The more stunning AI capabilities become, the faster data must run! Goldman Sachs' optical industry survey is out: CPO testing and coupling equipment are the first to realize revenue.

Title context: The more stunning AI capabilities become, the faster data must run! Goldman Sachs' optical industry survey is out: CPO testing and coupling equipment are the first to realize revenue. Text: Goldman Sachs, the Wall Street financial giant, recently released its research report "Optical Industry Survey: Ten Key Points on Technology, Demand, Supply and Competition," showing that the unprecedented trillion-dollar-scale investment by tech giants around AI computing infrastructure is simultaneously and substantially expanding the number, speed, and application scope of high-speed optical connectivity units in data centers. Growth opportunities are further accelerating from strong demand for high-speed optical module systems to lasers, fiber arrays, optical industry-related testing, and high-performance automated coupling equipment. Goldman Sachs had previously raised its 2027 and 2028 demand forecasts for 800G and above optical modules by 39% and 36%, respectively, to 144 million and 171 million units. After this latest round of research on leading Chinese data center optical product line companies, Goldman Sachs analysts believe there is still room for upward revision; at the same time, DSP, laser, and PCB supply may continue to constrain 2027 shipments. From an underlying engineering perspective, growing optical interconnect demand comes from more frequent and more stringent data exchange among more compute nodes: expert parallelism in Mixture-of-Experts (MoE) models requires distributing and aggregating data across GPUs; PrefillDecode Disaggregation requires transmitting the key-value cache (KV Cache); complex agent workflows increase the concurrency of model calls, tool execution, and data access. As clusters expand, network congestion causes expensive GPUs to wait for data, directly reducing the effective task volume that can be delivered per dollar and per watt. The distance, loss, and power constraints of high-speed electrical connections therefore drive broader optical connectivity. NPO/CPO shortens the electrical signal path, EML, CW, and ELS provide the light sources required for optical communication, and OCS supports reconfiguring connections based on workloads. Therefore, as demand related to frontier AI applications accelerates and expands, the requirements for transmission bandwidth and efficiency in AI cluster collaboration are becoming increasingly high. Goldman Sachs believes that the investment value in the data center optical interconnect/optical communication field depends on whether companies can obtain key materials, complete high-speed product certification, and convert demand into deliverable capacity. The report especially emphasizes that CPO commercialization takes time, but testing and coupling equipment have already begun to generate revenue, and the WINOX clocks of different links in the industry chain are not fully synchronized. Regarding ratings for stocks covered by Goldman Sachs, Goldman Sachs gives "Buy" ratings to Zhongji Innolight A/H shares, Robotechnik Intelligent Technology, VPEC, and FOCI. The target price for Zhongji Innolight A shares is based on a 2027 P/E of 35.6x, while the H-share target price assumes a 13% H/A premium; for VPEC, Robotechnik Intelligent Technology, FOCI, and ASMPT, more forward earnings are used and discounted respectively. For the Zhongji Innolight A-share target price, Goldman Sachs gives a target price of RMB 2,645, implying potential upside of 185.6% over the next 12 months. For the Zhongji Innolight Hong Kong share target price, Goldman Sachs gives a target price of HKD 3,267, implying potential upside of 179.2% over the next 12 months; for Robotechnik Intelligent Technology, Goldman Sachs' latest target price implies about 26% potential upside over the next 12 months; for Taiwan-based VPEC (2455.TW), Goldman Sachs' latest target price implies 54% potential upside over the next 12 months. Goldman Sachs expects high-speed optical module demand to reach 144 million units in 2027! The optical interconnect opportunity extends from modules to manufacturing bottlenecks Demand expansion and price resilience are the basis for Goldman Sachs' bullish view on high-speed optical modules. Goldman Sachs said that the surveyed Chinese companies gave a global demand range of 130 million to 200 million for 800G and 1.6T modules, with overall demand in the Chinese market expected to exceed 50 million units, mainly 400G and 800G, while 1.6T is beginning to enter the market; however, these scopes differ in product coverage and cannot be directly added together. On pricing, the surveyed companies expect the price decline for 800G modules in 2027 to be kept in the single digits, while 1.6T prices remain above USD 700 per unit, meaning shipment growth is still expected to revenue growth. The data communications product iteration cycle has shortened to about one to two years, significantly faster than the roughly five-year cycle in the telecom market, making R&D, supply chain management, mass production, and cross-regional delivery capabilities competitive thresholds. Goldman Sachs said that Zhongji Innolight secures supply through joint R&D, investment in suppliers, prepayments, and long-term agreements, reflecting the path by which a leader turns procurement capability into a competitive advantage. The surveyed companies also believe demand will be strong over the next two to five years, with some customers already planning beyond 2030; their stated payback period of about one to two years is an operating judgment of the surveyed industry and cannot be generalized as a certain return for all AI projects. Goldman Sachs said that upgrades in lasers and fiber arrays are causing supply constraints and improved domestic supply capability to occur simultaneously. Suzhou Dongshan Precision Manufacturing said that 200G electro-absorption modulated lasers (EML) have achieved mass production, but current output is still limited by tight digital signal processor (DSP) supply; the company has locked in supply of 10 million DSPs for 2027, supporting subsequent growth in laser and optical module shipments. Suzhou Everbright Photonics expects 2027 EML capacity to be slightly higher than continuous-wave lasers (CW), with EML still mainly 100G and supplemented by 200G. VPEC sees improvement in indium phosphide (InP) substrate supply and plans to increase metal-organic chemical vapor deposition equipment (MOCVD) from 62 units to 69 units by the second quarter of 2027, and is planning a new production base; its epitaxial wafer products are also upgrading from serving 70100 mW CW lasers to above 100 mW products required for NPO and 300400 mW products required for CPO, while also expanding 100G and 200G EML. In addition, FOCI's fiber array units (FAU) are upgrading from 40 channels serving 3.2T to 80 channels serving 6.4T, while 100 channels are still under development; August revenue rose 20% month over month, higher than 9% in July. These changes support an increase in the share of high-end products, but they cannot be used to conclude that heat dissipation, yield, and certification issues for high-power lasers have already been resolved. CPO's early gains are emerging in testing and automation equipment. Goldman Sachs expects CPO switch shipments in 20262028 to be 10,000, 92,000, and 131,000 units, respectively, while observing strong demand from Chinese and overseas customers for 3.2T near-packaged optics (NPO) engines. Robotechnik Intelligent Technology's second-quarter 2026 revenue grew 172% quarter over quarter, 141% higher than Goldman Sachs' forecast; together with its subsidiary FiconTEC, it expects equipment shipments over the next year to exceed half of the cumulative shipments over the past 25 years, and aims to shorten wafer-level testing time by 50%60% next year and increase chip-level testing speed to four times. Goldman Sachs analysts also wrote that high-value photonic integrated circuits (PIC) must identify defects as early as possible, so double-sided wafer testing, known good die (KGD) screening, and precise coupling become key to reducing scrap costs. Photon Technology's NightJar system is used to identify optical loss and defects in advance; Gallant Precision's dual fiber array active alignment equipment can simultaneously handle the transmit and receive ends, shortening coupling time by 50%; HYT's assembly systems cover applications such as optical modules, external light sources, OCS, NPO, and CPO; ASMPT's MEGA platform integrates high-precision placement, dispensing, UV curing, and 3D inspection. What Goldman Sachs analysts are uniformly bullish on is therefore a whole set of upgrades in photonic manufacturing capability. The more capable AI agents become, the more data center networks and high-speed data transmission must not become congested! The strong AI computing demand support linked to the AI computing industry chain has already been significantly reflected in the strong performance and long-term capacity agreements of industry leaders. Nvidia's revenue in the second quarter of fiscal 2027 was USD 96.2 billion, up 106% year over year, of which data center revenue was USD 89 billion, up 117% year over year; Anthropic's disclosed capacity arrangements include cooperation with Amazon for up to 5 GW, with Google and Broadcom for 5 GW, and obtaining more than 300 MW and more than 220,000 Nvidia GPUs of computing power through SpaceX. These agreements are at different delivery stages and support visibility into continued construction and expansion; they cannot all be counted as capacity already online. The market rebound also highlights how strong AI computing demand strengthens the bullish logic across the entire computing industry chain. South Korea's KOSPI rebounded about 22% on August 13 from its July 30 low, and the Philadelphia Semiconductor Index closed at 12,621 on August 17, rebounding more than 20% from its July 29 low, successively entering technical bull markets. The recovery of memory heavyweights such as Samsung and SK Hynix and AI earnings expectations reinforced each other, but in September South Korea's KOSPI benchmark index still showed a pullback, which also shows that there are still clear expectation fluctuations during the AI computing rebound. From an underlying technology perspective, the strong incremental growth in optical interconnect in the AI inference era comes from "more concurrent workloads + broader cross-node collaboration." OpenAI recently launched its most frontier AI large model Astra, which can improve computer operation, software engineering, and completion of complex workflows, enabling more tasks that were originally uneconomical to be handed to agents. OpenAI's product head's statement that demand is unprecedented to the point that the company may pause new Pro subscriptions is an important signal of pressure on AI computing service capacity. Therefore, the continued large-scale expansion of computing demand brought by the most frontier large models will also become a new round of strong growth catalysts for data center optical interconnect product lines. In systems using Mixture-of-Experts (MoE) expert parallelism, data needs to be distributed and aggregated across accelerators; PrefillDecode Disaggregation requires transmitting the key-value cache (KV Cache); agents calling tools, accessing databases, and sharing storage increase system-level data traffic. When the network slows data exchange, GPUs cannot deliver the corresponding effective task volume even if they have very high peak computing power. The bandwidth, distance, and energy efficiency advantages of optical connections, as well as the ability of NPO/CPO to shorten high-speed electrical signal paths, therefore have strong and obvious economic value. Demand growth does not require each task to consume more tokens: even if the efficiency of individual tasks improves, as long as new tasks and concurrency scale expand faster, total AI computing infrastructure resource scale and demand for high-performance network equipment can still grow; however, the number of ports, per-port speed, and optical adoption ratio are the most direct variables for estimating optical module demand. NavierStokes research provides a more specific scale reference: OpenAI said its stronger internal model organized about 10,000 concurrent agents, forming a solution in about 88 hours, after which Astra completed Lean formalization and verification in about 17 hours; all research attempts generated about 300 billion output tokens, of which the NavierStokes portion was about 130 billion, demonstrating the scale that scientific research-grade inference can reach. In addition, Recursive Self-Improvement (so-called RSI) further opens up demand space for scientific research computing: AI participation in code writing, experiment design, data generation, evaluation, and optimization of training tools may increase continuously running experimental and inference workloads. OpenAI has clearly advanced its research direction toward RSI. OpenAI's GPT-6 Astra large model, as well as the RSI technology path focused on by AI leaders, are expected to become the two core DRIVEs driving exponential expansion of AI computing demand, namely more powerful AI large models and broader use of AI application tools, and a next-generation AI training path with even stronger computing demand. These are important bases for strengthening the continued growth of AI computing infrastructure demand. Combined with Goldman Sachs' latest survey, the new growth curve has technical and commercial foundations. For investment, one should track actual high-speed module deliveries, laser yields, test equipment acceptance, and cash flow per share: only companies that can convert complexity and supply constraints into customer value have the opportunity to continue realizing the profit premium of AI optical interconnect.