AI infrastructure provider SuperX (SUPX) disclosed owned GPU servers operating in third-party data centers, offering on-demand and reserved-capacity services.

SuperX AI Technology Ltd. (SUPX) disclosed its operating model in a Form 20-F annual report filed with the U.S. Securities and Exchange Commission on October 29. Its GPU-as-a-Service (GPUaaS) offering uses GPU servers that the company owns and operates under hosting arrangements at third-party data centers, rather than establishing ownership of the buildings themselves.

The filing describes a two-layer cloud-software stack: a compute layer providing GPU hosting, model deployment, monitoring, and operations, and a model-services layer offering token-based access to hosted third-party AI models. Services are available on demand or through reserved capacity, with usage-based fees and possible monthly minimum-consumption commitments depending on the customer agreement; its unified AI Token Platform launched September 28, 2026.

The Denver AI Inference Cloud Hub, dated June 4, 2026, is described as SuperX’s first North American inference-cloud deployment and is operated by SuperX Singapore. The Singapore AI Innovation Centre, dated April 15, 2026, is hosted at ST Telemedia Global Data Centres’ STT Singapore 5 facility, where SuperX supplies NVIDIA Blackwell GPU servers, software, and workload orchestration, with workloads processed locally for applicable data-residency requirements.

Separately, SuperX offers hardware from other original equipment manufacturers, including B300 GPU systems and a GB300 NVL72 rack-scale platform integrating 72 Blackwell Ultra GPUs, 36 Grace CPUs, 288 GB of HBM3e memory per GPU, fifth-generation NVLink, ConnectX-8 networking, and liquid cooling. These product specifications are not a disclosed inventory of its hosted fleet, and the filing warns of competition from larger server vendors, systems integrators, hyperscalers, and specialist GPU-cloud providers.