Four CEOs on the Future of AI: CoreWeave, Perplexity, Mistral, and IREN

CoreWeave began in 2017 as a hedge-fund spin-off that mined cryptocurrency with Nvidia GPUs. When crypto volatility rose, the team repurposed the hardware for CGI rendering, batch computing, and eventually neural-network training. By donating A100 GPUs to the open-source Eleuther AI project, they learned how to run large-scale parallel workloads and identified a market gap: providing purpose-built GPU clusters for AI researchers. This early focus on scaling laws—recognizing that raw compute alone does not create value—positioned CoreWeave to serve the emerging demand for training-grade infrastructure before the ChatGPT boom.

To fund that infrastructure, CoreWeave created a novel financing vehicle known as the box. The company signs multi-year compute contracts—typically five-year deals—then bundles the contract, GPU purchase, and data-center costs into a single cash-flow waterfall. Lenders receive principal and interest before any residual profit returns to CoreWeave, allowing the firm to raise $35 billion in 18 months through GPU-backed loans. This structure pays off the capital within roughly two and a half years, proving that GPU assets retain economic life far beyond the industry-wide two-year depreciation myth.

The market now competes fiercely for the latest Nvidia chips, from H100s to the upcoming H300s. CoreWeave's first-to-scale deployments give clients immediate access to bleeding-edge hardware, which later migrates to inference workloads that generate recurring revenue. As more startups and sovereign funds enter the GPU market, allocation follows a simple first-come, first-served rule, reinforcing the sector's health. With inference serving as the primary monetization point for AI models, the longevity of GPUs—often exceeding six years—supports sustainable growth and underscores why financing models like CoreWeave's are reshaping AI infrastructure investment.