Y Combinator Demo Day: 9 Deep Tech Startups Drawing VC Buzz
Y Combinator’s latest Demo Day spotlights nine deep tech startups across nuclear data centers, robotics, optical chips and brain cell computing hardware.
Summary
Y Combinator’s latest Thursday Demo Day, current as of September 13, 2026, skewed toward deep tech. Nine companies drew mentions from at least two early stage VCs, whose consensus paired science fiction ambitions with more grounded valuations than recent cohorts.
Automarine, founded by an MIT computer scientist and naval engineer with an MIT nuclear engineering Ph.D., proposes seawater cooled floating data centers, a gas pilot by 2028 and nuclear power ships in 2032. Letters of intent exceed $4 billion, supporting one of the batch’s highest valuations. Dipole Labs’ optical switch keeps AI cluster traffic as light, avoiding heat and power intensive electrical conversion while reducing expensive GPU idle time. Isengard Industries, led by a former Australian Army officer and an entrepreneur who grew a Ukraine focused drone startup to $60 million revenue, locally mass produces lower cost jet strike and counter drones in allied countries. It has $10 million revenue and another top valuation. Lamb Labs’ Imperial College London AI Ph.D. and Oxford theoretical physicist hardcode model weights into custom Model Processing Units, removing inference memory bandwidth bottlenecks.
Praxis AI converts workplace videos into robot training data, works with publicly traded companies and has covered more than 150 environments. Six week old Nori reports nearly $500,000 in sales for its roughly $1,600 laptop controlled humanoid, designed to clean and fold clothes, versus about $20,000 for Neo. Cosmic Robotics’ heavy lifting autonomous machines install U.S. solar panels, with $25 million in contracts through 2027. Pursuing Mars construction alongside SpaceX, it targets a 2028 exploratory mission. Parasma is training human brain cells as lower energy AI hardware. Harvard founded Waddle Labs uses LLM agents instead of raw video or teleoperation models. Its Claude Code for robotics API claims any hardware can convert natural language into tested control code and complete setup in about 20 minutes.
Positives
- Automarine claims more than $4 billion in customer interest for floating data centers, helping secure one of the cohort’s highest valuations.
- Isengard Industries already generates $10 million in revenue from locally producible jet strike and counter drones.
- Nori reached nearly $500,000 in sales within six weeks while pricing its home humanoid at roughly $1,600.
- Cosmic Robotics has secured $25 million in contracts through 2027 for machines already installing U.S. solar panels.
- Praxis AI has collected robot training data across more than 150 environments and works with publicly traded companies.
- Waddle Labs says its API can configure robots from natural language in about 20 minutes.
Risks & concerns
- Automarine’s first gas pilot is not planned until 2028, while its transition to floating nuclear power ships must wait until 2032.
- Automarine’s more than $4 billion represents nonbinding customer interest through letters of intent, not recognized revenue.
- Nori’s early sales do not establish whether its $1,600 humanoid can reliably perform difficult household jobs such as stacking dishwashers.
- Cosmic Robotics’ 2028 Mars mission target adds an ambitious, untested objective beyond its current solar installation work.
- Parasma disclosed no timeline, performance results or commercialization plan for computing with trained human brain cells.
- Waddle Labs’ universal hardware compatibility and 20 minute setup remain company claims without disclosed customer or performance evidence.