Citi Asia Roadshow: AI Chain Divergence Intensifies, Optical Module Valuations Fully Priced While PCB Leaders Have the Edge
nashnova research
Citi's Sept 28 Asia roadshow feedback shows institutional investors still see AI as the dominant theme, but 'sector growth' alone no longer justifies adding positions — the next buy case demands order-by-order, earnings-by-earnings, certification-by-certification proof.
Optical modules — why are investors reluctant to chase?
Investors broadly believe Innolight's 2027 earnings expectations are already priced in, preferring to hold rather than add. This means → the market isn't bearish on optical modules — it simply thinks "the good news is in the stock."
CPO — co-packaged optics, integrating optical modules directly into chip packaging — and NPO — near-packaged optics, placing modules close to chips without full co-packaging — have not accelerated materially, yielding no new deliverable upside yet.
Eoptolink is seen as a key name for global AI optical demand, but investors are more sensitive to valuation and potential U.S. regulatory risk. O-Net Technologies, focused on passive components — optical parts with no embedded chips — is viewed as a more defensive pick with lower direct exposure to restrictions.
PCB — equipment lead times stretch to two years. Who can catch the orders?
Lead times for core equipment — exposure, lamination, laser drilling — have stretched from roughly 10–12 months to about two years. This means → companies with equipment already in place and mature processes lock in the demand window two years ahead of rivals.
But tight equipment ≠ all capacity is scarce. In plain terms = a handful of leaders are running full order books while many smaller players still face crowded competition and margin pressure — a "winners eat, the rest get squeezed" dynamic.
Shanghai Hua Hong (WUS Printed Circuit) is Citi's higher-conviction pick, highlighted for its AI server and networking PCB position and high utilization. Reference valuation sits at roughly 15–20× risk-adjusted 2027 earnings. Shenghong Tech and Guanghe Tech need more execution evidence — investors want certification progress and stable margins first.
Lenovo — what's holding it up? AI servers offsetting a PC slide
Investors expect Lenovo's Q3 earnings to be relatively resilient, supported by AI server orders and continued margin improvement in the infrastructure-solutions group.
The PC business is not in full recovery; the report even discusses a scenario of double-digit shipment declines, but price increases, a rising enterprise-client mix, and AI PC portfolio improvements may cushion revenue and profit.
Short-term share-price volatility from convertible-bond conversion is seen by investors as a trading-level factor, not a change in operating logic. In plain terms = a brief technical overhang from share supply, not a reason to rethink Lenovo's mid-term AI server story.
Advanced packaging — capex surges, yet investors prefer the equipment layer?
Citi places the highest conviction on the domestic advanced-packaging investment cycle. JCET's capex runs from ~RMB 4 bn in 2024 to ~RMB 8.5 bn in 2025, heading above RMB 10 bn in 2026, with further increases expected through 2027–2028.
Investors prefer the equipment layer, where earnings visibility is clearer. ASMPT is Citi's favored equipment beneficiary. This means → the market sees "selling the shovels" as higher-certainty than "doing the digging."
HBM — high-bandwidth memory, the high-speed stacked memory AI chips require — remains a key bottleneck. Expanding packaging capacity does not mean every critical component is ready.
Consumer electronics — memory rally fading. Who's waiting at low expectations?
Citi's read is that the steepest phase of memory price increases is likely past; sequential gains may narrow into next year, though supply-demand tightness could persist. Investors have not yet treated a cost reversal as a done deal.
Xiaomi and Longcheer Technology are listed as preferred names on the consumer-recovery side, built on low expectations — if memory prices stabilize or retreat, mid-to-low-end Android brands and ODMs could benefit earliest. For Xiaomi, institutions broadly expect Q3 2026 may mark an earnings trough.
The premium smartphone chain lacks a strong catalyst: iPhone 18 Pro's modest price hike and shorter delivery wait times raise demand-strength questions; foldable-phone shipment forecasts have also been trimmed. Luxshare faces near-term headwinds from a muted Apple cycle; the longer-term watch is optical and auto businesses, but investors want clearer optical-module delivery proof.
Software — Citi and investors disagree the most here. Who's right?
IT services and software show the widest gap between Citi's research view and investor feedback: most investors surveyed remain bearish, while Citi, after years of sector adjustment, is slightly more constructive on a bottom forming.
Chinasoft International drew relatively positive feedback; low expectations create re-rating room, but a sustained re-rating requires AI projects to contribute real revenue. Kingdee's AI agents — enterprise-level AI automation tools — could drive revenue acceleration, though near-term financial contribution remains hard to quantify.
This reflects the core tension in AI software: Citi sees emerging spend on AI agents and AI customer service, but investors are still waiting for "numbers that show up in the income statement." Q3 earnings, overseas cloud-vendor capex, and potential optical-communications regulatory developments into late 2026 and early 2027 form the key verification window for the next round of divergence.
市场有风险，内容仅供研究参考，不构成投资建议。