China chip stocks rally as new DeepSeek model fuels AI optimism

Chinese chipmaking stocks surged on Friday after DeepSeek released previews of its new flagship open-source model, with SMIC up 11%, Hua Hong Semiconductor up over 18%, Cambricon up 4% to 6%, and Hygon up over 10%. The update boosted investor sentiment around AI-driven chip demand and China’s push for semiconductor self-sufficiency, especially after DeepSeek signaled it was using Nvidia and Huawei chips and Huawei said its Ascend supernode fully supports DeepSeek V4. The move is positive for the sector, but the news is primarily sentiment-driven rather than a fundamental earnings update.

Analysis

The market is pricing this as a simple China AI-supply-chain sympathy rally, but the more durable takeaway is that domestic model progress is starting to validate a localized compute stack. That matters because it increases procurement confidence for Chinese AI hardware buyers who have been waiting for proof that inference/training workloads can run without frontier U.S. silicon. The near-term beneficiaries are not just the chip designers; the real second-order winners are foundry capacity, advanced packaging, HBM-adjacent supply, and power/thermal infrastructure that scale with server deployment velocity.

For NVDA, the signal is mixed rather than outright negative. In the next few months, any China-specific demand uplift is likely to be constrained by export controls and procurement risk, so this is not a clean fundamental tailwind; however, it does reinforce that NVIDIA remains the benchmark architecture even when Chinese players tout domestic compatibility, which can support premium sentiment globally. The bigger risk is a narrative shift: if investors conclude China can monetize AI with partially sanctioned stacks, the market may overestimate near-term substitution and underestimate the time required to reach competitive performance at scale.

The consensus may also be underappreciating how reflexive this can get. A strong local AI release can trigger two to three quarters of capex acceleration by Chinese hyperscalers and state-linked buyers, but that capex often compresses quickly if software adoption lags or if macro stimulus disappoints. Conversely, a drawdown in enthusiasm would likely come not from model quality, but from proof that deployment economics are weak — high inference cost, low utilization, or lack of enterprise demand — which would hit the more speculative domestic names first while leaving higher-quality infrastructure beneficiaries relatively intact.

Key Decisions for Investors

• Trade the sentiment gap: long a basket of China AI hardware beneficiaries via SMIC/Hua Hong/Hygon on a 1-3 month horizon, but size smaller than headline momentum suggests; use a 10-15% trailing stop because this trade is vulnerable to fast mean reversion if follow-on orders do not appear.
• Use NVDA as a relative short hedge against China AI exuberance only tactically: short NVDA against a long basket of local China compute names for 4-8 weeks if the market starts extrapolating domestic substitution too far; the pair works best if export-control headlines stay quiet.
• Buy call spreads on NVDA rather than outright equity if you want exposure to the global AI capex cycle; the asymmetric setup is that China validation supports the broader AI narrative, but direct China revenue upside is limited, so premium should stay contained.
• Watch for a second-wave trade in packaging and equipment suppliers over the next 2-4 quarters; any evidence of sustained local AI deployment should lift the less obvious upstream names before it fully shows up in chip ASPs.
• Fade the most extended Hong Kong chip names on strength if volume confirmation fades within 2-3 sessions; the current move looks sentiment-led, and without tangible订单/guideposts the risk-reward deteriorates quickly after a 15-20% spike.