Analysis
Y Combinator's latest Demo Day, held September 13, produced nine startups that at least two early-stage VCs independently flagged as the batch's buzziest, TechCrunch reported -- a cohort investors described as skewing far more toward deep tech than recent batches, with valuations they characterized as more grounded than the AI-application-heavy cycles of the past two years. Y Combinator has now run this same VC-survey format through several consecutive Demo Days, and this round's list looks structurally different from June's batch of 11 standouts, which leaned more toward software and agent tooling.
The nine companies, by focus:
- Atomarine -- floating, nuclear-powered data centers at sea, co-founded by an MIT computer science and naval engineering graduate and an MIT PhD candidate in nuclear engineering; the company says it has more than $4 billion in customer interest via letters of intent, with a gas-powered pilot planned for 2028 ahead of a nuclear transition targeted for 2032
- Dipole Labs -- optical networking hardware for AI data centers that avoids converting light to electricity and back, aimed at cutting the energy and heat overhead of high-bandwidth interconnects
- Isengard Industries -- locally manufactured jet-powered strike and counter-drone systems, founded by a former Australian Army officer who previously scaled a Ukraine-focused drone startup to $60 million in revenue; Isengard itself already has $10 million in revenue and investors called it one of the highest-valued companies in the batch
- Lamb Labs -- custom inference chips with AI model weights hardcoded directly into silicon ("MPUs"), founded by an Imperial College London AI PhD and an Oxford theoretical physicist
- Praxis Robotics -- real-world data collection for robot training, already working with publicly traded companies and reporting more than 150 captured environments
- Nori -- an affordable, roughly $1,600 humanoid robot for household chores like cleaning and folding laundry, reporting about $500,000 in sales just six weeks after launch
- Cosmic Robotics -- autonomous heavy-lifting robots, currently installing solar panels across the US under $25 million in contracts through 2027, with an exploratory Mars mission planned by 2028
- Parasma -- computing built on human brain cells, pitched as a far more energy-efficient substrate for AI workloads than silicon
- Waddle Labs -- an API layer, founded by two Harvard graduates, that generates robot control code from natural-language instructions, positioned as "Claude Code for robotics"
The through-line across most of these nine is physical infrastructure and hardware rather than pure software -- floating data centers, counter-drone hardware, custom silicon, humanoid robots and heavy-lift robotics all require capital-intensive manufacturing and longer paths to revenue than a typical SaaS batch company, a notable shift from YC's historical software-first center of gravity.
Isengard's reported $10 million in revenue stands out against a batch where most companies are pre-revenue or barely past launch -- Nori's $500,000 in six weeks is a strong early signal for a hardware product at that price point, but it's a fraction of Isengard's run rate. Defense-adjacent hardware companies with a founder who has already scaled a comparable business once before, in Isengard's case a Ukraine drone startup, tend to raise at a premium to unproven first-time founders in the same category, which is consistent with VCs flagging it as one of the highest-valued companies in the batch.
Atomarine's $4 billion-plus in letters of intent is the number likely to draw the most skepticism: LOIs are non-binding expressions of interest, not contracted revenue, and a pre-pilot floating-nuclear-data-center concept collecting that scale of interest says more about how starved the market is for AI power capacity than about Atomarine's actual execution risk, which remains almost entirely unproven at this stage.
TechCrunch's reporting notably didn't name any specific VC by quote, describing sourcing only as "one investor" or "two investors" throughout -- a reminder that buzz rankings compiled this way reflect informal VC chatter rather than disclosed term sheets, and none of the nine companies' actual valuations were confirmed on the record.
