Dimension Capital, a New York-based venture firm co-founded by Zavian Dar, Adam Goulburn, and Nan Li, has closed $800 million for its third fund, according to Venture Daily Digest. The fund targets companies operating at the intersection of science and computing, with a focus on AI infrastructure, biotech, drug discovery, and deep tech.
The Infrastructure Capital Wave
The close comes during a concentrated surge of venture capital flowing into AI infrastructure. In the past 72 hours alone, Khosla Ventures reportedly began raising up to $5.5 billion across multiple vehicles for early-stage AI infrastructure companies. PaleBlueDot AI closed a $255 million credit facility from Brookfield Asset Management. SkyPilot raised $20 million for cloud orchestration and GPU cost-control tooling.
Dimension’s thesis sits at a specific intersection: the computational layer where AI meets scientific discovery. Drug discovery, protein folding, materials science, and climate modeling all require specialized compute infrastructure that overlaps with the same GPU clusters and orchestration layers powering AI agent deployments. Firms building at that junction serve two demand curves simultaneously.
Where Dimension Fits
The $800 million fund positions Dimension in the mid-tier of infrastructure-focused capital raises this month, well above the seed and Series A rounds that dominate agent-native startups, but below the multi-billion-dollar vehicles from established firms like Khosla.
New York City startups raised a record $8.88 billion across 233 deals in Q2 2026, with AI companies capturing 51% of the capital. Dimension’s New York base and science-computing focus align with that concentration.
The Compute Thesis
The broader pattern: investors are increasingly betting on the platforms and infrastructure agents run on rather than the agents themselves. Compute orchestration, GPU allocation, data center capacity, governance tooling, and domain-specific hardware are attracting capital at rates that outpace the agent application layer. The bet is that whoever controls the infrastructure stack captures value regardless of which agents or models win at the application level.