Descriptions:
Harry Stebbings sits down with Roman Chernin, co-founder of Nebius, the AI infrastructure company spun out of Yandex that has grown to a reported $6.6 billion market capitalization, for a frank discussion of where the AI compute market actually stands. Chernin pushes back firmly on bubble narratives, arguing that enterprise AI adoption is still in the first one percent of potential use cases — with agentic coding having only emerged as a reliably proven application within the past several months.
The conversation covers Nebius’s full-stack differentiation strategy: building down to physical data centers, custom racks, and server infrastructure while simultaneously building up into inference platforms, model fine-tuning pipelines, and data flywheel services. Chernin explains the pattern he observes repeatedly — companies start with frontier API providers like OpenAI or Anthropic for speed of iteration, then migrate toward specialized or fine-tuned open-source models once they find product-market fit and need to optimize unit economics. He positions Nebius as the platform designed to capture that transition.
Stebbings also probes the six-month capital deployment constraint Chernin describes — ‘in the next six months, the capital cannot help, you have what you have’ — the consolidation risk if hyperscalers crowd out independent infrastructure players, and how Nebius differentiates from competitors like CoreWeave that share similar GPU inventory and major customers including Meta. The result is an unusually candid look at the business realities of competing in AI infrastructure.
📺 Source: 20VC with Harry Stebbings · Published June 08, 2026
🏷️ Format: Interview
