Chinese chip firms hit record high revenue driven by the AI boom and U.S. curbs

SMIC reported 2025 revenue up 16% YoY to $9.3B (LSEG estimates put 2026 revenue > $11B); Hua Hong posted Q4 revenue of $659.9M and guided $650–$660M, while Moore Threads guided 2025 revenue of ¥1.45–1.52B (up 231–247% YoY). CXMT saw revenue jump ~130% YoY to >¥55B (~$8B), driven by memory shortages and surging AI demand. Growth is being amplified by U.S. export curbs that have boosted domestic procurement, but Chinese firms still lag technologically (no ASML EUV access) and face risks of overcapacity in mature nodes.

Analysis

China’s forced onshoring of compute and memory is creating a captive, high-margin domestic market that cushions local suppliers from direct global competition and raises the value of local capacity investments. That captive-demand dynamic is a durable source of revenue but concentrates downside risk: if inventory cycles or global memory-price normalization occur, domestic players with slower process nodes will be first to suffer pricing pressure because their competitive moat is policy-dependent rather than technology-led.

Second-order supply-chain winners include pure-play tool and foundry partners who sit one step upstream from the capability race — firms that control critical masks, testing, or packaging tooling can monetize China’s scale even when EUV remains barred. Conversely, vendors with direct China-facing GPU/IP exposure face a bifurcated TAM: premium parts (EUV-era GPUs) remain out of reach, while lower-tier GPUs and mature-node silicon see volume and margin compression over time as capacity comes online.

Key catalysts and timing: near term (weeks–months) watch memory price signals and large OEM procurement cadence; medium term (6–18 months) watch capacity additions and HBM3 qualification timelines that will determine whether pricing holds; long term (2–5 years) the big policy/capability inflection is whether domestic tooling and advanced node competence materially narrow the performance gap versus TSMC/ASML, which remains the hardest and slowest vector to close.

The consensus leans bullish on pure revenue growth but underprices an inventory-led mean reversion and overweights policy-insulation as a sustainable moat. A pragmatic playbook monetizes the asymmetric outcomes: capture exposure to foundry/tool duopolies while hedging semiconductor OEMs with outsized China revenue sensitivity.

Key Decisions for Investors

• Pair trade — Long TSM (12–18 months) / Short NVDA (3–6 months): Buy TSM shares sized 2–4% of portfolio to capture foundry pricing upside and secular capacity tightness; offset with a short-biased option structure on NVDA (buy 3–6 month 30-delta puts and sell 3–6 month 10-delta puts, notional ~30–50% of the TSM position) to hedge Chinese GPU demand risk. Risk/reward: target +30–45% on TSM vs limited-cost hedge on NVDA with 3:1 asymmetric payoff if China demand disappoints.
• Long ASML (18–36 months) via LEAPS or tranche accumulation: Buy Jan-2028 calls or accumulate shares on pullbacks; thesis is structural pricing power from EUV/IP dominance as non-EUV alternatives remain immature. Risk: policy/licensing shift or capex pause could erase 20–30% in 3–6 months; position size 1–2% of portfolio.
• Tactical NVDA downside hedge (3–6 months): For short-term protection against China-share loss, implement a put spread (buy 30–60 day 25–30% OTM puts, sell deeper OTM puts to fund) sized to cover 30–50% of GPU exposure. Cost is modest; payoff if export frictions further curtail China revenue or if domestic substitutes materially accelerate.
• Selective overweight on data/research/info providers (MORN) — 6–12 months: Add a small position (0.5–1% portfolio) to capture recurring revenue tailwinds as clients increase spending on chip design and supply-chain due diligence. Downside limited but expect slower multiple expansion versus pure-capex beneficiaries.