Citi Survey: China's AI Supply Chain Poised to Gain Greater Share in Nvidia's GB300 and Vera Rubin
Taylor Wilson
After visiting 13 companies, Citi concludes that Chinese AI infrastructure suppliers will gain significantly higher procurement share in Nvidia's GB300 and Vera Rubin platforms — reversing a pattern where Japan, Korea, and Taiwan dominated the GB200 cycle.
Where do Chinese suppliers actually stand in Nvidia's supply chain?
Citi's field study found that Chinese suppliers hold a notably smaller procurement share in the current GB200 architecture than peers in Japan, Korea, and Taiwan.
As GB300 and Vera Rubin ramp, that gap is set to close. This means → the shift is driven by faster technical response and capacity commitment, not price competition alone.
Citi's AI infrastructure preference order: Kingboard Laminates > Han's Laser > Shengyi Technology. Citi has revised earnings forecasts upward for covered names accordingly.
Why are CCL and glass fabric the tightest, clearest bottleneck?
Shengyi Technology began shipping AI CCL — copper-clad laminate, the core substrate in PCBs — in late 2024. AI CCL's share of total monthly capacity is projected to rise from about 10% in 2025 to roughly 20% by end-2026, while total monthly capacity expands from 8.5 million to 10.4 million sheets.
In AI glass fabric — a specialty woven material inside CCL that determines signal quality — Honghe Technology plans to scale from about 5 million meters in 2025 to at least 33 million meters by 2027. Yet Citi estimates Kingboard Laminates will reach roughly 85 million meters by 2027, making it the segment's largest supplier.
In plain terms = electronic-grade glass fabric can't expand fast because the looms themselves are scarce. Citi estimates Kingboard's average selling price is up 60%–95% year-to-date. Citi has raised Kingboard's 2026–2028 earnings forecasts by 7%–16% and lifted the target price from HK$80 to HK$100.
Who benefits on the PCB equipment side?
Mitsubishi Electric's CO2 laser drilling equipment now has a 12-month lead time. That bottleneck opens a substitution window for Han's Laser (Dazu CNC).
Nvidia's proposed midplane (cable-free design) and backplane (scale-up network board) architectures both use more high-layer-count boards rather than HDI — high-density interconnect boards. This means → incremental demand for mechanical drilling equipment, directly benefiting Dazu CNC.
Citi expects Dazu CNC's Q2 2026 revenue could grow triple-digits year-on-year, roughly matching Taiwan peer Ta Liang's approximately 120% YoY pace in the same period.
In embodied AI, why do components beat finished machines?
Hengli Hydraulic and Rongtai have been shipping to a leading U.S. robotics company since last year. Both expect Optimus Gen 3 output of about 10,000–20,000 units this year, potentially scaling 10× to over 100,000 units next year.
Citi projects Hengli's robotics revenue — from planetary roller screws, precision parts that convert rotary motion into linear motion — will rise from under 1% of sales in 2026 to about 4% in 2027.
Citi's embodied-AI preference order: Hengli Hydraulic > Leaderdrive > UBTECH. This reflects a core judgment: components carry higher certainty than finished-machine OEMs, because whichever OEM wins still needs the same critical parts.
Why does Citi rate Envicool and construction-machinery names "sell"?
Envicool: Citi argues AI liquid cooling is not a high-barrier technology. It requires local design and installation, which limits revenue scalability. Citi's earnings forecast is more than 50% below consensus.
The sharper signal: Shenling Environment has already begun supplying Nvidia liquid-cooling work for a U.S. cloud provider — without Nvidia liquid-cooling certification. In plain terms = if a supplier can win orders without even being certified, the moat is shallower than the market assumes.
In construction machinery, China's overall equipment utilization in May 2026 was just 52.3%, down 7.2 percentage points year-on-year; the tower-crane rental price index remains near historic lows. This means → shipment data may look solid, but underlying demand is still weak. Citi maintains "sell" on CRRC, Dingli, Anhui Heli, and others.
What is the single variable that decides everything?
Whether Chinese AI infrastructure suppliers can convert rising share into sustained earnings improvement ultimately depends on two things: the mass-production pace of GB300 and Vera Rubin, and the stability of Nvidia's architecture roadmap.
In plain terms = if Nvidia's next-generation platforms ramp on schedule, the share gains translate into real profit. If production slips or the architecture changes significantly, Citi's upgraded forecasts could unwind.
This is the single most important variable to track — it simultaneously governs the realization timeline for CCL, glass fabric, PCB equipment, and components.
Content is for reference only, not financial advice.