MoE Capital Research
Overview
MoE Capital (also operating as "MoE Labs") is a newly launched early-stage venture fund investing at the intersection of agentic infrastructure and AI for Science. The name derives from the Mixture of Experts (MoE) architecture in machine learning, signaling their thesis that the best VC outcomes come from a structured collective of specialists, not generalists. The fund website went live on June 16, 2026, making this one of the most recently launched AI-focused funds in the market.
Investment Thesis
MoE Capital's thesis is anchored in "backing the Agentic AI Era with Expert-Led Conviction." The fund believes the current decade is the defining era for agentic AI: systems where autonomous agents drive the core product experience and transform how humans and software collaborate.
They invest exclusively in two areas:
1. Agentic Infrastructure — The picks-and-shovels layer for building and running agents:
- Training orchestration and reinforcement pipelines
- Evaluation and safety tooling
- RAG and memory systems
- Cost-efficient inference infrastructure
2. AI-Native Applications — Products where autonomous agents are central to the experience, not bolted on. This includes AI for Science, robotics, and AI-native enterprise tools.
The fund explicitly focuses on companies with "high frontier proximity" — founders deeply embedded in the technical frontier, ideally with research backgrounds or operator experience at top AI labs.
Expert Network and Differentiation
MoE Capital's primary differentiation is its "Mixture of Experts" network: a structured collective of frontier researchers and product leaders who provide portfolio companies with:
- Technical deep dives with researchers from OpenAI, Anthropic, xAI, Google DeepMind, Meta, and Recursive Superintelligence
- Faculty advisors from Stanford and Princeton
- Hiring support and talent introductions
- Design partnerships
- Founder salons and community access
- Customer connections and partnership introductions
The fund's belief is that the highest-leverage AI investing happens before categories exist, and their edge comes from being deeply embedded in the communities where those categories emerge.
Stage and Check Size
MoE Capital is an early-stage fund. Stage preferences and check sizes have not been publicly disclosed as of the fund launch date. Given the team composition (former AGI House operators, Princeton research faculty) and the early-stage positioning, the fund likely writes checks at pre-seed and seed stages.
Team
Henry Yin — General Partner
Henry is a researcher, founder, and force multiplier for AI builders. He is the co-founder of AGI House, one of the most prominent AI community hubs in Silicon Valley. Henry also created Apache DevLake, a widely used open-source engineering intelligence platform, and co-founded Merico as CTO. He holds a BS from Tsinghua University and dropped out of a PhD program at UC Berkeley. His deep community roots in the AI builder ecosystem — through AGI House, which has hosted hundreds of hackathons and events with top AI researchers and founders — give him unparalleled deal flow access to pre-announcement AI companies.
Naomi (Yue) Xia — General Partner
Naomi is a builder-operator with a strong investment track record in early-stage AI. She was most recently Investment Partner at AGI House Ventures, the venture arm of AGI House. Before that, she worked at J.P. Morgan, Inception Capital, and Kuaishou Technology. She holds an MBA in Business Analytics from The Wharton School. Naomi has been actively organizing AI community events and paper reading clubs, focusing on AI in fintech and enterprise applications. She is based in San Francisco, CA.
Mengdi Wang — General Partner
Mengdi is an Associate Professor at Princeton University's Department of Electrical and Computer Engineering and Center for Statistics and Machine Learning (CSML). Her research focuses on data-driven stochastic optimization, machine learning theory, reinforcement learning, generative AI, and AI for Science. She received her PhD in EECS from MIT in 2013 under Dimitri Bertsekas at the Laboratory for Information and Decision Systems (LIDS). She is a recipient of the Young Researcher Prize from the Mathematical Optimization Society (2016), Princeton SEAS Innovation Award (2016), NSF CAREER Award (2017), Google Faculty Award (2017), and MIT Technology Review 35-Under-35 Innovation Award (China region, 2018). Mengdi brings research credibility for the fund's AI for Science thesis and access to Princeton's academic network.
Portfolio
As of the fund's launch date (June 16, 2026), MoE Capital has assembled an initial portfolio of 8 companies. All portfolio companies are in stealth mode — only descriptive taglines appear on the portfolio page without company names. The portfolio spans the fund's investment themes:
- A company building the foundation of multimodal reasoning
- A platform for agent skills that evolve over time
- An adaptive intelligence solution for modern work
- A company providing human-AI interaction data for agent training
- AI agents for biomedical research
- A video-native music generation company
- A compounding intelligence platform for the enterprise
- A self-improving superintelligence system for knowledge discovery
The stealth status of all portfolio companies reflects the fund's stated belief that they invest "before categories exist" at the earliest possible stages.
Newsletter and Content Strategy
MoE Capital publishes a newsletter called "MoE Dispatch" on Beehiiv, focused on signals from the agentic frontier. The inaugural dispatch was authored by Naomi Xia in May 2026. The fund also maintains a research blog with essays on key frontier AI topics including world models, continual learning, and reinforcement learning, reflecting the research-led culture of the fund.
Geographic Focus
The fund is based in San Francisco and has strong roots in the SF Bay Area AI community through AGI House. Global research connections through Princeton (Mengdi Wang) and a community spanning London-based companies (Recursive Superintelligence is in their expert network) suggest openness to exceptional founders globally, particularly those with ties to top research institutions.
Decision Process
Given the three-GP partnership structure and the research-driven culture, investment decisions likely involve the full partnership. The fund's emphasis on "expert-led conviction" suggests deals are vetted through the broader expert network before commitment.
Competitive Positioning
MoE Capital sits at an intersection few funds occupy: deep research credibility (Princeton professor GP), operator network (AGI House community), and product instinct (ex-AGI House Ventures investment partner). They compete for deals with AI-specialist funds like Conviction Capital, Plural, and AI Grant, but differentiate through the proprietary MoE expert network that provides hands-on technical support unavailable from most VCs.