The 9 buzziest startups from Y Combinator’s latest Demo Day, according to VCs
Y Combinator’s latest Demo Day featured a cohort notably heavy on deep-tech startups, drawing investor attention for ideas from floating data centers to biologically based computing. TechCrunch polled early-stage VCs and identified nine companies that multiple investors flagged as the buzziest in the summer batch.

Why It Matters
The cohort’s concentration on capital- and science-intensive hardware and novel compute approaches highlights investor interest in solving core infrastructure constraints — power, cooling, data movement and physical labor — which could reshape how AI and robotics scale. VCs also noted that, despite ambitious technical visions, valuations in this batch appeared more conservative than in recent cohorts.
Key Facts
- Number of buzziest startups: 9 startups flagged by at least two investors
- Cohort character: Skewed toward deep tech compared with prior YC batches
- Investor quote on tech: An investor described the technology in the batch as "like science fiction"
- Valuation note: VCs said valuations were more grounded than in recent cohorts
- Automarine plans: Gas-powered pilot by 2028; transition to floating nuclear power ships in 2032; claims over $4 billion in customer interest via letters of intent (per company)
Y Combinator’s summer Demo Day tilted strongly toward hardware and infrastructure projects, according to investor feedback collected by TechCrunch. Instead of the more common consumer or SaaS ideas, many startups in this batch proposed capital-intensive solutions addressing energy, cooling, data movement and physical automation — areas VCs say are critical as AI and robotics scale. Among the companies that drew repeated investor mentions was Automarine, which proposes data centers on barges to exploit seawater cooling and plans a gas-powered pilot by 2028 before moving to floating nuclear power ships in 2032; the startup says it has secured more than $4 billion in customer interest through letters of intent. On the networking front, Dipole Labs presented an optical switch that keeps data in light form to avoid costly light-to-electrical conversions inside GPU clusters, aiming to improve energy efficiency and throughput in AI data centers. Several robotics and defense-focused startups also stood out. Isengard Industries is building jet-powered strike and counter-drones with an emphasis on local production and lower cost, and the company already reports $10 million in revenue. Cosmic Robotics is developing heavy-lift autonomous machines — used today for solar-installation work and supported by $25 million in contracts through 2027 — and frames that work as a step toward automating construction for future Mars missions. Meanwhile, Praxis AI is collecting large-scale video and operational data from real workplaces to train robots for real-world tasks, claiming data from more than 150 environments and partnerships with publicly traded firms. On the compute-hardware side, startups are pitching radical efficiency gains. Lamb Labs says it is building inference chips with model weights hardcoded into silicon to eliminate memory-bandwidth bottlenecks, while Parasma is exploring the use of human brain cells as an energy-efficient substrate for computation. Waddle Labs proposes an alternative path for general robotics by using layers of LLM agents to write control code, and consumer robotics newcomer Nori has launched a humanoid priced around $1,600, reporting nearly $500,000 in sales within weeks of launch. Across these ideas, investors singled out a mix of blue-sky ambition and commercial traction that made the batch one of the most talked-about yet.
Keep Reading

What’s behind the AI industry’s latest warnings of doom?

Insight Partners’ Devin Parekh on why the firm is diversifying while everyone else bets the farm on OpenAI and Anthropic

GOP rep on AI: ‘Let’s get on top of it’
