Nvidia Backed Neutral GPU Cloud
CoreWeave: the $465M/year cloud GPU startup growing 1,760% YoY
Nvidia was not just selling chips to CoreWeave, it was helping create a neutral GPU cloud that kept Amazon, Google, and Microsoft from fully controlling AI compute. The hyperscalers all pair cloud distribution with first party accelerators, AWS with Trainium and Inferentia, Google with TPUs, Microsoft with Maia, so backing CoreWeave gave Nvidia a large buyer whose business depended on keeping Nvidia GPUs central in the market.
-
The big clouds are competitors and customers at the same time. Microsoft Azure competes with CoreWeave on GPU cloud, but also signed a $2B plus contract with CoreWeave because demand from OpenAI outstripped its own supply. That shows why an outside supplier matters even to a hyperscaler.
-
CoreWeave won by looking more like AWS for GPUs than a cheap server rental shop. Heyday used Lambda for cheaper experiments, but used CoreWeave in production because it offered Kubernetes based clusters, autoscaling, VPC style networking, and public API exposure without extra infrastructure work.
-
This market is splitting by workload. CoreWeave serves enterprises reserving thousands of GPUs on long contracts, Lambda skews toward flexible training workloads, and Crusoe differentiates on energy and data center buildout. Nvidia benefits if all of these channels expand instead of one cloud dictating the stack.
The next phase is a fight between open GPU clouds and vertically integrated AI clouds. As AWS, Google, and Microsoft push their own chips harder, CoreWeave has to become valuable for software, reliability, and customer workflow, not just access to scarce Nvidia hardware. That is what turns a temporary supply advantage into a durable platform.