CoreWeave's origin story sounds almost accidental. Michael Intrator and Brian Venturo came from energy-focused hedge-fund investing, while Brannin McBee worked in data and finance. Their early GPU purchases were aimed at cryptocurrency mining, first on a pool table and later at much larger scale. The important founder decision came when the economics of crypto weakened and demand for GPU computing from machine-learning companies accelerated.
Rather than liquidating hardware into a downturn, the team repositioned the infrastructure as a specialized cloud. That pivot put CoreWeave in front of the most important physical constraint of the generative-AI boom: access to large clusters of advanced accelerators with power, networking and software configured for training and inference.
The moat is physical as much as software
AI cloud companies can look like ordinary software businesses from the outside, but CoreWeave's growth depends on acquiring chips, data-centre capacity, power and financing before customers need it. That makes capital allocation and supply-chain execution central founder skills.
The company differentiated itself from general-purpose hyperscalers by building around GPU-heavy workloads and offering more flexible configurations. As model developers demanded ever larger clusters, specialization became an advantage rather than a niche.
Going public changed the scrutiny
CoreWeave's 2025 IPO converted a venture-backed infrastructure story into a public-market test of the economics of AI compute. Investors can now see both sides of the model: extraordinary revenue demand and extraordinary capital requirements.
That tension is structural. If utilization stays high and customer contracts remain durable, expensive infrastructure can compound quickly. If demand slows or hardware depreciates faster than expected, leverage and fixed commitments become much more visible than they are in a conventional SaaS company.
The India expansion shows the next phase
On 6 October 2026 CoreWeave announced a planned 240MW deployment in Navi Mumbai with AdaniConneX, alongside a local office and hiring plan. The company intends to deploy Nvidia's Vera Rubin platform there for training, inference and agentic workloads.
For the founders, this is no longer a US capacity story. CoreWeave is becoming a global infrastructure operator whose success depends on repeating the same difficult playbook across power markets, data-centre partners and regulatory regimes. The company that started with a few GPUs is now testing whether specialized AI cloud can scale like a global utility.