Citi views China's Rmb2tn AI investment plan as positive for the domestic AI supply chain

Overview

This report is a Citi quick take on Greater China semiconductors, with the key event being Bloomberg's report that China plans to launch a new Rmb2tn AI investment plan. Citi believes the plan will directly strengthen demand visibility for the domestic AI computing supply chain through data center and network construction, as well as a procurement requirement of at least 80% domestic AI accelerators.

Core views

Citi's core view is that if this policy is implemented, it will benefit localization across China's AI supply chain, especially foundries, packaging and testing, semiconductor equipment, and domestic AI accelerators. The report believes the target of more than 80% localization of AI chips is achievable because after Nvidia H200 imports largely stalled, domestic AI chip makers have already taken a dominant position in China's AI accelerator market this year.

Analysis framework

The report uses an event-driven and supply-chain mapping approach, breaking down a macro/industrial policy investment plan into data center operations, AI chip procurement, wafer manufacturing, packaging and testing, and equipment demand, and on this basis judges which listed or unrated companies may benefit.

Methodology notes

• Industry chain analysisPolicy catalyst - AI localization supply chain mapping

  Maps the Rmb2tn AI infrastructure investment and the target of procuring more than 80% domestic AI accelerators to foundries, packaging and testing, equipment, and AI chip suppliers.

  By identifying the direct beneficiary chain from data center construction and domestic AI chip procurement, Citi judges that demand visibility and the domestic substitution logic for China's AI supply chain are strengthening.

• Rating frameworkCiti Research equity rating definitions

  Buy(1), Neutral(2), and Sell(3) are based on expected total return and risk over the next 12 months.

  The appendix explains that Buy usually corresponds to expected total return of at least 15%, with a higher threshold for higher-risk stocks; Sell corresponds to negative expected total return; covered stocks not assigned Buy or Sell are usually Neutral.

Asset mapping & comparison

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

• SMIC (0981.HK); Hua Hong Grace Semiconductor (1347.HK)

  Potential beneficiaries in the domestic foundry segment

  Strengths

  Rising demand for domestic AI accelerators could bring more local chip manufacturing orders, and the policy roadmap is clearer.

  Weaknesses

  The report does not provide quantified estimates of new orders, capacity utilization, or earnings elasticity.

  Comparison

  Compared with the backdrop of constrained foreign chip supply, domestic foundries are more directly tied to the local AI chip supply chain.

  Risks

  Advanced process capability, capacity bottlenecks, customer concentration, and the pace of policy execution may affect actual benefits.

• JCET Group (600584.SS); TongFu Microelectronics (002156.SZ)

  Potential beneficiaries in the packaging and testing segment

  Strengths

  Volume growth in AI accelerators usually drives demand for advanced packaging and testing, and domestic supply chain development may increase local packaging and testing orders.

  Weaknesses

  The report does not disclose specific packaging technology routes, customer mix, or order visibility.

  Comparison

  Compared with companies benefiting only from chip design, packaging and testing firms benefit from capacity expansion by more domestic AI chip suppliers.

  Risks

  Advanced packaging capability, price competition, capex, and customer qualification cycles are the main uncertainties.

• ASMPT (0522.HK); Shanghai Vital Deeptech (600641.SS)

  Potential beneficiaries among semiconductor equipment suppliers

  Strengths

  Data center expansion and domestic AI chip capacity build-out may drive equipment procurement demand, and the report includes them in the positive beneficiary chain.

  Weaknesses

  Vital Deeptech is rated Sell(3) in the company list, indicating that its fundamental rating is not fully equivalent to the event-driven beneficiary direction.

  Comparison

  The equipment segment is more exposed to the capex cycle, and its upside may lag chip procurement and manufacturing orders.

  Risks

  Equipment order confirmation, progress in domestic substitution, downstream expansion pace, and valuation volatility may affect investment returns.

• Cambricon Technologies (688256.SS); MetaX Integrated Circuits (688802.SS); Moore Threads Technology (688795.SS); Shanghai Biren Technology (6082.HK); Shanghai Iluvatar CoreX Semiconductor (9903.HK)

  Domestic AI accelerator suppliers or potential suppliers

  Strengths

  The target of more than 80% domestic AI accelerators directly increases procurement demand for local chips, and state-funded data centers may procure from a broader range of suppliers.

  Weaknesses

  Many companies are Not Rated, and the report does not provide earnings forecasts, market share, or product performance details.

  Comparison

  Compared with constrained Nvidia H200 imports, the substitution opportunity for domestic AI chips in the Chinese market is clearer.

  Risks

  There is uncertainty around chip performance, software ecosystem, mass-production capability, customer qualification, and policy procurement allocation.

• Baidu Inc (9888.HK); Baidu.com (BIDU.O)

  AI ecosystem and computing-demand related names

  Strengths

  Expansion of domestic AI infrastructure helps improve the availability of the AI application and cloud/computing ecosystem.

  Weaknesses

  The report does not present a core investment thesis for Baidu, and 9888.HK is Not Rated.

  Comparison

  Compared with the upstream semiconductor chain, Baidu is more of a beneficiary through AI applications and platform ecosystem exposure.

  Risks

  AI application monetization, cloud competition, capex efficiency, and the regulatory environment may affect the degree of benefit.

• NVIDIA Corp (NVDA.O)

  Foreign AI accelerator reference name and potential replacement target

  Strengths

  NVIDIA still has globally leading AI chip products and ecosystem advantages.

  Weaknesses

  The report says imports of Nvidia H200 have largely stalled, reducing its accessibility in related incremental projects in China.

  Comparison

  Domestic Chinese AI accelerator companies are relatively more direct beneficiaries in the context of this round of policy.

  Risks

  Changes in export controls, supply of substitute products, migration in customer demand, and geopolitical policy will all affect opportunities in the China market.

Key data

• Investment plan sizeRmb2tn, approximately US$295bnThe scale of China's AI data center and network investment plan reported by Bloomberg on June 9.
• Investment periodNext 5 yearsThe funds will support data center and network construction for the domestic AI industry.
• Domestic AI accelerator procurement ratioAt least 80%The report says data centers operated by Chinese telecom operators will use at least 80% domestic AI accelerators.
• Citi's view on localization feasibilityThe target of more than 80% should be achievableThe reason is that after Nvidia H200 imports were largely blocked, domestic AI chip makers have already taken a dominant share of China's AI accelerator market this year.
• Beneficiary foundriesSMIC; Hua HongThe report explicitly lists both as beneficiaries.
• Beneficiary packaging and testing companiesJCET; TongFuThe report explicitly lists both as beneficiaries.
• Beneficiary equipment suppliersASMPT; Vital DeeptechThe report explicitly lists both as beneficiaries.
• Company price and rating snapshotSMIC 0981.HK HK$75.0 Buy(1); Hua Hong 1347.HK HK$140.1 Buy(1); JCET 600584.SS Rmb75.29 Buy(1); TongFu 002156.SZ Rmb64.43 Buy(1); ASMPT 0522.HK HK$185.3 Buy(1); Vital Deeptech 600641.SS Rmb27.03 Sell(3)Prices are as of 2026-06-09 or 2026-06-08; refer to the company list in the report for specifics.

Impact & implications

If this investment plan moves forward as reported, China's AI infrastructure capex will provide a clearer demand roadmap for domestic AI chips and the upstream semiconductor manufacturing chain. For investors, the beneficiary logic expands from standalone AI chips to manufacturing, packaging and testing, equipment, and the data center procurement ecosystem, though the extent of realization depends on the pace of official execution, tender allocation, domestic chip performance, and capacity.

Risks

• Bloomberg's reported investment plan still requires verification of official details, budget arrangements, and execution pace.
• Although Citi believes the target of more than 80% domestic AI accelerators is achievable, it still depends on domestic chip performance, software ecosystem, capacity, and customer qualification.
• Procurement by state-funded data centers may cover a broader range of suppliers, creating uncertainty in benefit allocation and profit realization.
• Related stocks may have already priced in policy expectations, with elevated short-term valuation and share-price volatility risk.
• External export controls, import restrictions, cross-border regulation, and geopolitical changes may alter the supply chain landscape.
• OCR information from the report's charts is limited, and Figure 1 and Figure 2 lack verifiable detailed figures.

What to watch

• Whether China's AI investment plan receives official confirmation, as well as funding sources, project batches, and tender timing.
• Whether the actual share of domestic AI accelerators in telecom operator data center procurement reaches or exceeds 80%.
• Winning share and delivery capability of domestic AI chip makers in large data center projects.
• Follow-up changes in orders, capacity utilization, and capex at upstream companies such as SMIC, Hua Hong, JCET, TongFu, and ASMPT.
• Whether export and import restrictions on Nvidia H200 and other overseas AI chips to the Chinese market change.
• Whether smaller domestic AI chip suppliers can obtain meaningful orders from the multi-supplier procurement of state-funded data centers.