The All-In Podcast
Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN
Episode Summary
AI-generated · Mar 2026AI-generated summary — may contain inaccuracies. Not a substitute for the full episode or professional advice.
The episode features an interview with Michael Intrater, CEO of CoreWeave, conducted at Nvidia's annual GTC conference. Intrater detailed CoreWeave's unique journey, originating from an algorithmic hedge fund in 2017 that pivoted from crypto mining (specifically Ethereum, not Bitcoin) to building out GPU compute infrastructure. Initially, their focus expanded from crypto to CGI rendering, then to batch computing for medical research. A significant turning point was donating A100 GPUs to the open-source Luther AI project, which served as "tuition" for learning to run large-scale parallelized computing for neural networks.
CoreWeave then shifted to scaling for larger, more sophisticated clients, positioning itself as an infrastructure provider "above the Nvidia GPUs but below the models." They became a key partner for major AI entities, including Inflection as their first large commercial language model client, and later OpenAI, providing the supercomputers necessary for training and, increasingly, inference. Intrater highlighted inference as the monetization phase of AI investment, noting a massive increase in inference compute driven through their infrastructure as models move from research to productization.
Intrater also addressed the debate around GPU depreciation, firmly rejecting the notion that hardware becomes obsolete in 16-18 months. He stated CoreWeave's average client contract is 5 years, and they use a 6-year depreciation schedule, arguing that older GPUs like the A100 continue to appreciate in value and find new use cases, comparing it to the sustained demand for older iPhone models in developing markets. He described CoreWeave's innovative "box" financing mechanism, which allows them to raise capital by collateralizing client contracts, GPU purchases, and data center agreements, enabling them to secure $35 billion in 18 months and reduce their cost of capital by 600 basis points. The demand for GPU compute, according to Intrater, remains "relentless" and outstrips global supply, reinforcing their strategy of long-term contracts with financially robust partners.
👤 Who Should Listen
- Tech Professionals
- Early Adopters
- Software Engineers
🔑 Key Takeaways
- 1.CoreWeave originated from an algorithmic hedge fund in 2017, leveraging their risk management skills to navigate early crypto winters before pivoting to diverse GPU compute applications.
- 2.The company's evolution from crypto mining to CGI rendering, then medical research, and eventually neural networks demonstrates a strategic progression up the complexity stack for GPU utilization.
- 3.CoreWeave plays a crucial role in bringing Nvidia's bleeding-edge GPU architectures, such as H100s and GB200s, to commercial production at scale, serving major AI companies.
- 4.Michael Intrater dismisses the debate about rapid GPU obsolescence, explaining that CoreWeave's clients sign 5-year contracts, they use a 6-year depreciation schedule, and older GPUs find new, valuable use cases.
- 5.CoreWeave pioneered an innovative "box" financing model, creating separate entities for each client contract, GPU purchase, and data center agreement, enabling them to raise $35 billion in 18 months and significantly reduce their cost of capital.
- 6.The demand for GPU compute capacity is described as "relentless" and overwhelming global supply, necessitating CoreWeave's focus on long-term contracts with large, stable counterparties to manage risk.
- 7.Nvidia fulfills GPU orders chronologically rather than playing favorites, despite intense demand from major companies and sovereign entities.
💬 Notable Quotes
“We kind of live above the Nvidia GPUs but below the models.”
“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's it doesn't it doesn't in any way match up with the facts on the ground. The facts on the ground is they're buying it for 5 years.”
“The depth of the demand for the service we provide has been relentless and overwhelms the global capacity of the world to deliver enough compute to enable all of the demand for artificial intelligence to be sated.”
More from this guest
Michael Intrater
Listen to Full Episode
📬 Get weekly summaries like this one
No spam. Unsubscribe anytime. By subscribing you agree to our Privacy Policy.
Continue Exploring
