The CoreWeave thesis begins with scarcity. When every major AI company in the world is trying to acquire Nvidia H100 and H200 GPUs and finding lead times of six to twelve months, CoreWeave — which had the foresight to acquire large quantities of those chips before the AI boom made them impossible to get — became, overnight, a strategic asset.

The company is Nvidia-backed, which tells you something about the relationship. Nvidia does not invest in customers for charity. It invests in companies that help it sell more chips and that validate the GPU cloud thesis against competing architectures. CoreWeave validates both.

The financials are less clean than the thesis. CoreWeave carries significant debt from the capital expenditure required to build its GPU clusters. Rental revenue from AI companies is strong, but it is also concentrated — Microsoft is reportedly a large customer, which creates both a commercial opportunity and a dependency risk. If Microsoft builds out its own GPU capacity at the pace its Azure commitments suggest, CoreWeave's utilisation rates face pressure.

The IPO conversations with investment banks are real, per reporting from mid-2026. A valuation above $23 billion — the figure from a 2024 funding round — seems likely given the market conditions. The question is timing. Late 2026 is possible. 2027 is the base case.

The FreeMalta read: CoreWeave is the pure infrastructure play in this cohort. Less narrative than OpenAI, more defensible than a model company. The Microsoft concentration risk is real. The chip scarcity moat is temporary as manufacturing scales. List while the moat exists.