QumulusAI announced the purchase of 1,632 NVIDIA Blackwell B300 GPUs across 204 NVIDIA HGX B300 systems as it expands capacity to meet growing customer demand for AI infrastructure. The Atlanta-based neocloud provider said the acquisition represents one of its largest single infrastructure expansions and was financed primarily through Technology Finance Corporation and USD.ai. The deployment also includes 192 NVIDIA RTX PRO 6000 Blackwell GPUs to support visualization and specialized AI workloads.
The new systems will support multiyear customer contracts already secured by the company, including more than $124 million in three-year AI inference agreements announced earlier this year. QumulusAI said its distributed deployment model allows it to source hardware from multiple OEM partners, including Supermicro and Lenovo, while installing infrastructure across a combination of colocation facilities and company-owned data centers. The company describes this architecture as a hyperdistributed AI cloud designed to place GPU capacity closer to enterprise demand.
QumulusAI said its deployed GPU fleet expanded by more than 450% between June 2025 and June 2026. The company positions the new Blackwell HGX B300 systems for large-scale AI training and high-throughput inference, while the RTX PRO 6000 Blackwell GPUs provide additional flexibility for inference, visualization, and workstation-class AI applications. The company said its F.A.C.T.S. deployment model matches compute platforms to specific workload requirements rather than relying on a single GPU architecture.
• Purchase includes 1,632 NVIDIA Blackwell B300 GPUs in 204 HGX B300 systems.
• Additional deployment includes 192 NVIDIA RTX PRO 6000 Blackwell GPUs.
• Hardware sourced through multiple OEM partners including Supermicro and Lenovo.
• Expansion supports more than $124 million in previously announced three-year AI inference agreements.
• Purchase financed primarily through Technology Finance Corporation and USD.ai.
• QumulusAI reports deployed GPU fleet growth exceeding 450% over the past year.
• Infrastructure deployed across both colocation and company-owned data centers.
“Meeting AI demand today is a test of access, speed and flexibility, and you can’t win on one without the other two. Our model gives us the flexibility to source across multiple OEM partners such as Supermicro and Lenovo and to deploy across a national network of colocation and owned facilities. When demand moves, we move with it. That’s how 1,632 B300 GPUs go from purchase order to production at hyperspeed.” — Mike Maniscalco, CEO of QumulusAI.
🌐 Analysis: QumulusAI joins a growing group of neocloud providers that are rapidly expanding GPU infrastructure to serve AI startups, enterprises, and model developers that cannot always obtain sufficient capacity from traditional hyperscale cloud providers. The purchase also highlights continued market momentum for NVIDIA’s Blackwell generation, with HGX B300 platforms becoming the preferred building blocks for both AI training clusters and large-scale inference deployments.
