Customers

Anchor customers

The two biggest publicly-known customer relationships are Microsoft and Meta (covered in chapters 03 and 04). These two represent a significant share of total CoreWeave revenue.

Other large customers include hyperscaler-tier partnerships and frontier AI labs whose specific commitments aren't always publicly disclosed in detail.

Frontier model labs

CoreWeave serves several major frontier AI labs. The relationships are characterized by:

• Large dedicated clusters (hundreds to thousands of GPUs).
• Custom configurations and software-stack support.
• Multi-year commitments.
• Deep operational integration — CoreWeave engineers may work closely with the lab's infrastructure team during cluster bring-up.

Specific lab customer names are sometimes public, sometimes not. The pattern is consistent: high-end research compute that demands the cluster fabric and operational quality CoreWeave delivers.

Enterprise AI adopters

Beyond the marquee names, CoreWeave serves mid-to-large enterprises building AI capabilities:

• Financial services firms building AI-native products.
• Healthcare and pharma running large drug-discovery models.
• Media and entertainment with content-generation AI workloads.
• Other sectors with AI-driven product or research initiatives.

These are typically annual-reserved contracts in the seven-to-eight-figure range — meaningful revenue, less attention than the marquee deals.

Research and university

CoreWeave has partnerships with academic and research institutions, providing capacity for major research projects. Some are publicly announced; many are smaller and not publicized.

The research segment doesn't drive massive revenue but provides credibility and a pipeline of relationships that may evolve into commercial deployments.

Concentration profile

The customer concentration is the single most-discussed financial feature of CoreWeave:

• Top 1-2 customers are likely 35-50% of revenue.
• Top 5 customers are likely 60-75% of revenue.
• The long tail of mid-size enterprises is a smaller share.

This is structurally different from a hyperscaler's customer mix (where no single customer is a large fraction of revenue). It's a feature of CoreWeave's specialized positioning — selling concentrated capacity to compute-hungry buyers — but it creates risk.

Workload types

What customers actually run on CoreWeave:

• Large-scale training. The single biggest use case. Multi-thousand-GPU runs that require the InfiniBand fabric and the operational quality.
• Fine-tuning and customization. Often at hundreds of GPUs; shorter than training but still substantial.
• High-throughput inference. Production deployments serving large user bases.
• Research compute. Less-structured workloads from academic and lab users.

The common factor: workloads where dedicated, predictable cluster compute matters more than absolute lowest unit price.

Takeaway

CoreWeave's customer base is concentrated, enterprise-oriented, and structurally distinct from any other neocloud or hyperscaler. The concentration creates revenue visibility and risk in equal measure. The next chapter walks through the financial structure that supports this customer mix.