NVDA China Talks — Top AI Firms Request Approval for Limited H200 GPUs

Report Highlights

• Source: The Information (summarized by Shay Boloor @StockSavvyShay, PBOC-related)

• Requester: China (on behalf of its top AI firms)

• Ask: approval to purchase a limited number of NVIDIA H200 GPUs

• Status: under US government review

Purpose

• Ease AI training chip shortage

• Break training workload bottleneck for top AI firms

Parallel Policy

• Inference workloads to remain on domestic processors

• H200 for training only · dual-track strategy

Target Chip (NVIDIA H200)

• Top-tier Hopper architecture

• HBM3e 141GB · bandwidth ~4.8 TB/s

• Materially better training throughput than H100

Scenarios

1) US approves → NVDA China revenue resumes · upside optionality restored

2) Denied → training bottleneck persists · pressure on domestic substitutes rises

3) Partial approval → ongoing negotiations on volume and eligible firms

Investment Implications

• China revenue optionality restored (previously largely priced out)

• CUDA/software moat re-confirmed for training workloads

• Inference market share risk persists → domestic processor validation news is the key forward trigger