NVIDIA GPU Hardware In Stock — What Is Actually Available Right Now
GPU hardware is the one category where "in stock" matters more than anything else: factory allocation queues for current-generation NVIDIA systems run from weeks to many months. This page describes what Haink typically has available from stock in Hong Kong and Dubai, and what realistically requires allocation. Stock rotates — request current availability before planning a deployment.
NVIDIA H200 NVL (Passive PCIe, 141 GB) — Almost Always in Stock
The H200 NVL is the workhorse of PCIe AI infrastructure: 141 GB of HBM3e per card, passive cooling for standard 2U GPU servers, NVLink bridge support for 2- and 4-way card pairing. It fits Dell PowerEdge R760xa, HPE ProLiant DL380a Gen11, Lenovo ThinkSystem SR675 V3 and equivalent PCIe GPU platforms. Because it does not depend on 8-GPU HGX baseboard allocation, the H200 NVL is the fastest route to H200-class LLM inference and fine-tuning capacity — and it is almost always available from Haink stock. Memory capacity per card also makes it efficient for serving models in the 70B–180B parameter class with quantization.
NVIDIA DGX Spark — Typically in Stock
The DGX Spark personal AI supercomputer (GB10 Grace Blackwell, 128 GB unified memory) is typically available from stock. It is the practical desktop option for developers prototyping locally on models up to roughly 200B parameters before scaling to data center hardware.
Blackwell Ultra B300 — Available on Allocation
B300-class systems (Blackwell Ultra, 288 GB HBM3e per GPU) can be supplied through Haink's distribution channels on allocation. Lead times depend on OEM allocation cycles and configuration; contact Haink with target specifications for a realistic delivery estimate. For the technical comparison between generations, see H100 vs H200 vs B200 vs B300.
GPU Servers and Surrounding Infrastructure
PCIe GPU server platforms (Dell R760xa, HPE DL380a Gen11, Lenovo SR675 V3, Supermicro GPU systems) are typically available from stock or short lead time, along with NVIDIA networking (ConnectX NICs, InfiniBand and Ethernet switching), high-wattage PSUs, and NVLink bridges. Cooling for dense GPU racks is covered in liquid cooling for AI GPU servers.
Export Compliance
Advanced GPUs are dual-use items under US export regulations. Haink screens destinations and end users and supplies only export-compliant transactions — see export controls and dual-use IT hardware. For restricted destinations or use cases, Haink will decline the transaction; for compliant ones, documentation is handled as part of delivery.
Frequently Asked Questions
Is the NVIDIA H200 NVL really available from stock?
Yes — the H200 NVL (passive PCIe, 141 GB HBM3e) is almost always available from Haink stock in Hong Kong, and frequently in Dubai. It is the fastest way to deploy H200-class capacity because it avoids HGX baseboard allocation queues.
Can I buy NVIDIA B300 systems from stock?
B300-class systems are supplied on allocation rather than from open stock. Haink can quote realistic lead times through its distribution channels based on your configuration and destination.
Is the NVIDIA DGX Spark in stock?
Yes, the DGX Spark is typically available from stock and ships within days to most Asian, Middle Eastern and African destinations.
Can you ship GPUs to any country?
No. Advanced GPUs are export-controlled dual-use items. Haink screens every transaction for destination and end-user compliance with US export regulations and supplies only compliant orders.