reports nearly $500,000 in sales. Its humanoid robot promises to help clean and fold clothes and can be controlled via a laptop app. Priced at about $1,600, it is far cheaper than other humanoids such as Neo, which is priced around $20,000. A central question in consumer robotics is whether a practical, affordable at‑home humanoid is achievable; Nori represents another attempt to meet that challenge.
Cosmic Robotics
- What they’re building: autonomous robots able to lift heavy objects
- Why it’s notable: the founders have a long‑term vision of building a city on Mars; the nearer‑term objective is robotics technology for heavy‑duty tasks. The company says its technology is already being used to install solar panels across the U.S., and it has $25 million in contracts through 2027. Cosmic Robotics aims to start an exploratory mission by 2028, putting its timeline in apparent competition with SpaceX’s Mars ambitions.
Parasma
- What they’re building: training human brain cells to perform computation
- Why it’s notable: Parasma is exploring whether human neuronal tissue can serve as a lower‑power alternative to current AI hardware, representing another approach to the industry’s energy‑efficiency challenge.
Waddle Labs
- What they’re building: an API layer that writes robot control code
- Why it’s notable: investors and technologists are looking for the equivalent of a ChatGPT moment in robotics. Waddle Labs takes a different route by using LLM agents to generate executable control code rather than training foundation models on raw video or teleoperation data. Founded by Harvard graduates, the company markets itself as “Claude Code for robotics.” It claims that by hooking hardware into Waddle’s API, developers can instruct robots in natural language; the AI agents will autonomously generate control code, verify execution, and set up the robot in roughly 20 minutes.
Why this cohort matters
The startups in this YC batch span a range of hard‑tech domains — from optics and specialized silicon to maritime nuclear infrastructure and biologically based compute. Some firms already show revenue traction (for example, Isengard and Nori), while others target longer timelines for technological transitions (Atomarine’s planned nuclear ships, Parasma’s brain‑cell compute). Collectively, they suggest renewed investor appetite for capital‑intensive, deep‑tech approaches even as valuations appear more measured than in previous cohorts.
