B300 servers, GB300 NVL72 racks and H200 for documented Hong Kong projects, with the US export licence these installations require. DGX Spark and workstation GPUs for desk-side AI are usually available now.
Haink is headquartered in Hong Kong. First transaction? Verify serials with NVIDIA and inspect at our Sheung Wan office before paying.
| Platform | Spec | How it is supplied |
|---|---|---|
| HGX B300 | 8 × B300 288 GB, ~14.5 kW; liquid at cluster density | Allocation to documented projects |
| GB300 NVL72 | 72 × B300 in one liquid-cooled rack, ~132 kW | Project-quoted as a complete rack |
| HGX H200 | 8 × H200 141 GB, ~10.2 kW; air-coolable | Project order; lower entry cost |
| H200 NVL nodes | 2–4 × H200 NVL PCIe for inference | Project order; standard air-cooled racks |
| HGX B200, H100 | Previous platforms | On request |
| Server platforms | Supermicro, Dell, HPE, Lenovo | Chosen per project |
| Fabric | 800G / 400G InfiniBand, Spectrum-X, ConnectX | Planned with the cluster |
| DGX Spark, RTX PRO 6000 | Desk-side and workstation AI | Usually available |
Desk-side and workstation GPUs available now — snapshot 30 August 2026. Indicative prices in USD, confirmed by written quotation. Availability changes daily.
| Model | Spec | Price | Availability |
|---|---|---|---|
| NVIDIA RTX PRO 6000 | Workstation / server | $23,000 | ~1 week · HK |
| NVIDIA RTX 5090 | Workstation | $5,672 | ~1 week · HK |
| NVIDIA DGX Spark | 128 GB / 4 TB | $4,850 + ~$85 ship | Beijing / Shenzhen |
How a project runs: our procurement process → · What to prepare: what we ask before a supply request →
B300 servers and GB300 NVL72 racks for new clusters and H200 where a lower entry cost fits, for documented projects with the required US export licence. DGX Spark and workstation GPUs are usually available now — current prices are in the table above.
Yes, for advanced data-center GPUs. Hong Kong is treated like Mainland China under current US export rules; we prepare the end-user documentation and quote the project with the licence timeline included.
Yes — authorized-channel sourcing, serial numbers provided before payment and verifiable with NVIDIA or the OEM, and physical inspection welcome at our office.
Yes, as an option: 2 MW of liquid-cooling-ready capacity with our Tier 1 data-center partner, available now — see GPU cluster hosting.
Yes — configured servers with GPUs, fabric and power, tested before delivery, from a single node to a full cluster.
Running AI on this infrastructure? Haink also builds the LLM & ML software that runs on it — model, pipeline and GPUs under one contract.
Tell us the workload, the site and the target date — we come back with the platform, a costed bill of materials, the licence timeline and hosting options.
Not a quote? Ask a question