A-Share AI Infrastructure Boom: CPO Optical Modules Lead Rally as NewEase Surges Past $60B Market Cap

Full-Blown Boom Across the AI Compute Hardware Chain: A-Share Market Establishes a Structural Investment Theme—Domestic Substitution and Global AI Capex Cycles Converge as Core Catalysts

On April X, the A-share market exhibited a pronounced “technology-led rally”: the Shanghai Composite Index rebounded above 4,100 points (closing at 4,106.26, +0.52%), the ChiNext Composite Index hit an all-time intraday high, and growth-oriented indices—including the STAR 50 and CSI 1000—led broad-based gains. Most strikingly, sub-segments of the AI compute hardware chain—including Coherent Pluggable Optics (CPO), optical modules, compute leasing, advanced packaging, and high-speed connectors—saw a wave of涨停 (trading limits reached): over twenty stocks—including Dongshan Precision, Changguang Huacore, Annuoqi, Hengwei Technology, and Hanggang Co., Ltd.—hit their daily upper limits. New Access (300502.SZ) surged nearly 8% in a single day, reaching a new all-time high and pushing its total market capitalization past RMB 600 billion for the first time. Foxconn Industrial Internet (FII, 601138.SH) surged straight to its daily limit in the afternoon (RMB 67.55 per share). Invek (002897.SZ) staged an extreme V-shaped intraday reversal, signaling intense investor demand. Total market turnover soared to RMB 2.58 trillion, with over 2,900 stocks closing higher. This rally is no short-term thematic speculation—it is the most certain structural investment theme emerging from the powerful convergence of three forces: accelerated deployment of AI infrastructure investments, deepening domestic substitution, and upward revisions to global AI capital expenditure (capex) forecasts.

CPO & Optical Modules: Breaking the High-Speed Interconnect Bottleneck—Chinese Vendors Ascend to the Global Top Tier

Optical communications constitute the “vascular system” powering AI large-model training and inference. As GPU cluster scale expands toward tens of thousands of cards, traditional copper cabling can no longer meet TB/s-level bandwidth demands—within server racks, between racks, and across data centers—making 800G/1.6T optical modules essential. CPO (co-packaged optics) technology integrates optical engines heterogeneously with switching ASICs, significantly reducing power consumption and latency; it is rapidly transitioning from lab prototypes to mass production. Against this backdrop, Chinese optical module vendors have achieved leapfrog advancement by leveraging vertical integration capabilities and rapid response advantages. New Access—now a leading 800G supplier—has secured large-scale orders from top North American cloud providers; its 1.6T silicon photonics module samples are already undergoing customer validation. Zhongji Xunyi continues to lead global market share. Cambridge Industries Group and TFC Communications have built formidable moats in high-speed optical components. Notably, the optical module index has risen over 75% year-to-date—far outpacing semiconductor equipment and design indices—confirming the market’s strong conviction in optical interconnects as the “first and non-negotiable gate” enabling AI compute scalability: breaking the physical bottleneck of optical interconnects is the indispensable prerequisite for unleashing AI compute potential.

Compute Leasing: Asset-Light Model Lowers AI Adoption Barriers—Deep Integration Between IDCs and Server OEMs

The explosive rise of the compute leasing theme reveals a pivotal shift in AI commercialization pathways. SMEs and vertical-domain developers struggle to bear the hundreds-of-millions-to-billions-RMB costs of procuring and maintaining GPU clusters. Compute leasing—offering “pay-as-you-go, instant-on” access—dramatically lowers the barrier to AI application deployment. Companies such as Annuoqi and Hengwei Technology leverage their own high-performance AI computing centers to deliver end-to-end services—from model training to inference deployment. Hanggang Co., Ltd. capitalizes on regional advantages in electricity supply and land resources to accelerate construction of a Yangtze River Delta AI computing hub. More fundamentally, the boom in compute leasing directly drives upstream server shipments: Foxconn Industrial Internet—the world’s leading AI server OEM—reaching its daily trading limit reflects not only strong earnings expectations (Q1 AI server orders up >300% YoY) but also epitomizes broader supply-chain synergy: IDC operators purchase servers → server OEMs procure GPUs, optical modules, and thermal management components → upstream semiconductor and photonic device suppliers ramp capacity. This virtuous cycle—“application-driven demand → infrastructure build-out → hardware volume expansion”—is transforming AI compute investment from abstract “concept” into verifiable revenue and cash flow.

Deepening Domestic Substitution: From “Functional” to “Superior”—Supply Chain Security Evolves into Performance Competitiveness

While consumer sectors—including film & television and pharmaceutical distribution—weakness during this correction highlights investors’ extreme focus on “hard tech.” Underpinning this shift is a qualitative evolution in the domestic substitution thesis: previously anchored in “self-reliance as a baseline,” it has now matured into a dual driver of “performance leadership” and “cost advantage.” Take photonic chips as an example: Changguang Huacore’s high-power VCSEL chips are already deployed in LiDAR systems and AI photonic computing. Luvip’s photomasks match international leaders in yield and precision, supporting domestic advanced packaging capacity ramp-up. Guanshi Technology’s OCA optical adhesives have broken the Japan–Korea duopoly and are now used in AI-enabled end-device displays. Crucially, domestic solutions demonstrate differentiated advantages in AI-specific use cases—for instance, Cambricon’s MLU series AI accelerators deliver superior inference energy efficiency versus select international competitors, prompting server OEMs to proactively integrate them. When domestic hardware transitions from “backup option” to “first choice,” supply chain security ceases to be merely a defensive narrative—it becomes a core variable enhancing overall compute cost-performance ratios.

Upward Revision of Global Capex: China’s Compute Infrastructure Enters “Acceleration Phase”—Northbound Fund Allocation Logic Undergoes Fundamental Shift

Market consensus is being continuously reinforced by overseas data. According to Counterpoint Research, global AI server shipments are projected to grow 85% YoY in 2024, with China’s share rising above 30% for the first time. TSMC has raised its CoWoS (Chip-on-Wafer-on-Substrate) packaging capacity to 100,000 wafers per month—yet demand still vastly exceeds supply. Microsoft and Meta both significantly raised full-year capex guidance in their latest earnings reports, directing substantial increases toward AI infrastructure. Against this backdrop, northbound fund allocation logic toward A-share tech stocks has undergone a fundamental shift: investors now prioritize “inclusion in global AI giants’ supply chains,” “capacity for continuous technological iteration,” and “possession of scarce, scalable manufacturing capacity”—over traditional metrics like P/E multiples or near-term profitability. New Access’s ~890% stock price surge over the past year and FII’s sustained foreign institutional buying exemplify this new paradigm. As China’s AI compute hardware chain becomes an indispensable infrastructure provider for global AI expansion, its valuation benchmark will undergo systemic upward re-rating.

Conclusion: The Theme Is Far From Exhausted—Three Key Evolutionary Frontiers to Watch

This rally is far from over. Going forward, investors should closely monitor three critical evolutionary vectors:

1. Commercialization timeline: the pace of mass delivery for 800G optical modules and the commercial rollout schedule for 1.6T modules;
2. Business model upgrade: whether compute leasing can evolve beyond “selling raw compute resources” toward “selling intelligent services”—e.g., pre-trained model APIs and domain-specific fine-tuning platforms;
3. Penetration acceleration: the real-world adoption rate of domestic GPUs and AI chips within server deployments.

As AI compute hardware advances from “functional” to “superior” and “user-friendly,” this golden industrial chain—spanning semiconductors, optical communications, servers, and IDCs—will continue to define the long-term value anchor for China’s technology equities. It is more than just a stock-market theme: it is the steel spine underpinning China’s substantive competitive barriers in the global AI race.