Huawei's 910D Arrives, H20 Departs

Why This Matters

Nvidia faces simultaneous competitive and regulatory threats in China. Huawei's Ascend 910D approaches H100 performance while H20 export bans and tariffs eliminate Nvidia's ability to serve Chinese customers. The gap between Street estimates and likely reality creates meaningful downside risk.

The Core Investment Thesis

Nvidia's China business faces 50-90% revenue decline depending on scenario. Street estimates appear too optimistic, and the competitive moat (CUDA) is eroding as DeepSeek's open-source GPU software enables alternatives. The stock may underperform as China reality sets in.

Key Arguments

Argument #1: Huawei 910D Closes Performance Gap

Chinese domestic alternatives are approaching competitive parity with Nvidia's offerings.

Data: Huawei Ascend 910D: 2 PFLOPS in BF16 — within ~10% of H100 performance. Chinese datacenters pay ~5¢/kWh versus 13¢ in Silicon Valley. Inferior power efficiency becomes less economically penalizing in China's cost structure.

When performance is 90% equivalent and operating costs are 60% lower, the total cost of ownership calculation favors domestic alternatives even without considering regulatory pressure.

Argument #2: Export Controls and Tariffs Compound

Nvidia faces simultaneous revenue loss from bans and margin compression from tariffs.

Data: H20 Ban impact: $5.5 billion one-time charge. Annual revenue collapse estimate: $8-10 billion. 32% duties on Taiwan-origin chips add ~$18,000 per DGX H100 server. Gross margin drag: 120-150 basis points.

The dual impact of lost China sales and margin compression on remaining business creates earnings pressure not fully reflected in consensus estimates.

Argument #3: CUDA Moat Eroding

Nvidia's software ecosystem lock-in is weakening as alternatives emerge.

Data: DeepSeek's open-source GPU software undermines CUDA's proprietary ecosystem lock-in. Risk of industry-wide price war as competitors gain traction. Compelled migration away from Nvidia in China creates precedent for other markets.

CUDA has been Nvidia's deepest moat. If Chinese developers successfully migrate to alternatives under regulatory pressure, the playbook exists for others.

Scenario Analysis

• Base Case (60% probability): 50% China revenue decline by FY-27; high single-digit stock drag. Street estimates gradually revise lower.
• Bear Case (25% probability): 80-90% China revenue collapse plus 25% global tariffs; mid-20% downside. CUDA erosion accelerates globally.
• Bull Case (15% probability): Supply constraints limit Huawei; China impact contained to ~10% of datacenter segment. SMIC yields fail to scale.

Bottom Line

Nvidia faces structural China headwinds that consensus estimates underappreciate. The gap between Street projections ($10B+ China datacenter) and likely reality (<$5B) creates meaningful downside risk. Monitor Huawei benchmark performance, Nvidia margin guidance, and SMIC yield sustainability for scenario confirmation.

Verdict: Consensus too optimistic on China; downside risk material