Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN
Michael Intrator, CEO of CoreWeave, shares how they built the AI infrastructure company from crypto mining origins to a $35B hyperscaler. Key insight: Computing "decommoditizes at scale" - anyone can run a GPU, but building supercomputers that can train world-changing models requires specialized inf
1h 37mKey Takeaway
Michael Intrator, CEO of CoreWeave, shares how they built the AI infrastructure company from crypto mining origins to a $35B hyperscaler. Key insight: Computing "decommoditizes at scale" - anyone can run a GPU, but building supercomputers that can train world-changing models requires specialized infrastructure. Their innovation: structuring 5-year customer contracts (Microsoft, OpenAI, etc.) into isolated "boxes" that enable low-cost financing, driving their cost of capital down 600 basis points in 2 years while proving GPUs remain valuable far beyond the 16-month depreciation myths.
Episode Overview
This episode features four AI CEO interviews from Nvidia's GTC conference, focusing primarily on Michael Intrator of CoreWeave and Arvin Srinivas of Perplexity. CoreWeave's journey from crypto mining to AI infrastructure reveals how they became a major hyperscaler through innovative financing and long-term customer relationships. The discussion covers GPU economics, infrastructure constraints, and the evolution of AI services. Perplexity's segment highlights their multi-model approach and innovative features like live data integration and AI-powered browsing.
Key Insights
Computing Decommoditizes at Scale
While anyone can run a single GPU, the real competitive advantage lies in building clusters large enough to train transformative models. CoreWeave focuses exclusively on AI infrastructure rather than general cloud services, allowing them to optimize every layer between Nvidia GPUs and foundation models for this specific use case.
The "Box" Financing Innovation
CoreWeave structures each major customer contract into an isolated financial vehicle ("the box") containing the customer contract, GPU purchase orders, and data center agreements. Revenue flows into the box, pays expenses and debt first, then returns profit to CoreWeave. This structure enables 5-year financing with confidence that principal is repaid within 2.5 years, driving their cost of capital down 600 basis points.
GPU Depreciation Myths vs. Reality
Despite market claims that GPUs become obsolete in 16-18 months, CoreWeave's 5-year average contract length proves otherwise. As new architectures launch, previous-generation GPUs move from training to inference to new use cases. A100 prices actually appreciated through the year as new companies and applications emerged needing that capacity.
Risk Management Through Diversification
Coming from hedge fund backgrounds, CoreWeave's founders applied rigorous risk management from day one. They diversified from crypto mining into CGI rendering, batch computing, medical research, and eventually neural networks - always viewing GPU compute as having optionality across multiple use cases rather than betting on a single application.
The Path from Crypto to AI Leadership
CoreWeave donated A100s to open-source AI researchers (Eleuther AI) as "tuition" to learn how to run large-scale parallel computing for neural networks. Those researchers later became customers at their day jobs, launching CoreWeave's commercial AI business. This created crucial information flow across different AI use cases and market segments.
Notable Quotes
"Computing decommoditizes at scale right like when you know anybody can run a GPU but can you run a cluster that's large enough to train a model that can change the world and that's a different question."
"Our average contract is 5 years. So any commentary by anyone either inside or outside of the industry that this stuff becomes obsolete in 16 months or whatever nonsense they're spewing, it doesn't it doesn't in any way match up with the facts on the ground."
"If people are willing to pay me for it, it still has value."
"We have dropped our cost of capital by 600 basis points."
"The cost of a million tokens when ChatG3 came out and it was $32 and change and now a million tokens cost nine cents."
Action Items
-
1
Apply Risk Management Principles to New Ventures
Like CoreWeave's hedge fund-trained founders, approach new technology investments with disciplined risk management. Diversify across use cases and maintain optionality rather than betting everything on a single application. Structure financial commitments to ensure sustainability through market volatility.
-
2
Invest in Learning New Technologies Through Practice
When entering unfamiliar territory, consider 'paying tuition' through initial investments that allow hands-on learning with lower stakes. CoreWeave donated GPUs to open-source projects to learn neural network infrastructure without high-pressure commercial SLAs, which later converted to paying customers.
-
3
Structure Long-Term Contracts for Predictability
In capital-intensive businesses, secure multi-year commitments from creditworthy customers before making major infrastructure investments. This enables better financing terms and protects against short-term market fluctuations. CoreWeave's 5-year contracts provide the foundation for their entire business model.
-
4
Focus on Information Flow Across Market Segments
Build relationships across different use cases and customer types to gain early signals about emerging trends and needs. CoreWeave's diversified customer base (from LLMs to high-frequency trading to search) provides valuable intelligence about where to invest next and what products to build.