Citi China Industrials AI Infrastructure and Robotics Tour: More Visible Earnings Upgrade Potential for AI-Infra

Overview

Citi held a China industrial AI infrastructure and robotics tour from June 9 to 12, 2026. The core conclusion is that AI-Infra remains the direction with the greatest certainty of growth and may bring more earnings upgrades. The report notes that Chinese suppliers previously had lower share in NVDA’s GB200 architecture than peers in Japan, Korea, and Taiwan, but are likely to increase share during the GB300 and Vera Rubin cycles through technology upgrades, response speed, and capacity expansion investment. Beyond AI infrastructure, demand related to embodied intelligence, industrial automation, the Apple equipment chain, and excavators is also more optimistic; real estate- and FAI-related verticals remain weak.

Core views

First, within the AI infrastructure chain, upstream CCL, AI-fabric, and PCB equipment have better supply-demand dynamics, and Citi prefers KBL > Han’s Laser > Shengyi. Second, in the embodied intelligence chain, early positioning should favor component suppliers over complete-machine OEMs, ranked Hengli > Leader Drive > UBTech. Third, the Apple supply chain and AI-related manufacturing are driving a recovery in machine tool and automation orders, benefiting Han’s Laser, OPT, Yiheda, and Inovance. Fourth, excavator demand is supported by mining, exports, and Hengli’s share gains, but overall construction machinery is still dragged down by real estate and FAI. Fifth, Envicool, Topband, and some construction machinery names face pressure from competition, costs, and insufficient demand.

Analysis framework

The report is based on Citi’s field research interviews with 13 Chinese industrial, AI infrastructure, and robotics-related companies. It compares orders, capacity, share, gross margin, customer mix, and valuation across supply chain links from the bottom up, and combines industry data such as NVDA architecture iterations, Apple capex cycles, and construction machinery utilization to form a stock preference ranking.

Methodology notes

• Supply chain researchAI infrastructure share migration analysis

  Changes in supplier share from GB200 to GB300 and Vera Rubin

  Through company interviews, the report assesses the opportunity for Chinese suppliers to gain share in NVDA’s next-generation GPU architectures, focusing on technology curve, response speed, capacity expansion commitments, and whether suppliers have entered the NVDA ecosystem.

• Industry chain comparisonUpstream components outperform downstream OEMs

  Early embodied intelligence allocation should favor core components

  The report believes that in the early stage of humanoid robots/physical AI, component companies such as Hengli have both core business support and robot optionality, while complete-machine OEMs may see valuations diluted by more newly listed companies.

• Valuation frameworkPE and PEG comparison

  KBL still has rerating room relative to Grace Fabric

  The report compares KBL and GF’s 2027E PE and three-year EPS CAGR, concluding that KBL pairs lower PE with high growth, and its AI-fabric business may drive further rerating.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Kingboard Laminates Holdings (1888.HK)

  Beneficiary of AI-fabric and upstream CCL; Citi’s top pick in the AI-Infra chain

  Strengths

  AI-fabric capacity expansion, stronger-than-expected profitability, 2027E capacity of about 85m meters potentially exceeding GF, and valuation still low relative to growth

  Weaknesses

  Business rerating depends on continued delivery of AI-fabric demand and capacity ramp-up

  Comparison

  Citi ranks KBL > Han’s Laser > Shengyi; KBL’s 2027E 16x PE is below GF’s 178x PE, while three-year EPS CAGR is close

  Risks

  AI-fabric demand below expectations, capacity rollout or product mix changes, changes in NVDA architecture material solutions

• Han’s Laser Technology (002008.SZ)

  Beneficiary of AI PCB equipment and the Apple equipment supply chain

  Strengths

  Benefits both from demand for AI PCB equipment and from the upward Apple capex cycle in 2026-2027

  Weaknesses

  Non-AI equipment demand may still be affected by volatility in the traditional manufacturing cycle

  Comparison

  Within the PCB equipment chain, Citi prefers Han’s Laser > Han’s CNC-H > Han’s CNC-A

  Risks

  Timing of Apple equipment demand, AI PCB capacity expansion pace, weaker-than-expected order delivery

• Shengyi Technology (600183.SS)

  CCL supplier benefiting from share gains in GB300 and Vera Rubin

  Strengths

  Rising AI-CCL mix, expanding total capacity, and preparations in Q-fabric, PTFE, and in-house solutions for Ultra Rubin

  Weaknesses

  PTFE on the PCB side is technically difficult, and some applications may be limited to high-frequency areas

  Comparison

  Citi ranks Shengyi third in its preferred AI-Infra chain, behind KBL and Han’s Laser

  Risks

  Uncertainty in NVDA’s final material solution, PCB-side yield and mass-production constraints, peer competition

• Han’s CNC (3200.HK / 301200.SZ)

  Supplier of PCB drilling equipment, benefiting from AI PCB and spillover SLP demand

  Strengths

  Demand for mechanical drilling equipment, orders for ultrafast laser equipment, and tight supply of CO2 laser equipment create growth opportunities; 2Q26E revenue may grow triple digits YoY

  Weaknesses

  Valuation differences between A-shares and H-shares are significant, and selection of non-AI orders may affect revenue mix

  Comparison

  Citi prefers HC-H over HC-A because HC-H trades at about a 50% discount to HC-A

  Risks

  AI customer capacity expansion pace, equipment delivery, valuation premium compression

• Hengli Hydraulic (601100.SS)

  Dual beneficiary of embodied intelligence components and excavator hydraulic parts

  Strengths

  Strong core hydraulic business, rising share in excavator pumps and valves, robot planetary roller screws provide optionality, and production lines are running at full load

  Weaknesses

  Robot revenue contribution remains low in the short term, below 1% in 2026E

  Comparison

  Citi ranks embodied intelligence names Hengli > Leader Drive > UBTech and prefers Hengli over downstream OEMs

  Risks

  Humanoid robot mass-production pace below expectations, weak traditional construction machinery demand, customer concentration

• Leader Drive (688017.SS)

  Beneficiary of core embodied intelligence components

  Strengths

  Benefits from the allocation logic that robot components outperform OEMs

  Weaknesses

  The report provides fewer company details than for Hengli

  Comparison

  Ranked after Hengli and before UBTech

  Risks

  Robot industrialization pace, valuation volatility, uncertainty in order delivery

• UBTECH Robotics (9880.HK)

  Humanoid robot OEM, with new product launches providing catalysts

  Strengths

  Launches of new products such as Walker S3, Walker C, Cruzr Y1, and bionic robots, with Thinker-WM model capabilities recognized

  Weaknesses

  OEM valuations may be diluted by listings of more robot companies such as Unitree and Zhiyuan

  Comparison

  Citi prefers components within the embodied intelligence chain, so UBTech ranks below Hengli and Leader Drive

  Risks

  Delivery targets, commercialization progress, intensified competition, and valuation dilution

• Precision Tsugami China (1651.HK)

  A name to watch for recovery in machine tools and factory automation

  Strengths

  The Apple supply chain and AI liquid cooling are driving strong order growth, with orders up 100%-150% YoY in April and May 2026

  Weaknesses

  Automotive demand is slowing

  Comparison

  Its order improvement is corroborated by Japan’s machine tool export data to China and growth in Inovance automation orders

  Risks

  Electronics demand cycle, Apple new product timing, sustainability of AI liquid-cooling orders

• Envicool (002837.SZ)

  Negative-view name in AI liquid cooling

  Strengths

  One of the few Chinese suppliers to obtain NVDA liquid-cooling certification

  Weaknesses

  Liquid-cooling technology barriers are low, competitors such as Shenling can serve US CSPs without NVDA certification, and the company has difficulty building its US team

  Comparison

  Citi’s earnings forecasts are more than 50% below market consensus, and it reiterates a Sell view

  Risks

  Intensified competition, earnings below consensus, overseas expansion below expectations

• Topband (002139.SZ)

  Negative-view name under rising material costs and customer price pressure

  Strengths

  ESS and humanoid robot brushless motor businesses may provide rerating catalysts

  Weaknesses

  Global customers in home appliances and power tools may continue to squeeze margins, while exposure to related new businesses is limited

  Comparison

  Citi reiterates a Sell view

  Risks

  Rising material costs, customer bargaining pressure, margin pressure

Key data

• Tour period2026-06-09 to 2026-06-12Citi China industrial AI-Infra and robotics tour
• Number of companies visited13 companiesThe companies drawing the most investor attention were AI-related names such as Han’s CNC, Shengyi Tech, DTECH, and an NVDA-certified PSU supplier
• Shengyi AI-CCL capacity mixFrom about 10% to about 15% by mid-2026, and about 20% by end-2026Driven by GB300 and Vera Rubin demand
• Shengyi total monthly capacityFrom 8-8.5m sheets to 9.6m sheets by mid-2026 and 10.4m sheets by end-2026Used to support AI-CCL expansion
• KBL earnings forecast revision2026-2028E raised by 7%-16%Driven by stronger-than-expected AI-fabric profitability; target price raised from HK$80 to HK$100
• KBL AI-fabric gross profit contributionFrom 6.9% in 2026E to about 17.4% in 2028ECiti estimate
• KBL AI-fabric capacityAbout 85m meters in 2027EHigher than GF’s estimated 2027E capacity of about 33m meters
• Han’s CNC short-term revenue growthMay achieve triple-digit YoY growth in 2Q26EDriven by demand for AI PCB equipment and spillover SLP demand
• Optimus output expectation10k-20k units in 2026, at least 100k units in 2027Based on Hengli and Rongtai’s judgment of customer timing
• Hengli robot revenue contributionBelow 1% in 2026E, about 4% in 2027EMainly from humanoid robot-related components such as planetary roller screws
• PTC order growth100%-150% YoY growth in April and May 2026Driven by orders from electronics, the Apple supply chain, and AI liquid cooling
• PTC electronics order mixFrom 10.1% in 2025 to 21.4% in 1Q26Mainly from new product demand in the Apple supply chain
• PTC AI liquid cooling order mixFrom 7.2% in 2025 to 18.8% in 1Q26Driven by orders from Chinese secondary suppliers
• Bozhon electronics business revenue guidanceFrom Rmb3.5bn in 2025 to about Rmb4.5bn in 2026About 90% is Apple-related business
• China construction machinery utilization52.3% in May 2026, down 7.2 percentage points YoYIndicates overall construction machinery demand remains insufficient
• Envicool earnings gapCiti’s 2026E and 2027E earnings forecasts are more than 50% below consensusThe company faces intensified liquid-cooling competition, low technical barriers, and difficulty building its US team

Impact & implications

The report’s investment implications lean toward structural opportunities: upstream AI infrastructure materials and equipment have stronger earnings elasticity due to shortages, share gains, and improving overseas customer mix; in the early stage of embodied intelligence, investors should focus first on core components; and the Apple equipment chain and AI manufacturing can drive better automation orders. At the same time, weakness in the real estate and FAI chain, low construction machinery utilization, intensified liquid-cooling competition, and rising material costs mean the industrial sector’s overall recovery cannot be linearly extrapolated, and investors need to distinguish between incremental demand driven by AI and exports and traditional segments weakened by domestic demand.

Risks

• Changes in NVDA GB300, Vera Rubin, or Ultra Rubin architecture plans may result in Chinese suppliers gaining less share than expected.
• AI-fabric, CCL, and PCB equipment capacity expansion may be too fast or demand may fall short of expectations, worsening supply-demand dynamics.
• PTFE PCB mass production is highly difficult, which may limit the broad application of PTFE CCL in AI server PCBs.
• The humanoid robot mass-production pace may be below supplier expectations, and the 2027 output target for Optimus is uncertain.
• Real estate and FAI demand may remain weak, dragging on construction machinery and traditional industrial automation demand.
• Rising material costs and margin pressure in domestic business may offset part of the gains from overseas growth.
• Intensified competition in AI liquid cooling and limited certification barriers may cause revenue and profit at companies such as Envicool to come in below market expectations.
• More robot OEM listings may divert valuations from existing names such as UBTech.

What to watch

• The actual share gains of Chinese suppliers in NVDA GB300, Vera Rubin, and Ultra Rubin.
• Capacity additions and price changes for AI-fabric, e-glass fabric, and CCL at KBL, Shengyi, and Grace Fabric.
• Order delivery by Han’s CNC and Han’s Laser in AI PCB equipment, SLP, and the Apple equipment chain.
• The progress of Optimus Gen 3 ramping to 1,000 units per week in 2H26 and whether output reaches at least 100k units in 2027.
• 2Q and 2H automation order and revenue delivery at PTC, Bozhon, and Inovance.
• Whether China’s tower crane rental rate, tower crane utilization rate, and overall construction machinery utilization recover from low levels.
• Envicool’s progress in building its US team, its NVDA liquid-cooling business, and share changes versus competitors such as Shenling.
• The order pull from Apple’s 2027 new product cycle on the automation equipment chain.