Nvidia Denies Requiring Prepayment for H200 GPU Shipments to China
#Hardware

Nvidia Denies Requiring Prepayment for H200 GPU Shipments to China

AI & ML Reporter
2 min read

Nvidia refutes claims it demanded upfront payments from Chinese customers for unreceived H200 GPUs amid ongoing US export restrictions and supply constraints.

Featured image

Nvidia has formally denied allegations that it required Chinese customers to make upfront payments for its H200 GPUs before receiving hardware. In a statement to Reuters, a company spokesperson asserted: "We would never require customers to pay for products they do not receive." This rebuttal addresses industry rumors about potential cashflow tactics amid tightened US export controls on advanced AI hardware.

The H200 represents Nvidia's current flagship datacenter GPU, featuring 141GB of HBM3e memory and 4.8TB/s bandwidth - specifications critical for large language model training and high-performance computing workloads. Since its November 2025 launch, the H200 has faced extraordinary demand coupled with manufacturing constraints, particularly affecting customers in China where US export regulations limit shipment volumes.

Industry analysts note several factors contributing to market tension:

  1. Export Compliance: The H200 falls under US restrictions prohibiting direct sales to Chinese military entities and requiring special licenses for civilian use cases
  2. Supply Allocation: Global H200 allocation remains tight, with typical lead times exceeding 6 months
  3. Secondary Market Premiums: Unauthorized resellers reportedly charge 200-300% markups on available units

While Nvidia's denial specifically addresses payment-before-delivery practices, it doesn't clarify whether Chinese customers face:

  • Extended payment terms compared to other regions
  • Higher deposit requirements
  • Conditional allocation based on purchase history

The company continues developing China-specific products like the HGX H20 (an export-compliant L40S variant) to maintain market presence. However, performance compromises remain significant - the H20 delivers approximately 80% lower FP64 performance than the H200 according to Nvidia's architecture documentation.

Recent financial disclosures reveal China accounted for 9% of Nvidia's Q4 2025 datacenter revenue, down from 22% before export restrictions intensified. The ongoing situation demonstrates how geopolitical factors continue disrupting global AI hardware supply chains, forcing companies to navigate complex compliance landscapes while maintaining customer trust.

Third-party logistics providers interviewed by Reuters suggest some Chinese firms have begun prepaying for GPU access through Singaporean and Taiwanese intermediaries, though these arrangements operate outside Nvidia's official channels. The company maintains its standard global payment terms apply uniformly across regions.

Comments

Loading comments...