Jason
Cui

The check was $150 million. The company had no product shipping at scale. The founders were the team that had built vLLM - the open-source inference engine that most of the AI industry runs on. Jason Cui signed off on it anyway. That's not risk tolerance. That's pattern recognition at the layer where it counts.

Cui joined Andreessen Horowitz as a Partner on the Infrastructure & AI team after a career that ran from Harvard's computer science program through product roles at Uber and Hulu, into the serverless compute org at Databricks, and then briefly through the founder's chair at Jemi - a creator commerce platform he co-built with Annie Hwang and sold to Brat TV in 2023. The pivot from operator to investor wasn't a retreat. It was a vantage point upgrade.

At a16z, his mandate sits at the intersection of foundational AI, developer tooling, and scientific platforms. Not AI apps. Not AI wrappers. The plumbing. The inference engines. The data layers. The orchestration infrastructure that every LLM-powered product quietly relies on and rarely gets credit for building.

He graduated Harvard with a computer science degree in 2018 and moved through the operator track with deliberate speed. Product management at Hulu. Product at Uber. Then a detour that told you something about how he thinks: he didn't go straight to VC. He went to go build something.

Jemi was a bet that creators needed better infrastructure to monetize their audiences - before "creator economy infrastructure" became a pitch deck category. He and Annie Hwang raised from Y Combinator, General Catalyst, and Kleiner Perkins. Forbes put him in the 30 Under 30 for Consumer Technology in 2021. Brat TV acquired the company in 2023. The outcome was a strong signal that he understood what building toward an acquisition actually looks like from the inside.

That operator scar tissue - the kind you get from shipping products, managing developer experience at Databricks scale, and selling a company - is rare in venture capital. Most infrastructure investors have either built deep technical systems or managed product roadmaps. Cui has done both, which explains why founders at the infrastructure layer respond to him differently than they do to analysts with strong conviction and no calluses.

One of Cui's more articulate public positions is on AI and scientific discovery - a space where most investors see either pharma adjacency or long-horizon risk. He doesn't. He sees a structural shift where AI accelerates the pace of discovery, and better discovery in turn produces better training data and stronger AI systems. A feedback loop. An ecosystem play, not a moonshot bet.

He works with the SAIR Foundation - an organization building community around AI for Science - and has written and spoken extensively on how enterprise AI agents fail without robust context layers. His March 2026 piece with Jennifer Li, "Your Data Agents Need Context," identified a specific failure mode in enterprise AI deployment: agents that have access to data but no semantic understanding of what that data means. The fix isn't more data. It's ontology. It's context architecture. It's infrastructure, again.

In January 2026, a16z led a $150 million seed round into Inferact - one of the largest seed rounds in AI infrastructure history. The company was founded by the maintainers of vLLM, the open-source project that has become the de facto inference engine for deploying large language models at scale. Cui co-led the deal alongside Matt Bornstein and Raghu Raghuram.

The bet makes sense in the context of Cui's thesis. If AI infrastructure is the critical layer, then inference is the critical chokepoint - the place where model capability meets production reality. Whoever builds the most efficient, most scalable, most developer-friendly commercial inference engine captures outsized value as the broader AI market matures. Inferact's founders weren't just technically credible. They had already written the code that most of the industry runs on.

In "Search Wars: Episode 2," co-authored with Jennifer Li, Sarah Wang, and Steph Zhang, Cui's team mapped the transformation of search from an ad-optimized link retrieval system to an AI-native index. The argument wasn't that Google was dying. It was that the surface area of "search" was expanding to include interfaces that never existed before - AI chat, vertical agents, voice, code completion - and that each new surface was a new infrastructure problem.

That's the thesis, restated. Infrastructure. Again. From a different angle.

He spoke at the Databricks Data + AI Summit in 2025, drawing on his product experience to talk about where AI infrastructure is actually constrained. Not by model capability. By data context, by deployment architecture, by the unsexy operational layer that determines whether an AI system is useful or merely impressive in a demo.

Cui is part of Gold House, the network championing Asian Pacific representation in business and creative culture. He's been a Venture Scout for Gold House Ventures, and the community element of his work - connecting people across the AI and tech ecosystem - seems genuinely important to how he operates rather than being a checkbox activity.

The Harvard in Tech spotlight that featured him and Hwang after founding Jemi described a duo who were building something real, not just executing on a pitch deck. The framing of Jemi as "creator infrastructure" rather than "creator app" said something about how Cui thought about markets even before he formally became an investor.

He lives in San Francisco, competes in jiu jitsu, makes music, and works at a firm that has backed some of the defining companies of the last decade. The combination of founder experience, operator credibility, and genuine intellectual curiosity about AI's structural impact on science and enterprise software makes him a different kind of infrastructure partner - one who has been on the other side of the table and still chose to come back to the investing seat.

a16z's Jason Cui on AI for Science, Infrastructure & the Next Wave of Breakthroughs - a wide-ranging conversation on where he thinks the infrastructure layer of AI is heading and why scientific discovery is the underappreciated frontier.