"Turpentine VC" | Venture Capital and Investing · Erik Torenberg

E62: Wing's Chris Zeoli on AI Infrastructure Opportunities

·49 min·3 clips

Meta is spending $20 billion this year, and OpenAI is spending $7 billion on inference and training.

1. Turpentine VC E62, "Wing's Chris Zeoli on AI Infrastructure Opportunities," focuses on Wing Capital's view of enterprise AI infrastructure. 2. Chris Zioli, a Wing Capital partner, joins Eric Torenberg to explain why Wing backs data-heavy enterprise companies. 3. The episode asks how data, models, and infrastructure fit together in AI investing. 4. Chris says Wing avoids consumer, international, and fintech and concentrates on how data is used in the modern enterprise. 5. He names Snowflake and Cohesity as early Wing-backed foundations for enterprise data infrastructure. 6. Chris describes a flow from emails and documents into embeddings, RAG pipelines, vector databases, and production applications. 7. He cites Voyage AI as an investment focused on better embeddings models. 8. He points to Pinecone as the leading vector database company in Wing's portfolio context. 9. Chris says internal AI applications are easier than customer-facing bots because governance and security risks are lower internally. 10. He adds that small-model production work, including Gradient AI, matters where latency or domain expertise are important. 11. Chris uses Jensen's framework to contrast the internet's retrieval architecture with a new era of retrieval plus generation. 12. He says simple chat tasks like "who the first president of the US is" need little generation, while reasoning-heavy tasks do. 13. He argues that large language model training is capital intensive and that open and closed players are fighting for scale. 14. He says model performance is showing diminishing marginal returns, with many systems clustering around a 90% accuracy benchmark. 15. Chris frames OpenAI, Anthropic, Meta, Google, and Amazon as competitors in both consumer and enterprise AI. 16. He says Meta can spend $20 billion this year and OpenAI can spend $7 billion on inference and training. 17. The conversation turns to infrastructure detail, including GPU clouds, data-center upgrade cycles, networking, memory, cooling, and power. 18. Chris also says cybersecurity is shifting because exposing more enterprise data to models creates new governance risk for every data worker. 19. This interview is technical, investment-focused, and heavy on concrete company examples, spending figures, and stack-level tradeoffs. 20. Listeners interested in enterprise AI, semiconductor supply chains, and infrastructure investing will get the most value; general consumer-AI chatter listeners may skip it.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from "Turpentine VC" | Venture Capital and Investing