Culper Research shorts Nvidia over alleged China trade violations

Culper Research said it is short Nvidia, alleging the company still derives over 20% of fiscal 2026 compute revenue from China despite prior claims that the China business went to zero after April 2025 U.S. restrictions. The report alleges illegal GPU diversion, intermediary financing links involving Megaspeed and Alibaba, and references a March 2026 DOJ indictment tied to $2.5B of smuggled Nvidia-powered servers. The claims point to renewed China exposure and export-control risk, which could pressure Nvidia shares.

Analysis

This is less about a single short-seller note and more about the market re-pricing the durability of the China AI revenue bridge. If the diversion thesis is even partially right, the most vulnerable leg is not just NVDA’s direct China line, but the entire “non-China” channel used to launder China demand through Southeast Asian build-and-ship routes. That would hit margin expectations unevenly: OEM/assembler names with the highest exposure to proxy shipments and least transparent end-demand should de-rate fastest, while pure-play semicap and U.S. cloud infrastructure vendors may prove more insulated.

The second-order winner is domestic Chinese AI and networking vendors. A sustained blockade on Nvidia-class accelerators accelerates substitution into local silicon, which likely improves unit share but at the cost of lower near-term performance and higher total system cost. That creates a subtle negative feedback loop for the whole global AI stack: as China localizes, U.S. export controls become less effective on revenue capture but more effective on fragmenting standards and reducing the addressable market for U.S. compute ecosystems over the next 4-8 quarters.

For BABA, the key issue is not direct earnings leakage but regulatory and governance overhang: any credible financing or facilitation link to a prohibited procurement chain raises the probability of secondary scrutiny, banking friction, and board-level capital allocation constraints. The asymmetry is better in SMCI than NVDA for a tactical short because the stock is more hostage to shipment integrity, working-capital optics, and customer-concentration rumors. NVDA can absorb headline risk better, but if Q1/Q2 shipment data confirms sequential China contraction, the multiple compression can persist for months.

Contrarian risk: the market may be overestimating how much of this is incremental information versus already-discussed export-control leakage. If management can show a clean channel audit and China mix stabilizes through alternative product segmentation or non-China demand re-acceleration, the stock can rebound quickly on any AI capex print. The real pivot will be in the next two earnings cycles and customs/trade data, not the accusation itself.

Key Decisions for Investors

• Short NVDA into strength or via call spreads over the next 2-6 weeks; prefer defined-risk structures because the stock can absorb headline noise, but valuation multiple risk remains if China revenue transparency worsens.
• Short SMCI outright or buy puts for 1-3 months; highest leverage to shipment-disruption and intermediary scrutiny, with better downside convexity than NVDA if trade records or DOJ follow-on actions expand.
• Reduce/hedge BABA exposure with short-dated downside protection for 1-2 months; the risk is not earnings compression but a governance/regulatory overhang that can re-rate the ADR lower on any financing-link confirmation.
• Pair trade: long domestic China AI beneficiaries against short NVDA/SMCI for 3-6 months; if import restrictions harden, local substitution accelerates even if absolute demand slows.
• If already short NVDA, take partial profits into any sharp opening gap and re-enter on evidence of shipment deceleration in customs data; the catalyst path is data-confirmation, not the initial headline.