Macquarie says now is the 'best time' to buy Chinese AI chip stocks. This one's its favorite

Macquarie initiated bullish coverage of China’s AI chip players, arguing the “best time to invest” has arrived on AI/LLM momentum and PRC support tied to Nvidia GPU import restrictions. The firm’s top picks are Shanghai-listed Cambricon (favorite) with a 2,060 yuan target, implying >50% upside from Friday’s close, and Hong Kong-listed Biren Tech with a 140 HK$ target, more than double Friday’s close. It rates Hygon underperform due to expected market-share pressure and limited upside from agentic AI development, with Huawei’s Ascend highlighted as the leader in shipments.

Analysis

This is less about a near-term technology leap and more about procurement rerouting under policy pressure. When a buyer base is pushed to localize, the first money goes to the vendors that can ship acceptable performance with the least integration friction; that tends to lift multiples before it meaningfully lifts earnings. The structural losers are NVDA and AMD on China mix, but the more important second-order effect is ecosystem entrenchment: once domestic cloud and LLM teams standardize around local silicon, interconnect, packaging, compilers, and cluster orchestration become sticky, reducing future pull-through for imported accelerators even if licenses remain open.

Over the next 1-3 months, the cleanest relative trade is local AI-chip leaders versus U.S. GPU suppliers, because the market will price policy optionality faster than it can verify unit economics. BABA and BIDU are interesting only as indirect beneficiaries: if their internal chip efforts lower inference and training costs, cloud margins can improve, but the upside is more about strategic self-sufficiency than a direct semiconductor earnings story. The risk is that the move is too front-loaded; if domestic chips remain confined to smaller-scale inference or government clusters, the revenue upside for the chip names will lag the narrative.

The contrarian miss is assuming AI demand automatically translates into broad-based domestic chip share gains. Frontier training still depends on power efficiency, software compatibility, and supply-chain access that China cannot fully replace yet, so the best outcome may be incremental share in less demanding workloads rather than a wholesale displacement of foreign silicon. Falsifiers: any easing in U.S. export restrictions, delayed domestic product launches, or cloud guidance showing continued reliance on foreign accelerators to meet performance targets.

Key Decisions for Investors

• Long MQBKY / short NVDA as a 1-3 month relative-value expression: use on pullbacks, target 20-30% spread outperformance, and cut if China AI-capex commentary shows no real shipment conversion.
• Long BABA and BIDU as smaller, indirect beneficiaries of domestic AI self-supply over 3-6 months; treat this as a margin-improvement optionality trade, not a core semiconductor call.
• Avoid chasing AMD rallies tied to China headlines; use any strength to fade into the next earnings window unless management can show China revenue stabilization and no further mix deterioration.
• Set a watch item on MQBKY/China domestic chip shipment data and cloud capex commentary; if evidence of design wins does not appear within 1-2 quarters, exit the thematic long because the policy trade is likely overdone.