For nearly two years, CoreWeave symbolized the urgency of AI infrastructure development. A former player in cryptocurrency mining converted to the GPU cloud, the American company has established itself as a key supplier of computing power for generative AI, taking advantage of privileged access to Nvidia chips and a market in chronic shortage. The IPO, the failure of a strategic acquisition and the rise in financial constraints have, however, highlighted the fragilities of a model that has become very capital-intensive and illustrates the questions about the AI bubble.
An IPO which marks the passage under market constraints
CoreWeave’s IPO, finalized in spring 2025, represents a tipping point for the company. Until then, the company operated in a world of private capital where the priority was the speed of deployment; it is now deciphered by financial analysts with a completely different reading framework.
While financial documents reveal a rapid growth trajectory, it remains highly concentrated. Microsoft represents more than 60% of turnover, while two clients concentrate most of the revenue. At the same time, despite nearly $2 billion in revenue in 2024, the company has high losses and a cost structure dominated by hardware investments.
Debt as an accelerator… and as a structural constraint
To support its growth, CoreWeave uses financial arrangements that can be described as innovative, in particular a debt backed directly by its Nvidia GPUs. This financial engineering allows it to be deployed quickly, but also reinforces the capital intensity of the model.
Debt exceeds $8 billion, in a context where each new generation of chips requires additional investments to remain competitive, a model that relies on the continuity of demand and the ability to regularly refinance the infrastructure without major deterioration in financing conditions.
This dependence on the cost of capital becomes more sensitive when the security enters a correction phase. Debt, which has long been a lever for growth, weighs heavily on the company.
The takeover of Core Scientific, an attempt at industrial security
It is in this context that CoreWeave launched, in the summer of 2025, a takeover offer for Core Scientific, a former major player in bitcoin mining converted to the operation of high energy density data centers. The operation, valued at $9 billion and entirely in shares, aimed to internalize part of data center capacities, secure access to energy and reduce several billion dollars in long-term rental commitments. On an industrial level, the logic was coherent, on a financial level, the structure of the deal immediately aroused distrust.
The absence of a protection mechanism against the volatility of CoreWeave stock directly exposed Core Scientific shareholders, thus activists, hedge funds and proxy advisors opposed the merger. CoreWeave’s refusal to review the terms seals the failure of the project.
A stock market correction which recomposes the reading of the model
In the weeks that followed, CoreWeave shares fell significantly, a correction that went beyond a simple stock market adjustment and led investors to re-examine the solidity of the economic model.
In addition to a high customer concentration, while the main customers have the means to internalize their infrastructures, they note a dependence on Nvidia, a strategic supplier and dominant player in the value chain and an increasing exposure to energy costs and the constraints linked to data centers, which have become heavy industrial assets.
CoreWeave, revealing a regime change for AI
If CoreWeave’s trajectory does not call into question the reality of the demand for computing for AI. It highlights a change in regime. So after a phase of rapid expansion, financed by urgency and scarcity, the AI infrastructure is entering a period of selection.
From now on, the question is no longer just about accessing GPUs, but about demonstrating the economic sustainability of a capital, energy and debt intensive model. CoreWeave remains a central player in the global AI ecosystem. But its future will depend on its ability to reduce its customer dependence, stabilize its financial structure and clarify its industrial strategy in the face of hyperscalers.
The company was listed at the close of December 15 at $75.35 compared to $187 at its highest in June 2025.
