Demand for “optical interconnection” is bursting with demand in the AI reasoning era! “Standing in the Light” Lumentum (LITE.US) rushed NPO and expanded OCS to meet a time of rising profits

The Zhitong Finance App learned that the minutes of the Citigroup Global TMT conference show that Lumentum (LITE.US) management, the most core participant in the “Google TPU AI Computing Power Chain” and an indispensable core optical/optical module component supplier of the “Nvidia and AMD AI GPU AI Computing Power Chain”, shows that AI infrastructure is still in the early stages of construction, and the expansion of computing power and physical restrictions on copper interconnection will jointly drive the high-speed interconnection networks within large AI data centers towards optics. The revenue and earnings per share growth rate for the 2027 and 2028 fiscal years will maintain Strong growth trajectory.

The Citigroup analysts' bullish investment logic for Lumentum comes down to three mutually reinforcing growth paths: AI cluster expansion drives demand to continue to exceed production capacity, optical circuit switching technology trajectory (OCS) opens up a larger system-level market, and near-package optics (NPO) and co-packaged optics (CPO) are jointly expanding demand for high-power lasers and external light sources. Citi maintained a “buy” rating and a target price of $1,200, which was about 21.3% higher than the closing price of $988.98 on September 9 used in the report, and retained upward catalytic observations as of October 14. Lumentum's stock price has soared about 170% since this year, and the increase was as high as 340% in 2025.

Lumentum management recently raised its earnings target for FY2028 to $40, mainly based on the continued increase in the visibility of OCS and NPO business growth; Citi's latest calculation model in the minutes still anticipates the company's overall revenue of about 6.828 billion dollars and 10.532 billion US dollars for the 2027 and 2028 fiscal years, respectively, which means that it is expected to achieve significant year-on-year increases of 126.5% and 54.3%, while the core earnings per share forecast are $23.03 and $37.21, respectively.

Why can Lumentum's stock price trend from 2025 to be called “extreme”? The main logic is that Lumentum can simultaneously benefit from Google's TPU AI computing power industry chain and the AI GPU computing power industry chain dominated by Nvidia. In the future, whether it is Nvidia's AI GPU computing power infrastructure cluster or Google's TPU cluster (TPU represents the AI ASIC technology route), it will be inseparable from “optical interconnection and CPO silicon optical high-speed interconnection capabilities represented by Lumentum”.

Lumentum has joined the S&P 500 index, the benchmark stock index for US stocks, and the Nasdaq 100 Index (“NASDAQ 100” for short), which has the title of “Global Technology Stock Weather Vane”. In other words, Lumentum has now obtained passive capital allocation qualifications for two core US stock benchmark indices at the same time, which has further strengthened its institutional holding base as a core weighted growth stock for AI Optical Interconnect.

After the inclusion of the benchmark index, it can be said that Lumentum was officially promoted from the “niche supply chain target” of AI data center optical interconnection to the global technology core asset pool. It also means that the main AI computing power investment narrative is rapidly shifting from “a single-point computing power competition around AI GPU/ASIC” to an “artificial intelligence inference end full-stack computing power system driven by AI agents”. In this AI mainline narrative transformation, data center CPUs, memory chips, and optical interconnect/optical communication industry chains are probably the biggest winners.

Recently, the stock prices of many data center optical product suppliers such as Lumentum have continued to be strong. There is also market news that the “AI chip hegemon” Nvidia (NVDA.US) is adjusting the next-generation AI rack computing power system architecture. That is, the next generation AI computing power cluster Rubin Ultra may reduce the HBM configuration of each rack and instead use optical interconnection technology architectures to connect multiple AI racks. As the prices of HBM/DRAM memory systems and NAND memory chip products have continued to soar in recent years, the focus of AI computing power infrastructure investment seems to be shifting from GPU/TPU single-point performance and HBM capacity to overall data center architecture efficiency. Among them, high-speed optical interconnection is probably the most critical direction.

Out of stock extended to 2027: Lumentum turns near-endless demand for AI optical interconnection into strong fundamentals

Lumentum management judged that AI infrastructure is still in the early stages of construction, and the expansion of computing power and physical limitations of copper interconnection are jointly promoting the optical high-speed interconnection network within the data center. Using interconnection between data centers (DCI) as a benchmark, the management described the interconnection demand intensity of the scale-out network (scale-out) as about 10 times. The future migration of AI computing power accelerator clusters to tight coupling interconnection (scale-up) to optics will also open up room for a stronger level of growth.

According to Citigroup Minutes forecast data, shipments of related modules will approximately double from 2026 to 2027, and will increase further in 2028, while the supply shortage of electrically absorption modulated lasers (EML) and continuous wave lasers (CW) is expected to continue throughout 2027. The price of the new contract remains at or higher after the previous increase, indicating that demand has not declined significantly due to price increases. In terms of product structure, 800G EML still dominates, and 1.6T is expected to increase; CW production capacity is also used to fill customer gaps, seek future design orders, and support its own relevant modules. Its yield and profit margins are improving, but it has not yet reached the EML level.

OCS and pump lasers form two significant incremental businesses. As the largest OCS customer's own hardware business expands, the number of switching ports required has increased; new uses such as dynamically dividing accelerator resources and bypassing abnormal performance nodes have also made management believe that the total potential OCS market estimate of about 8 billion US dollars is low, and may extend business visibility to 2028 and early 2029. The management used an empirical estimate of about 1.5 ports corresponding to 1 accelerator, and emphasized that profit growth in FY2028 will gradually be realized rather than concentrated at the end of the year.

The Citigroup analyst team notes that the pump laser business related to cross-cluster and cross-park connectivity is also strong: it has been growing for ten consecutive quarters, signed long-term agreements with all major network equipment manufacturers, and the business volume is expected to quadruple in the next five quarters. These data and anticipated prospects mean that Lumentum's growth sources have covered laser chips, optical modules, and data center optical switching systems in a broad sense.

Whether supply can be delivered depends on high performance manufacturing and customer certification. The production capacity of ultra-high power lasers is progressing in three stages: Lumentum's existing plant in the US continues to climb until 2027; the UK plant will increase production capacity from the second quarter of 2027 and continue until 2028; the third factory is aiming to start in mid-2028, and the key constraint on early release is the customer certification cycle.

Lumentum's management believes that the combination of high power, low relative intensity noise, narrow line width, and no mode hopping performance forms the threshold for large-scale manufacturing, and sees Coherent as the only comparable competitor for ultra-high power lasers at present. Regarding the Chinese competitor, management said it has yet to see a new EML external supplier with substantial influence, while admitting that the industry needs a second source of supply to support 2028 demand.

The more capable AI agents are, the more unblocking the network and data transmission: the “next HBM system” is AI optical interconnection

NPO first expanded the market, bringing Lumentum an opportunity to benefit across architectures. In an interview with Citi, Lightmatter, another leader in optical technology, predicted that NPO will enter the market in 2027, be widely deployed in 2028, and that the large-scale window for CPO for scale-up will be in 2028-2029; Lumentum predicts that by 2028, NPO and CPO will each account for about half of the optical scale-up market. Of these, about three-quarters of NPOs use external light sources to avoid the thermal environment near accelerators.

Even with an integrated light source, they tend to use fewer, more powerful lasers, with a power range of about 120-150 mW and above. Lumentum has received the first CPO customer's external laser light source (ELS) project, and delivery is expected to begin in the second half of 2027, covering 15% to 20% of the customer's related equipment deployment; NPO customers have already received widespread interest, but no official project winners have been announced. Meanwhile, increased demand for CPO orders has not changed the certification schedule.

The insufficient test capacity indicated by Lightmatter indicates that commercialization also depends on mass production infrastructure; its L20 unidirectional 6.4 Tbps bandwidth and single-fiber bidirectional transmission also reflect that the industry is increasing bandwidth while reducing wiring costs.

Looking at the underlying engineering, the growth in optical interconnect demand comes from more frequent and stricter data exchanges between more computing nodes: experts in hybrid expert models (MoE) require parallel data distribution and aggregation across GPUs; prefill—decode disaggregation (Prefill—Decode Disaggregation) requires transmission of key-value caches (KV Cache); and complex agent workflows increase the amount of concurrent model calls, tool execution, and data access. As clusters expand, network congestion will cause expensive GPUs to wait for data, directly reducing the amount of effective tasks that can be delivered per dollar or watt. The distance, loss, and power consumption constraints of high-speed electrical connections therefore drive a wider range of optical connections. NPO/CPO shortens electrical signal paths, EML, CW, and ELS provide the light sources required for optical communication, and OCS supports reorganization of connections according to workload.

Regarding the AI optical interconnection growth logic surrounding Lumentum, the deployment scale and communication requirements are expanding, rather than “the stronger the model, each task will inevitably consume more tokens” — OpenAI just emphasized that Astra reduces retries and single-task token consumption. Judging from Goldman Sachs's growth expectations for SanDisk, NAND will tend to take on large-scale data persistence and partial cache tiering in the future, and optical interconnect will undertake cross-node handling. Both will benefit from the increase in data capacity and communication demand for AI inference systems that handle heavyweight workloads, respectively. How much revenue Lumentum can realize ultimately still depends on order share, certification progress, and rising production capacity; the Citigroup analyst team also listed slowing AI construction, increased competition, and falling short of expectations in CPO/OCS adoption as the main risks.