CoreWeave has carved out a unique niche by rejecting the generalist approach of the major public clouds. Gone are the thousands of ancillary services and architectures designed for versatility. The company has set itself a clear mission: to provide the best possible infrastructure for AI-intensive workloads, whether it's model training, ultra-low-latency inference, or complex graphics rendering.

To do this, it deploys massive clusters of the latest generation of GPUs (Nvidia Hopper, Blackwell) in its own data centers in the United States and Europe.

It then offers this raw power to its customers through a business model based on multi-year "take-or-pay" contracts, guaranteeing stable and predictable revenue. In other words, customers commit to paying for reserved resources, whether they are used or not. This model allows it to invest heavily in expanding its capabilities (32 data centers, over 250,000 GPUs in 2025) while securing essential cash flow in an extremely capital-intensive industry.

Technological differentiation through vertical integration

CoreWeave attracts AI leaders thanks to its unique performance promise. The company doesn't just offer GPUs: it has built an integrated cloud platform optimized for end-to-end AI.

Its bare metal architecture, with no virtualization layer, enables enhanced isolation and maximum performance. In-house software innovations such as CKS (CoreWeave Kubernetes Service) and SUNK (Slurm on Kubernetes) facilitate the management of massive clusters for training models on tens of thousands of GPUs.

CoreWeave has also developed Nimbus, its own control plane based on NVIDIA's Bluefield DPU chips, which provides intelligent network scheduling and enhanced security at the node level.

The demonstration speaks for itself: according to MLPerf v5.0 benchmarks, CoreWeave outperforms the H200 by 2.86× on the Llama 3.1 405B model in inference, and its GB200 cluster for training is 34 times larger than the second-ranked cluster.

 

CoreWeave is not just a provider of raw power: it has become the trusted partner of the biggest names in AI. In addition to Microsoft (which accounted for more than 50% of its revenue), the company has signed a five-year contract with OpenAI valued at nearly $12bn, which also includes an equity stake in CoreWeave.

Other big-name clients include Meta, IBM, Cohere, Mistral AI, and Jane Street. This privileged position among the global AI elite is a testament to its recognized technological excellence, but it also raises structural concerns: the extreme concentration of its revenue makes CoreWeave vulnerable to any strategic changes by its major clients, some of which are also its competitors.

CoreWeave, the AI Hyperscaler™, was the only AI cloud provider to receive the highest Platinum rating based on SemiAnalysis' ClusterMAX™ rating system.

Dizzying growth

By 2025, analysts expect revenue of $5bn, supported by the order book and the ramp-up of the OpenAI contract.

The company has achieved the feat of generating an adjusted EBITDA margin of 64% by 2024, a sign of exceptional potential profitability despite heavy investment. It has raised more than $12.9bn through asset-backed financing (notably its GPUs) and contracts signed with investment-grade customers.

However, this dependence on external financing is a point of concern. CoreWeave remains heavily indebted, with negative free cash flow, and will need to continue to convince the markets to support its growth rate.

A strategic position in a highly competitive market

The AI infrastructure market is expected to grow by more than 30% per year through 2030. CoreWeave is establishing itself as the only pure player at hyperscale, facing two types of competitors:

  • Generalist hyperscalers (AWS, Azure, Google Cloud)

These giants have huge financial resources and product portfolios. They can compete with CoreWeave on access to GPUs (or even develop their own silicon), offer integrated bundles, and leverage long-standing customer relationships.

CoreWeave is countering this with faster execution, cutting-edge technological specialization, and services tailored to the specific needs of AI labs. But the threat is existential, particularly from Google Cloud (TPUs, Vertex AI) and Microsoft Azure, which is both a customer and a competitor.

  • Specialized neo-clouds (Lambda, Together AI, Crusoe)

With less capital, these players stand out for their agility, open-source focus, or green credentials. They sometimes target the same customers as CoreWeave, but struggle to compete in terms of raw capacity and reliability. CoreWeave maintains its lead thanks to its size, access to capital, and operational excellence.

An ambitious valuation, a bet on the future

CoreWeave went public in March 2025, raising $1.5bn in the largest IPO in the AI sector to date. The share price is currently trading at around $155 per share, giving it a market capitalization of $75bn. Its stock market performance since the IPO has been impressive: +300% in just a few months. The current valuation leaves no room for error. The stock is trading at almost 20x its estimated revenue for 2025.

Investors expect strong growth in the coming years in a very popular sector, hence the very high ratios. Indeed, revenue for 2026 is estimated at $11.6bn and for 2027 at $16.5bn.

Investors who are willing to pay the current price for the stock are betting on three things: explosive demand for specialized AI infrastructure, CoreWeave's outstanding operational execution, and solid economic fundamentals (strongly growing operating margins, recurring revenue). But there are many risks: customer dependence, competitive pressure, ongoing need for financing, and GPU scarcity.

A tech gem to watch very closely

CoreWeave is establishing itself as the technology champion of AI infrastructure, combining raw power, software innovation, and operational excellence. In an industry where the race for performance is key, it has become the preferred partner of AI leaders. But its future will depend on its ability to manage its growth, diversify its customer base, and maintain its technological lead against giants with virtually unlimited resources.

In a world where AI is becoming infrastructure, CoreWeave wants to be the equivalent of what Amazon was for e-commerce: the invisible but essential backbone of innovation. It's a bold bet... but not an unrealistic one.