Source GPUs and AI servers

§1What we handle

• Renting capacity — reserved blocks or on-demand clusters through our rental partners; we collect offers for the model, size and region you need, so you can compare them against the index.
• Buying hardware — GPU servers (eight-GPU HGX/OAM-class systems, PCIe servers), components and rack integration through our hardware partners.
• Mixed plans — rent now, own later, or the reverse; tell us both horizons.

§2What to include

Model and quantity are the minimum. Location matters twice: for delivery and duties when buying, for latency and data-residency when renting. The timeline decides whether stock or a factory slot is needed. A budget range is optional, but it lets us say early if a plan is not workable — for reference, the index median for an H100 is $3.49 per GPU-hour as of 16 Oct 2026.

Useful extras in the notes field: server form factor (SXM or PCIe), interconnect (InfiniBand or Ethernet), storage, the power available per rack and whether the site has liquid cooling.

§3Terms in plain words

The Dream View Co., Ltd. is a procurement agent registered in Phuket, Thailand. Terms are agreed individually for each request. GPUs and GPU servers are available both to rent and to buy, through different partners, so a rental request and a purchase request get equal attention. If you are undecided, we can price both. NVIDIA and AMD accelerators are subject to US export controls; we only supply where the buyer, destination and end use allow it, and we will ask about them.