WAIC 2026 field research strengthened UBS's confidence in China's AI technology supply chain
AI summary card
WAIC 2026 field research strengthened UBS's confidence in China's AI technology supply chain
UBS believes that segments including China's LLMs, AI computing power, SuperPod, networking interconnects, advanced packaging, and server ODM will have clearer visibility for rapid growth in 2026 and beyond.
- WAIC 2026 attracted more than 1,100 exhibitors and over 400,000 attendees, while more than 40 institutional investors participated in the UBS tech tour.
- China's AI ecosystem is accelerating. The newly established WAICO involves 29 countries and signed AI governance-related initiatives; China also signed about US$3 billion in new contracts with overseas companies.
- Domestic AI infrastructure has made notable progress, including Huawei Atlas 950 SuperPoD, Biren computing power, Xizhi optical interconnect, and multiple SuperPoD solutions.
- Supply chain companies reported positive end-demand for AI infrastructure, spanning GPUs, memory interface chips, advanced packaging, AI servers, switches, and AI power supplies.
- UBS names Zhipu as its top pick in China's AI LLM space and rates MiniMax Buy; in the AI-related semiconductor supply chain it prefers NAURA, JCET, and Han's Laser.
Report interpretation
Overview
This report summarizes the key findings from UBS's WAIC 2026 tech tour. UBS held C-level or IR meetings with 12 companies, covering LLMs, GPUs, networking, analog chips, advanced packaging, ODMs, and semiconductor equipment, and also participated in booth visits and thematic activities. The overall conclusion is that China's AI ecosystem and technology supply chain capabilities continue to strengthen, with positive feedback on AI infrastructure demand and rapid progress in domestic computing power, SuperPod, networking interconnects, and edge AI applications.
Core views
UBS's core view is that visibility for growth across China's AI industry chain will further improve in 2026 and beyond. On the LLM side, Zhipu and MiniMax are seeing rapid ARR growth, driven mainly by API demand, token usage, and inference optimization. On the hardware side, demand is improving for domestic AI accelerator cards, SuperPod, optical interconnects, advanced packaging, AI servers, and AI power supplies. On the supply chain side, companies such as JCET, Montage, Huaqin, and Southchip reported stronger AI-related demand, although some segments are still constrained by upstream packaging materials, computing power rental costs, or weak consumer electronics demand.
Analysis framework
The report is based on on-site observations at WAIC 2026, booth visits, thematic activities, and communications with management or IR teams at companies including MiniMax, Zhipu, Phancy, Xunce, Biren, Montage, Xizhi, Southchip, Dosilicon, JCET, Huaqin, and PowerTech, summarizing industry chain demand, product roadmaps, commercialization progress, supply constraints, and valuation methods.
Methodology notes
Extracting industry chain trends through on-site WAIC visits and company meetings
UBS combines new products displayed at the exhibition, management communications, investor participation, and supply chain feedback to assess demand strength in China's AI ecosystem, computing infrastructure, and semiconductor supply chain.
Price-to-earnings valuation
The report states that the PE method is used to value NAURA, Luxshare, Lens Tech, AAC Tech, and Han's Laser.
Price-to-book valuation
The report states that the PB method is used to value JCET.
Price-to-annual recurring revenue ratio
The report states that Price/ARR is used as the valuation method for MiniMax.
Price-to-sales and sum-of-the-parts valuation
The report states that Zhipu is valued using P/S, with SOTP analysis used for cross-validation.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- Zhipu (2513.HK)Top pick in China's AI LLM space
- Strengths
- ARR reached US$1bn and achieved internal targets ahead of schedule; API demand and higher pricing support growth, and the H-share placement is expected to support computing power expansion.
- Weaknesses
- Computing power remains a key constraint, and uncertainty around computing rental costs may affect the pace of margin improvement.
- Comparison
- UBS lists it as its top pick in China's AI LLM space and believes its fundamentals are positive and valuation attractive.
- Risks
- Tight domestic computing supply, rising rental prices, pressure from model capability iteration, and intensifying competition.
- MiniMax (0100.HK)LLM company rated Buy
- Strengths
- ARR is ramping rapidly, with a rising revenue contribution from API and Token Plan, while inference engineering optimization and cluster operations help maintain healthy gross margins.
- Weaknesses
- Larger model parameter scales bring computing and cost pressure, and commercialization targets still depend on sustained demand growth.
- Comparison
- It is one of UBS's key covered names in China's LLM space alongside Zhipu, but the report names Zhipu as the preferred pick.
- Risks
- Computing costs, model competition, changes in API pricing, and fluctuations in enterprise customer demand.
- JCET (600584.SS)Beneficiary of advanced packaging
- Strengths
- AI-related business, wafer-level packaging, and bumping are running at full capacity; the Rmb7.8bn Shanghai advanced packaging project is focused on 2.5D, and AI PMIC assembly margins are above the company average.
- Weaknesses
- The ramp-up cycle for new capacity is relatively long, and part of the growth depends on domestic AI accelerators and overseas spillover demand.
- Comparison
- UBS prefers JCET within the AI-related China technology semiconductor supply chain and values it using PB.
- Risks
- Capacity build-out execution, packaging material supply, customer demand fluctuations, and pricing cycles.
- Montage (6809.HK / 688008.SH)Supply chain company related to memory interfaces, CXL, and interconnects
- Strengths
- Contribution from DDR5 Gen3/4 is increasing, Gen5 is expected to ship in H226, and CXL, MRDIMM, PCIe retimer, and longer-term Ethernet/optical interconnects have growth potential.
- Weaknesses
- Volume ramp for some new products depends on next-generation CPU platforms and ecosystem maturity.
- Comparison
- Benefiting from AI server memory bandwidth and interconnect demand, but the pace of commercialization depends on platform iteration.
- Risks
- Tight upstream materials such as BT substrates and e-glass cloth, CPU platform delays, and customer adoption progressing more slowly than expected.
- Huaqin (3296.HK / 603296.SS)Server ODM and SuperPod supply chain company
- Strengths
- Data center revenue is targeted at Rmb70bn in 2026, up 50% year-on-year, driven by switches, AI servers, and SuperPod deployments; profit growth and product mix improvement are becoming clearer.
- Weaknesses
- Traditional smartphone ODM volumes are under pressure, and the transformation requires sustained scaling of new businesses.
- Comparison
- It is expanding from smartphone ODM into mobile, PC, and data center hardware platforms, with a more diversified growth mix.
- Risks
- Data center order volatility, supply chain management, customer concentration, and weak traditional consumer electronics demand.
- Southchip (688484.SS)AI power and analog chip supply chain company
- Strengths
- Focused on applications such as AI server power, memory power, HBM, MRDIMM, SOCAMM, LPCAMM2, and enterprise SSDs, with accumulated capabilities in charge pumps, IP, and proprietary processes.
- Weaknesses
- Consumer electronics demand is weak in the short term, and as a later entrant into the AI power market it still needs to prove its ability to scale.
- Comparison
- Expanding from consumer electronics into AI, automotive, and industrial applications.
- Risks
- Rising wafer costs, tight capacity, and slower-than-expected ramp-up of AI server power products.
- Biren (6082.HK)Domestic GPGPU and AI accelerator company
- Strengths
- Benefiting from China's AI GPU demand and localization trend; the BR20X series is expected to launch in H226 and ramp by the end of 2026, with a higher degree of supply chain localization.
- Weaknesses
- Product performance, ecosystem compatibility, and capabilities in advanced process nodes/packaging still need continuous improvement.
- Comparison
- Positioned in the high-end GPGPU segment for the China market, targeting leading global peers.
- Risks
- Maturity of the domestic supply chain, customer onboarding pace, software ecosystem, and availability of advanced packaging.
- Xizhi Lightelligence (1879.HK)Optical interconnect and optical computing company
- Strengths
- Launched a 51.2T CPO switch prototype; optical interconnect makes up the main portion of revenue; NPO adoption is accelerating, and dOCS may become a lower-cost alternative for some AI workloads.
- Weaknesses
- Optical computing commercialization is still at an early stage and broader application scenarios need to be validated.
- Comparison
- It has differentiated positioning in domestic optical interconnect and SuperPoD interconnect solutions.
- Risks
- Changes in CSP demand, technology path selection, mass-production yield, and ecosystem adoption pace.
Key data
- Number of WAIC 2026 exhibitorsMore than 1,100The report states that WAIC 2026 had more than 1,100 exhibitors.
- WAIC 2026 attendanceMore than 400,000The report states that there were more than 400k attendees.
- Institutional investor participation in UBS tech tourMore than 40Reflects strong institutional investor interest in China's AI technology industry chain.
- Countries covered by WAICO29 countriesChina announced the establishment of the World Artificial Intelligence Cooperation Organization.
- New contracts signed by China with overseas companiesAbout US$3 billionThe report describes this as cUS$3bn.
- Scale of Huawei Atlas 950 SuperPoD1,024 chips / 1 EFLOPS FP8 / 2 EFLOPS FP4 / 256TB unified memory addressingUsed for trillion-parameter model training and inference, emphasizing low-latency and high-bandwidth interconnects.
- Deployment volume of Huawei Ascend 384 SuperPoDMore than 750 unitsCovers vertical industries including internet, telecom, finance, education, healthcare, transportation, and manufacturing.
- Zhipu ARRReached US$1bn by JulyManagement said it had already reached its internal end-2026 target ahead of schedule.
- Zhipu H-share placementHK$31bnManagement believes this can support further computing power expansion.
- MiniMax ARR trajectoryAround US$100m by end-2025, around US$150m disclosed at the Q1 earnings call, targeting US$1bn by end-2026Driven by rapid growth in API and Token Plan.
- Huaqin data center revenue guidanceRmb70bn in 2026, up 50% year-on-yearDriven by switches, AI servers, and SuperPod deployments.
- JCET Shanghai advanced packaging investmentRmb7.8bnFocused on 2.5D packaging, expected to complete in H228 and begin production in 2029.
Impact & implications
The report has a positive investment implication for China's AI industry chain: LLM commercialization, domestic computing power substitution, SuperPod deployment, optical interconnects, advanced packaging, AI servers, and AI power supplies could become key growth drivers after 2026. In terms of stock selection, UBS prefers companies with improving fundamentals, attractive valuations, or critical positions in the industry chain, such as Zhipu, MiniMax, NAURA, JCET, and Han's Laser, while also warning that technology stocks remain highly exposed to technological change, competition, macro cycles, and valuation uncertainty.
Risks
- The technology sector changes rapidly, competition is intensifying, and it is highly exposed to macro cycles.
- Technology companies' operating models are volatile, making financial forecasting difficult.
- Traditional and non-traditional valuation metrics have limited explanatory power for technology stock prices, making valuation challenging.
- Tight domestic AI computing supply and rising computing rental prices may constrain margin improvement for LLMs.
- Some supply chain segments are constrained by upstream supply such as BT substrates, e-glass cloth, and wafer capacity.
- Weak consumer electronics demand may drag on analog chips, smartphone ODMs, and related hardware supply chains.
- The commercialization pace of domestic AI accelerators, SuperPod, optical interconnects, and advanced packaging still depends on customer adoption, ecosystem maturity, and supply chain capabilities.
What to watch
- Progress in ARR delivery, API demand, pricing strategy, and inference cost optimization at Zhipu and MiniMax.
- Domestic AI computing supply, computing rental prices, and adoption rates for domestic AI accelerator cards.
- Deployment progress of Huawei Atlas 950 SuperPoD and other SuperPod solutions.
- Shipments of Montage DDR5 Gen5, MRDIMM, CXL, and PCIe retimer, as well as the maturity of the platform ecosystem.
- Build-out of JCET's new 2.5D advanced packaging capacity, demand for AI PMIC assembly, and progress of the Shanghai project.
- Delivery of Huaqin's data center revenue, switches, AI servers, and SuperPod orders.
- Ramp-up of Southchip's AI server power and memory power products starting from 2027.
- Customer onboarding progress for Xizhi's CPO, NPO, dOCS, and optical computing solutions.
- The timing of volume ramp-up in PowerTECH's high-end SoC testing, SiC testing, and memory testing businesses.