VCs pick deep-tech standouts from Y Combinator's summer batch
Y Combinator's latest Demo Day featured a batch skewed toward deep tech, with startups ranging from floating nuclear-powered data centers to optical networking hardware. VCs highlighted nine buzziest startups, including Automarine, Dipole Labs, and Isengard Industries, noting valuations were more grounded than recent cohorts. The startups address power shortages, AI data center efficiency, and drone production.
The deep-tech tilt marks a notable departure for Y Combinator, a program historically associated with software startups. Investors characterized the ideas as feeling "like science fiction," yet valuations were described as more grounded than in recent cohorts, suggesting a cooling of the speculative fervor that had marked earlier batches. The common thread across many picks is the AI infrastructure bottleneck — power, data movement, and compute efficiency.
Several startups attack this problem from distinct angles. Automarine's floating data centers address power scarcity and community opposition to land-based facilities, claiming over $4 billion in customer interest. Dipole Labs targets the energy wasted in optical-to-electrical conversions within GPU clusters. Lamb Labs hardcodes model weights into silicon to eliminate memory bandwidth constraints. Isengard Industries, already generating $10 million in revenue, focuses on defense drone production within allied countries.
This batch's emphasis on physical infrastructure — power generation, networking hardware, drone manufacturing — could signal a shift in venture priorities toward capital-intensive deep tech over pure software. If these startups succeed, they may help alleviate AI's growing energy demands and data center siting conflicts, potentially affecting energy markets and local communities. However, long timelines, such as Automarine's 2032 nuclear target, mean near-term societal impact may be limited. The more grounded valuations could foster a healthier, more sustainable startup ecosystem.