Behind the video
The Collapse is Starting: Is this the Black Swan?
Overall Thesis
CoreWeave faces liquidity pressure and changing AI infrastructure demand, amplified by a feedback loop between falling shares, weaker bonds and rising credit default swaps. Kevin doubts this constitutes an imminent systemic collapse, describes the broader selloff as a buying opportunity, and identifies future AI disruption of employment as the greater systemic concern.
Narratives
CRWVCoreWeaveCoreWeave's liquidity shortfall and exposure to changing AI infrastructure demand make it vulnerable to a self-reinforcing financing squeeze. Kevin nevertheless thinks default fears are overstated because supportive investors could provide financing, although existing shareholders could suffer dilution.
Key Arguments
- He estimates cash, receivables and recently raised debt provide roughly $11.7 billion against $18 billion of bills over the next 12 months.
- Local model deployment and smaller agentic models could reduce demand for some centralized GPU capacity.
- Falling shares make equity issuance harder, while falling bond prices increase the cost of additional borrowing.
- Rising credit default swap prices can reinforce perceptions of distress and trigger further selling.
- NVIDIA and other financial backers could provide support before bankruptcy becomes unavoidable.
Risks acknowledged
- Training demand will persist for quite a while.
- Bringing more capacity and power online could improve operations.
- The stock remains above its IPO price, leaving some potential to issue equity.
- A bailout could preserve the company while diluting shareholders.
- Avoiding bankruptcy does not establish a positive return for the stock.
AAPLAppleKevin presents Mac Studios as a comparatively inexpensive way to run large AI models locally. This supports an operational opportunity for Apple without establishing a forecast for its shares.
Key Arguments
- He estimates four Mac Studios could run the discussed model for about $40,000.
- He contrasts that cost with substantially more expensive GPU configurations.
- Interest in bringing AI workloads in-house could benefit Apple's hardware business.
Risks acknowledged
- The hardware comparison concerns a particular model and configuration.
- Other suppliers also offer hardware for local AI deployment.
DELLDell TechnologiesDell could benefit operationally from companies running AI models on their own premises. Kevin identifies its hardware offerings as an alternative to renting centralized compute, without predicting Dell's stock direction.
Key Arguments
- Dell sells hardware suitable for companies bringing AI workloads in-house.
- Investor attention is shifting toward local model deployment and agentic computing.
Risks acknowledged
- Apple is another supplier positioned to serve local AI workloads.
NVDANVIDIANVIDIA appears both as a supplier potentially exposed to changing compute requirements and as a major CoreWeave shareholder capable of supporting it. Kevin challenges Jensen Huang's favorable financing commentary but makes no NVIDIA share-price prediction.
Key Arguments
- Smaller agentic models and local deployment could change demand for high-end centralized GPUs.
- Kevin describes NVIDIA as a roughly 10% CoreWeave shareholder.
- He believes NVIDIA could provide the financing CoreWeave needs to continue operating.
Risks acknowledged
- Training workloads will continue to require substantial compute.
- Kevin disputes the claim that CoreWeave's cost of capital is decreasing while its bonds are selling off.
- Potential financial support for CoreWeave is speculative.
METAMeta PlatformsKevin uses a large Meta model to illustrate the memory and hardware requirements of local AI deployment. The discussion supplies technical context rather than an investment thesis or stock forecast for Meta.
Key Arguments
- He estimates the discussed model requires about 700 gigabytes of RAM.
- He compares several hardware configurations that could accommodate that model.
Hedges & Caveats
- Kevin discloses no position in CoreWeave.
- He acknowledges CoreWeave's weak balance sheet and expects potential rescue financing to dilute shareholders.
- He questions the market-implied 64% default probability rather than ruling out bankruptcy.
- He expects AI model training to continue for quite a while despite a short-term shift in investor attention.
- His closing buy-the-dip recommendation does not clearly identify a specific asset.
- He expects the AI bubble to collapse eventually but supplies neither a specific tradable instrument nor a deadline.