Search for a command to run...

Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
This episode dives deep into CoreWeave, a crypto-mining-turned-AI-infrastructure company that sits at the heart of the current AI boom and potentially its bubble. (01:15) Host Eli Patel interviews Verge senior reporter Liz Lopatto about her extensive investigation into CoreWeave's origins, financials, and its unique relationship with NVIDIA. The conversation reveals how CoreWeave went from mining Ethereum in New Jersey to becoming a $50 billion public company that leases NVIDIA GPUs to major AI companies like Microsoft, Meta, and OpenAI. (02:20) However, the story uncovers complex financial maneuvering, circular funding arrangements, and a troubling dependency on NVIDIA that raises serious questions about the sustainability of the AI infrastructure buildout.
Editor-in-Chief of The Verge and host of the Decoder podcast. Patel leads The Verge's coverage of technology, policy, and their intersection with culture and business.
Senior reporter at The Verge specializing in AI, finance, and technology business models. She has extensive experience covering complex financial arrangements and tech industry trends, and is currently working on a major investigative piece about CoreWeave.
Many investors believe investing in AI infrastructure companies like CoreWeave is safer than betting on AI companies directly. (04:02) However, Lopatto's investigation reveals that CoreWeave has minimal competitive moats compared to traditional cloud providers like Amazon Web Services. While AWS offers comprehensive software services and infrastructure management, CoreWeave essentially just leases out NVIDIA chips with basic connectivity. (06:10) This means CoreWeave competes primarily on price and energy access, making it vulnerable to both larger tech companies building their own data centers and potential oversupply in the market.
The AI industry is riddled with circular financing arrangements where companies invest in each other and then pay each other for services using the same money. (28:39) Microsoft invests billions in OpenAI, which then pays Microsoft for computing services. Similarly, CoreWeave gives OpenAI $350 million in company stock while OpenAI has a $22 billion contract with CoreWeave. (29:52) While this creates the appearance of robust demand and revenue, it's essentially "passing around the same $5" and may not represent genuine market validation of AI services.
NVIDIA has positioned itself as both CoreWeave's primary supplier and financial backer, creating an unprecedented arrangement where the chip supplier guarantees demand for its customer's services. (11:23) When CoreWeave's IPO faltered, NVIDIA bought shares and promised to purchase any excess compute capacity that isn't sold to customers. This arrangement gives NVIDIA significant leverage over major tech companies like Microsoft and Amazon, as it can influence data center capacity and maintain its chips as the default standard for AI workloads. (35:15)
CoreWeave pioneered a new financing model where NVIDIA GPUs serve as collateral for massive loans, but this creates a critical timing problem. (19:20) The company has taken on enormous debt across multiple instruments - chip loans, revolving credit lines, vendor financing, and secured loans - all betting that AI will generate sufficient revenue before the chips depreciate or wear out. (22:53) If AI adoption follows a normal technology curve (like e-commerce taking 25 years to reach 25% of retail), rather than the explosive growth assumed in these debt structures, CoreWeave faces serious financial difficulties.
Major tech companies are using special purpose vehicles (SPVs) and companies like CoreWeave to shift the risks of AI data center construction away from their core businesses. (15:28) This allows companies like Microsoft to maintain pristine credit ratings while still participating in the AI buildout. (17:42) If the AI boom fails to materialize, companies like Meta, Google, and Amazon can absorb losses through their other profitable business lines, while specialized AI infrastructure companies like CoreWeave have no fallback options. (33:18)