NVIDIA and CoreWeave have found a balance between partnership and competition

A host reflected on how NVIDIA and CoreWeave’s relationship developed amid NVIDIA’s earlier ambition to act as an intermediary between cloud providers and customers. The host said the idea behind DGX Cloud Lepton—aggregating compute resources and competing with AWS, Azure, and Google Cloud—worried neocloud providers: refusing to participate could mean NVIDIA steered customers to competitors, while participating risked weakening direct customer relationships and compressing margins. NVIDIA acquired Lepton in April 2025; an industry estimate discussed at the time put the deal at $300 million to $900 million, but the host stressed that this was a rumor.

Rather than becoming the expected intermediary that allocates demand, the product evolved, in the host’s assessment, into a “unified AI platform” with proprietary software for managing GPUs: node groups, development environments, batch jobs, inference, storage, observability, and reservations. Users can still use their own hardware and maintain a direct relationship with its owner. The host added that initial DGX reviews did not establish an advantage over working directly with CoreWeave, and SemiAnalysis tests suggested the system performed less well than some alternatives.

The hosts separately discussed CoreWeave Forge, through which the company is expanding its cloud with CPUs, storage, and management tools. One host saw the move mainly as intensifying competition with roughly 300 neocloud companies, rather than as a broad attack on market participants. The host pointed to rising value per unit of compute and expressed optimism about CoreWeave’s business; the comparison with Elon Musk and Jensen Huang’s $60 billion-per-gigawatt figure was presented as a conditional benchmark, not a confirmed outcome.