China’s SMIC says foreign clients shifting orders back to China

SMIC said overseas orders are rising as the AI boom and tight foreign foundry capacity push customers to shift production into China. The company added 9,000 12-inch equivalent wafers of capacity in Q1, ran at 93% utilisation, and expects depreciation expense to rise about 30% from last year. However, advanced 7nm expansion remains constrained by U.S. export controls, keeping the overall read modestly positive but mixed.

Analysis

The first-order read is capacity relief for China-based foundries, but the more important second-order effect is margin compression for non-China legacy nodes globally. When AI and HBM-related products consume leading-edge wafer starts, older-node supply gets rationed, which should support pricing and utilization for SMIC’s mature-node mix even if advanced-node constraints remain binding. The real beneficiary set is broader than SMIC: local equipment, materials, and EDA-adjacent suppliers tied to capacity expansion should see a longer demand runway as overseas customers de-risk away from crowded foreign fabs.

For Nvidia, the near-term impact is not direct incremental China demand; it is a cleaner supply-chain backdrop for global AI buildout. If foundry bottlenecks persist, hyperscalers will keep prioritizing AI capex, which extends the replacement cycle for compute and memory but also raises the odds of periodic inventory whiplash in less strategic end-markets like smartphones and industrials. That creates a favorable backdrop for high-end compute names while pressuring downstream OEMs that rely on stable access to mature-node capacity.

The key risk is policy, not demand: any tightening of export controls or pressure on customers using China capacity could slow the migration of overseas orders into the country. Over 3-12 months, the more relevant catalyst is whether new Chinese fabs ramp on schedule; if they do, SMIC can re-rate as a capacity monetization story, but depreciation will cap operating leverage until utilization stays elevated for several quarters. Consensus may be underestimating how durable this is: the AI boom is not just a winner for fabs serving AI chips, it is also a structural tailwind for every non-AI customer forced to pay up for scarce legacy capacity elsewhere.

Key Decisions for Investors

• Maintain a tactical long NVDA position into AI-capex revisions over the next 1-3 months; the setup favors multiple support from supply bottlenecks, but trim on any sign of hyperscaler digestion or weaker order visibility.
• Pair trade: long SMIC-related China capacity beneficiaries / short legacy-node exposed global semiconductor suppliers for 3-6 months, aiming to capture pricing scarcity while hedging broad chip beta.
• Add optionality via NVDA call spreads or diagonal spreads into the next earnings cycle; the thesis is not a huge demand upside, but a higher probability of sustained supply discipline and positive guidance revisions.
• Avoid chasing broad smartphone/consumer semiconductor names for 1-2 quarters; they are the most likely losers if mature-node capacity stays tight and order timing remains volatile.
• If SMIC rallies on capacity scarcity, consider fading after a sharp move unless utilization stays above 90% for multiple quarters; depreciation is a near-term drag that can limit earnings leverage.