CoreWeave

Full briefing

CoreWeave was founded in 2017 in Roseland, New Jersey, originally as a cryptocurrency mining operation using consumer-grade NVIDIA GPUs. When GPU-based proof-of-work mining became unprofitable around 2018, co-founders Michael Intrator, Brian Venturo, and Brannin McBee pivoted the company to provide cloud GPU rental services for graphics rendering and scientific computing. The AI boom of 2022–2023 transformed CoreWeave into one of the fastest-growing companies in technology, as demand for GPU capacity for LLM training vastly outstripped what traditional cloud providers (AWS, Azure, GCP) could deliver. CoreWeave's technical differentiation lies in its GPU-native infrastructure design. Unlike hyperscalers that provision GPU instances as a small subset of a broader compute portfolio, CoreWeave builds data centers purpose-designed for dense GPU deployment — with cooling systems, power density (upward of 50 kW per rack), and network fabrics optimized specifically for NVIDIA's NVLink and InfiniBand interconnects. The company operates clusters of NVIDIA H100 SXM5 and H200 SXM5 GPUs with NVLink Switch fabric providing 900 GB/s bidirectional bandwidth between each GPU pair within a node, and NVIDIA InfiniBand NDR (400 Gb/s) between nodes. Arista Networks Ethernet switches serve the front-end management and storage networking fabric. NVIDIA made a strategic investment in CoreWeave of approximately $100 million in 2023, a highly unusual move for a chip company. This investment signaled NVIDIA's interest in ensuring that its largest GPU customers had the capital to absorb GPU production at scale, creating a quasi-guaranteed demand floor. CoreWeave in turn received preferential access to GPU allocations during the H100 supply constraint period of 2023–2024, when demand for H100s significantly exceeded supply. The company has signed long-term GPU supply agreements with NVIDIA that extend into the Blackwell generation. Major CoreWeave customers include Microsoft (a reported $10 billion multi-year commitment), OpenAI, Cohere, and IBM. Microsoft's reliance on CoreWeave capacity — to supplement Azure's own GPU build-out — was disclosed in CoreWeave's IPO filing and represents one of the largest infrastructure dependency relationships in the AI industry. CoreWeave IPO'd on Nasdaq (CRWV) in March 2025 at approximately $23 per share, raising $1.5 billion in the offering and valuing the company at approximately $19 billion. CoreWeave's supply chain position is unusual: it is simultaneously a large customer of NVIDIA (for GPUs), a customer of Arista (for networking), a customer of major power and cooling infrastructure suppliers, and a supplier to AI model companies. Its debt load — incurred to finance GPU purchases — stood at approximately $8 billion at IPO, reflecting the capital intensity of building GPU clusters at hyperscale. This leverage means CoreWeave's financial stability is closely tied to the continued high utilization rates of its GPU fleet, making it a sensitive indicator of the overall health of the AI training market.