20VC: Nebius Co-Founder on AI Infrastructure Bubbles | The Real Impact of Open Source on OpenAI & Anthropic | How Price Elastic is Demand for Compute | Could Nebius Sell 10x More Compute If They Had It & more with Roman Chernin
Roman Chernin, co-founder of Nebius, discusses the company's AI infrastructure strategy, arguing we are at the very beginning of AI adoption rather than in a bubble. He outlines Nebius's four-layer product stack from bare metal capacity to managed inference, and explains why diversification of customers and vertical integration are critical to long-term survival against hyperscaler competition.
Summary
Roman Chernin, co-founder of Nebius (a $6.6B market cap AI infrastructure company), joins 20VC to discuss the state of AI infrastructure, Nebius's competitive positioning, and his vision for the future of compute. He opens by firmly rejecting the 'bubble' narrative, arguing that enterprise AI adoption is still in its first few percentage points across use cases, with coding being the only truly scaled use case that has emerged in just the last few months.
On the question of open source models threatening frontier providers like OpenAI and Anthropic, Chernin argues this is already happening but is not destructive — rather, it's a natural maturation pattern where customers crack use cases on frontier models, then shift to tunable open source models for economics and specialization. He uses the Jevons Paradox to explain that cheaper inference doesn't reduce demand but expands it, citing the DeepSeek moment when Nebius stock fell 40% yet the company had its best-ever sales week simultaneously.
Chernin describes Nebius's four-layer product strategy: (1) bare metal capacity sold in megawatts to hyperscalers and large labs; (2) multi-tenant managed cloud sold in GPU-hours to research-heavy teams; (3) managed inference (Token Factory) sold in tokens to vertical AI companies and enterprises; and (4) an emerging agentic orchestration layer where developers think in terms of task outcomes rather than individual model calls. He argues this full-stack vertical integration — both downward into physical infrastructure and upward into product — differentiates Nebius from pure-play infrastructure competitors.
On capital and competition, Chernin acknowledges Nebius's $20-25B CapEx program this year is roughly 8x smaller than hyperscaler competitors. He explains that capital helps on 24-month timelines but is irrelevant on 6-month horizons, and that the real bottleneck shifts depending on time horizon — execution in the near term, capacity constraints in the medium term, and capital in the long term. He expresses concern about community pushback on data center permitting and acknowledges Nebius must do better at community engagement, especially as ~70-75% of new capacity is now being built in the US.
Chernin identifies consolidation — not competition — as the biggest threat to Nebius, arguing the company thrives in a diversified, democratized ecosystem. He warns that a world controlled by three to five super-companies would reduce Nebius to a physical infrastructure provider with limited value creation. He also discusses the importance of NVIDIA relationships, arguing that engineer-to-engineer respect is the most reliable foundation for that partnership. The interview closes with reflections on his daughters' futures, where he now prioritizes empathy and creativity over math and engineering skills.
About this episode
<p dir="ltr">Roman Chernin is Co-Founder and Chief Business Officer of <a href="https://nebius.com?utm_source=chatgpt.com">Nebius</a>, one of the fastest-growing AI infrastructure companies in the world. Today, Nebius operates some of the largest AI compute clusters globally and serves leading AI labs, enterprises, and developers. Today, Nebius has a market cap of $57BN. </p> <p dir="ltr">AGENDA: </p> <p dir="ltr">00:00 — Why AI Infrastructure Is Not a Bubble</p> <p dir="ltr">05:00 — The Real Impact of Open Source on OpenAI & Anthropic</p> <p dir="ltr">11:00 — Jevons Paradox: Why Cheaper AI Creates More Demand</p> <p dir="ltr">13:00 — The Four Layers of AI Infrastructure Explained</p> <p dir="ltr">19:00 — If Nebius Had 10x More Capacity Tomorrow</p> <p dir="ltr">26:00 — The Shift from Training to Inference and Agents</p> <p dir="ltr">31:00 — How Token Factory Cuts AI Costs by 70%</p> <p dir="ltr">44:00 — Sovereign AI, Europe, and the Future of Model Building</p> <p dir="ltr">49:00 — Competing Against Hyperscalers with 10x More Capital</p> <p dir="ltr">59:00 — The Biggest Threat to Nebius Isn't Competition—It's Consolidation</p> <p dir="ltr"> </p> <p dir="ltr"> </p>
Key Insights
- Chernin argues we are not in an AI infrastructure bubble, pointing out that coding only became a reliably scaled AI use case a few months ago, suggesting most enterprise adoption is still in its first percentage points.
- Chernin claims the DeepSeek moment perfectly illustrated Jevons Paradox in AI: Nebius stock fell ~40% that week on bubble fears, yet the company simultaneously had its best-ever commercial sales week as cheaper inference unlocked new production workloads.
- Chernin describes a consistent customer maturation pattern where companies start on frontier closed models (OpenAI, Anthropic), crack a use case, then shift to tunable open source models for economics and specialization — which he argues is already the present, not the future.
- Chernin argues that Nebius's primary differentiator versus other neo-clouds is 'full-stack integration' — deep vertical ownership both downward into physical infrastructure (data centers, racks, servers) and upward into managed product layers — rather than just GPU capacity.
- Chernin states Nebius's four product layers serve fundamentally different customer populations: bare metal serves a 'dozen customers in the world,' managed cloud serves hundreds, managed inference serves thousands, and the agentic layer will serve tens of thousands of developers.
- Chernin claims that optimizing inference through techniques like model distillation, speculative decoding, and caching can reduce token costs by up to 70%, meaning the nominal GPU price is far less important than the total cost of ownership delivered by a managed platform.
- Chernin identifies consolidation — not competition from hyperscalers or other neo-clouds — as the single greatest existential threat to Nebius, arguing the company's business model depends on a diverse, fragmented ecosystem of AI builders.
- Chernin observes that enterprise customers like Revolut face a 'cold start problem' where early AI adoption is slow due to foundational investments in evaluation infrastructure and CI/CD pipelines for AI — but once solved, they grow on the same exponential trajectory as AI-native startups.
- Chernin argues that Nebius's $20-25B CapEx this year is approximately 8x smaller than hyperscaler competitors, and that additional capital primarily accelerates the 18-24 month build horizon, not the 6-month execution window where current constraints dominate.
- Chernin contends that on the bare metal layer, Nvidia GPU infrastructure is not truly a commodity at scale because satisfying the most demanding customers like Meta or Microsoft requires highly sophisticated physical infrastructure that is extremely difficult to execute reliably.
- Chernin argues that engineer-to-engineer respect is the most reliable foundation for a productive NVIDIA relationship, claiming that if NVIDIA's engineers respect your engineers, the commercial partnership follows — rather than the relationship being primarily driven by purchasing power.
- Chernin states he has shifted his advice to his teenage daughters away from math and engineering skills toward empathy-based human communication and creativity, arguing these will be the scarcest and most valuable capabilities as AI commoditizes hard technical skills.
Topics
Transcript
Who has power against Nvidia? We are in the capital intensive game and we're competing with the most capitalized companies in the world. Our CapEx program this year is 20, $25 billion. Our competitors, hyperscalers have eight times bigger. In the next six months, the capital cannot help. Six months is too short time. You have what you have. You need to deliver. Consolidation. Yeah. I think that the main threat for Nebios as a business is the world will be too much consolidated. It's like a shark. You're alive when you move, right? So we have to move. The AI infrastructure race is on. CapEx spend has never been greater. At the center of this, Nebius. Today, I'm joined…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
20VC: Is Seed Investing Dead Without a $1BN Fund? | Does Ownership and Price Matter When Companies Can Be $1TRN Exits | Are AI Revenue Numbers Real and What to Watch Out For with Venky Ganesan, Menlo Ventures
Venky Ganesan from Menlo Ventures discusses how venture capital has fundamentally changed in the AI era, arguing that seed investing still exists but founders must understand that venture is now focused on outlier companies with explosive growth rather than steady revenue growth. He emphasizes the importance of IRR over ownership percentages, the dangers of gaming metrics, and the need for VCs to play the current market game while maintaining authentic values.
20VC: The Future of Datacentres: What You Need to Know | Why Everyone Gets GPU Depreciation and AI's Energy Costs Wrong | Who Really Makes Money From AI & Why Most Moats Don't Exist with Chase Lochmiller
Chase Lochmiller, CEO of Crusoe Energy, discusses how data centers and AI infrastructure are becoming critical bottlenecks, arguing that energy-distributed computing rather than centralized hubs will power AI's future, and that vertical integration across data centers, GPUs, and inference services creates sustainable competitive advantages in the AI infrastructure space.
20VC: $1BN ARR in 18 Months; The Untold Story of Higgsfield | Spending $4M Per Month on Models | Why Moats in AI are BS | Scaling a Content Team to 150 People with Alex Mashrabov
Alex Mashrabov, founder of Higgsfield, discusses how his AI video generation startup achieved $1 billion ARR in 18 months—faster than Coursera—by focusing on product-market fit, strategic model routing between open-source and proprietary AI models, and building a 400-person team primarily based in Kazakhstan. He shares insights on AI moats being largely meaningless, the $4 million monthly spend on model inference, and his vision for Higgsfield to become bigger than Shopify by owning the distribution layer for AI-generated content.
20VC: Jensen Huang Declares AGI Has Arrived | GPT Astra and Fable 5.1 Accelerate the Model Race | Tesla Launches Cybercabs | Index Pulls Out of Town & Anthropic Pulls From Descartes Acquisition
In this 20VC episode, Rory O'Driscoll and Jason Lemkin discuss major tech news including Jensen Huang's AGI claims, AI agent risks and rule-breaking behavior, Elon's Cybercab launch, venture capital conflicts around AI assistants, and high-growth companies like Wonderful raising at massive valuations with significant secondaries.
20VC: The $100 Billion AI Assistant Race: Town vs Instinct vs GrokBot | We Spend $75K Per Engineer on AI Tools | Why the AI Assistant Market Is Not a Bubble & AI Assistants Will Replace Every App on Your Phone with JD, Founder of Town
JD, founder of Town, discusses how his AI assistant platform operates within email and calendar to automate work tasks, argues that network effects at the agent level will be the true moat in the AI assistant market, and explains why the speed of market development means companies can only learn at human speed while building at machine speed.