China Galaxy Securities: Large supply-demand gap in phosphorus-cell captive; Industry chain card placement and value reappraisal.

China Galaxy Securities released a research report stating that AI computing power drives the upgrade of optical interconnects, and phosphide gallium covers the entire industry chain from "material bottleneck" to "value reassessment": with the large-scale deployment of 800G optical modules, the introduction of 1.6T acceleration. CPO/SiPh in the next generation of AI clusters is transitioning from sample to mass production, and phosphide gallium (InP) as the only direct bandgap III-V material that can support long-distance high-speed single-mode emission at 1310/1550nm, its strategic position has shifted from "optical communication substrate" to "AI infrastructure bottleneck". By 2025, the global demand for InP devices is approximately 2 million pieces, with actual production capacity of only about 600,000 pieces, resulting in a supply-demand gap of over 50%, with long production expansion cycles, slow yield ramp-up, upstream gallium resources associated with zinc mines, and extremely weak supply elasticity. The main points of China Galaxy Securities are as follows: Upstream gallium resources face dual constraints of "associated + control", with purity bottleneck blocking the industry chain. 90% of global native gallium comes from by-products of zinc mines, with China's refined production accounting for about 70% of the global total; After the Ministry of Commerce implemented export controls on gallium phosphide and other materials in February 2025, the prices of East and West gallium decoupled, and further constraints were placed on the availability of raw materials at the substrate end due to the purification of high-purity gallium above 6N levels (electrolysis-vacuum distillation-regional melting multistage process), leading to increased scarcity in the upstream and downstream. The midstream substrate oligopoly structure is stable, with the core being the upgrade to 6-inch wafers and expansion cycles. The substrate market share is smaller, but its technological barriers and market concentration are much higher than epitaxy: Substrate manufacturing is a crystallization process from 0 to 1, with the core bottlenecks in the upstream, heavily relying on ultra-high purity gallium above 6N levels (99.9999%); The long crystal growth process is extremely difficult, InP single crystal growth requires extreme environmental control at high temperatures and pressures, with long expansion cycles, slow equipment delivery and yield ramp-up; Highly oligopolistic: Sumitomo (42%), AXT/Beijing Tongmei (36%), JX Nippon (13%) together monopolize over 90% of the market share. Sumitomo's Osaka plant is expected to expand production threefold, AXT raised $632.5 million for expansion, filling the gap is expected. Whereas epitaxy grows thin films on substrates, it has lower market concentration and higher market share than substrates. Therefore, the bank believes that InP substrate as the most upstream manufacturing link in the industry chain has the highest technological barriers and the tightest supply and demand, and is the most bottlenecked link in the current AI optical module chain. Shortage of downstream EML drives technology path differentiation, strengthening the non-replaceability of InP on the light source side. Optical chip along EML (current) CW+SiPh (high-speed development) CPO+ELS (2026-2028) Silicon-based III-V single-chip (2030+) four generations of evolution; EML is facing capacity constraints from companies like NVIDIA, with deliveries delayed until after 2027, driving accelerated penetration of CW+SiPh, but CW's InP base still relies on Sumitomo/AXT, with demand for InP increasing rather than decreasing. Profitability in the industry chain is concentrated on "substrate/epitaxy + high-performance optical chips", with vertically integrated manufacturers enjoying increased bargaining power. Substrate technological barriers and manufacturer concentration are highest; epitaxy is more dispersed as MOCVD can be procured, but it is highly customized and has strong customer stickiness; downstream EML/CW/UHP chips and CPOELS module ASP have 2-3 times elasticity. IDMs (Lumentum, Coherent) with full-stack capabilities in InP single crystal growth, epitaxy, chips, and modules, along with leading substrate material suppliers (Sumitomo, AXT/Tongmei), will benefit from the premium in filling the gap. Investment recommendations Industry supply falls short of demand, with high prosperity. It is recommended to focus on related leading companies in the optical communication sector: AXT (AXTI), Sumitomo Electric Industries (5802.T), IQE (IQE.L), Applied Optoelectronics (AAOI.O), Lumentum (LITE), Coherent (COHR), Ciena (CIEN.N), etc., focusing on profit capabilities and dual upward valuation trends. Risk warning 1. Risks of demand fluctuations due to CSP capital expenditure falling short of expectations; 2. Risks of disturbances in upstream gallium supply and export controls; 3. Risks of substrate and epitaxy expansion and slow yield ramp-up falling short of expectations; 4. Risks of alternative technical paths and slower-than-expected penetration of CPO.