Key Takeaways
- The AI industry is experiencing significant investment, but concerns persist about a potential 'bubble.'
- CoreWeave, a former crypto miner, pivoted to provide NVIDIA GPU infrastructure for AI, securing multi-billion dollar deals.
- CoreWeave's business model relies heavily on debt financing, with NVIDIA as both supplier and investor, guaranteeing demand.
- Investor sentiment is cooling due to concerns about high debt, chip depreciation, circular financing, and potential overcapacity.
- A slowdown in demand for AI inference is considered a critical factor for how any AI bubble might burst.
Deep Dive
- CoreWeave, founded by former commodities traders as an Ethereum mining firm, pivoted to AI after crypto's 2022 decline.
- The company now builds data centers and leases NVIDIA GPUs to AI firms like Meta and OpenAI.
- It has secured multi-billion dollar deals, including a $14 billion deal with Meta and a $6.5 billion deal with OpenAI.
- CoreWeave describes its financing model as 'creative financing' to support its data center build-outs.
- NVIDIA is both CoreWeave's main supplier of GPUs and an investor, holding equity in the company.
- NVIDIA had initial reservations about supplying chips for crypto mining but shifted its focus to AI as a strategic market after ChatGPT's release.
- NVIDIA guarantees CoreWeave's income by purchasing excess compute capacity, effectively backstopping demand for its own chips.
- CoreWeave operates over 250,000 NVIDIA chips, with its business model relying on buying these GPUs and selling their usage at a higher price.
- CoreWeave takes on significant debt, secured by NVIDIA chips, to fund rapid data center build-outs.
- This model allows companies like Microsoft to rent compute power without taking on the risks of data center construction or rapid chip depreciation.
- The use of Special Purpose Vehicles (SPVs) by companies like Meta is a related strategy to limit risk and maintain credit ratings.
- Concerns exist about whether CoreWeave can sustain profitability by solely accepting AI industry risk.
- CoreWeave pioneered using debt for chip acquisition, a strategy now followed by other 'neo-cloud' companies like Lambda and Crusoe.
- A key concern is whether AI workloads can generate sufficient profit before chips depreciate, creating a 'chronology problem.'
- The rate of chip depreciation is disputed, raising questions about their reliability as collateral for substantial loans.
- Slower AI adoption by businesses, similar to online shopping's 25-year ramp-up to 25% of retail, suggests a more gradual integration than currently assumed.
- CoreWeave's stock dropped after reporting third-quarter earnings, attributed to negative investor sentiment, a lowered revenue forecast, and a customer contract delay.
- Investor anxiety across the AI sector is increasing, fueled by reports of circular deals and debt concerns.
- Circular financing involves a company (e.g., OpenAI) receiving funds from an investor (e.g., Microsoft) and then using those funds to pay the same investor for services.
- Building infrastructure is inherently challenging, with factors like construction schedules and weather causing delays.
- Sam Altman of OpenAI has commented on the possibility of an AI bubble.
- A slowdown in demand for AI inference is identified as a key indicator for how the AI bubble might pop.
- CoreWeave lacks diversified businesses to absorb potential data center overcapacity, unlike companies like Amazon or Meta, making it more vulnerable if AI demand is unmet.
- NVIDIA's significant investment in CoreWeave and similar companies secures chip sales and gives it leverage over larger competitors.