Mantle DC launches GPUaaS with 144 Blackwell GPUs - Data Center Dynamics

Mantle DC, a newly launched GPU-as-a-Service provider in the US, has announced the deployment of 144 NVIDIA Blackwell GPUs, marking a rapid entry into the AI infrastructure space. The company, founded... Read original →

Mantle DC’s rapid deployment of 144 Blackwell GPUs signals a strategic pivot from AI model hype to infrastructure-scale competition. Its NVL72 racks paired with 1.2PB all-flash storage intensify pressure on upstream EUV capacity, redirecting focus toward HBM and advanced packaging, while accelerating CXL-based memory disaggregation in storage stacks. Amid tightening U.S. export controls on AI chips to China, this fully domestic deployment mitigates geopolitical risk but inflates operational costs—each NVL72 rack draws over 100kW, potentially eroding >30% of gross margins through power and cooling. Competitors like CoreWeave may respond by bundling software toolchains or locking in enterprise SLAs. Within 18 months, the GPU-as-a-Service market will bifurcate: hyperscalers dominate full-rack training workloads, while independent providers like Mantle must pivot to inference specialization or vertical-specific fine-tuning to survive margin compression.