QumulusAI signs $124m deals for 1,280 Blackwell GPUs

QumulusAI has secured more than $124m in three-year AI infrastructure agreements with Hyperbolic and an unnamed AI inference platform. The contracts include $21.9m in upfront commitments and cover 1,280 Nvidia Blackwell GPUs across 160 Lenovo and Supermicro bare-metal servers. The deployment site remains undisclosed, making Denton and other US colocation markets key follow-up points.

• Hyperbolic and another inference platform signed three-year infrastructure commitments
• Denton approval and $45m note show capital-intensive expansion path

The fact

QumulusAI has secured more than $124m in three-year AI infrastructure agreements with open access AI cloud provider Hyperbolic and an unnamed AI inference platform. The contracts include $21.9m in upfront commitments and cover 1,280 Nvidia Blackwell GPUs across 160 Lenovo and Supermicro bare-metal servers using B300 and B200 chips. Cisco Nexus One will power the cluster fabric. QumulusAI has not disclosed the deployment site, but it previously won approval for a planned 20MW modular data centre in Denton, Texas.

The Assessment

The contracts give QumulusAI something more concrete than an AI infrastructure roadmap: customers committing cash before the deployment site is public. That matters because Blackwell procurement, bare-metal server delivery and data centre capacity all require capital before revenue is fully realised. The signal for BTW readers is that QumulusAI is using customer pre-commitments to de-risk a GPU buildout, while testing whether smaller AI infrastructure providers can move faster than hyperscalers on production inference deployments.

What to Watch

Watch whether QumulusAI places the cluster in Denton or another US colocation market, discloses the second customer, and converts its $45m note into timely Blackwell delivery.

Also read: AI workload volatility raises data centre power waste

Also read: AI data centres face medium-voltage equipment supply bottleneck