Morgan Stanley Raises China AI Chip Market Forecast to $91 Billion, Lifts Cambricon Target Price to 1,528 Yuan
Taylor Wilson
Morgan Stanley raised its 2030 China AI chip market forecast by 36% to $91 billion, triggered by U.S. moves to close an export loophole for Nvidia's Blackwell chips — tighter controls, paradoxically, widen the addressable market for domestic GPUs.
Why the sudden 36% upgrade?
The U.S. Commerce Department is closing a loophole that let Nvidia Blackwell chips reach Chinese firms via their overseas subsidiaries. This means → the channel for Chinese cloud vendors to buy Nvidia is narrowing, and the incremental market for domestic GPUs opens by default.
After the revision, China's AI chip market compounds at 23% annually from 2025 to 2030, reaching $91 billion — up from the prior $67 billion estimate.
In plain terms = the stricter U.S. controls get, the more bullish Morgan Stanley becomes on China's own chip business — demand doesn't vanish, it just changes supplier.
Where does the $91 billion come from?
Morgan Stanley identifies four new demand pools: ① capex from Chinese cloud vendors building offshore data centers with domestic GPUs; ② ByteDance plans to ramp AI spending sharply in 2026–2027, with 2027 capex conservatively estimated at ~$80 billion; ③ Kingsoft Cloud added to the model, with 2026 capex projected above ¥15–20 billion; ④ government and SOE market space raised from $7 billion to $9 billion.
Behind pool ④ sits a headline number: Bloomberg reported on June 9 that China plans to spend ¥2 trillion on data centers over five years.
This reflects a broadening funding base — AI compute is no longer funded by internet giants alone; government and state-owned enterprises are becoming major buyers.
When do domestic chips actually ship at scale?
Morgan Stanley flags 2026 as the pivotal year — the window for domestic chipmakers to enter cloud vendors' procurement systems.
On-the-ground research shows cloud vendors keep expanding capacity, yet the compute gap persists; at the same time, domestic chip qualification is accelerating noticeably.
This means → competition is shifting from "can the chip work at all" to ecosystem maturity, software optimization, and large-scale cluster deployment.
Who gets access to advanced foundry capacity?
Access to advanced-node foundry capacity is the core competitive dividing line. Firms that pass the U.S. BIS CCATS review can use TSMC lines, gaining a clear edge in cost and power efficiency — Biren Technology's 7nm/6nm products fall into this category.
Morgan Stanley expects China to develop a more stable domestic supply chain beyond SMIC's southern fab by 2027–2028.
Equipment import data offers a side view: April 2026 China semiconductor equipment imports hit $2.4 billion, down 6% year-on-year; imports from the U.S. fell 43%, while only Singapore posted a gain — up 43%. In plain terms = Western equipment is flowing in at a shrinking rate; domestic substitution is filling the gap.
Cambricon and Iluvatar — how did targets move?
Cambricon (寒武纪) stays Overweight. Revenue forecasts for 2026–2028 rise 6%, 10%, 9% respectively; target price lifts to ¥1,528. SMIC's supply chain for Cambricon is stabilizing, supporting volume shipments of the MLU580 and MLU690 high-end compute chips in H2 2026. Bull-case valuation: ¥2,887; bear case: ¥770.
Iluvatar CoreX (天数智芯) also Overweight. Revenue forecasts up 6%, 10%, 8%; target raised to HK$688. Its new Tiangai 300 chip carries a far higher ASP than the prior Tiangai 150; the 2027 gross-margin assumption rises from 51.3% to 52.4%. Bull case: ¥1,261; bear case: ¥344.
This reflects a selective bet — Morgan Stanley is not backing the entire sector, but the few companies that have locked in advanced-node capacity and are about to ramp.
Can this forecast be trusted — what is the key checkpoint?
Morgan Stanley itself names the test: whether domestic AI chips enter cloud vendors' procurement systems at scale in 2026.
If Cambricon, Iluvatar, and peers deliver on shipment volumes in H2 2026, the $91 billion forecast gains its first anchor; if adoption stalls, the number needs a haircut.
In plain terms = the report gives a big number but also hands you a ruler — check back at the end of 2026, and the answer will be clear.
Content is for reference only, not financial advice.