In the era of AI inference, demand for "optical interconnects" is booming! Lumentum (LITE.US), "standing in the light," is ahead of NPO, expanding OCS to embrace the moment of profit surge

According to The Zhitong Finance APP, the minutes from the Citi Global TMT Conference reveal that Lumentum (LITE.US)—the core participant in the "Google TPU AI Computing Power Chain" and an indispensable key optical/optical module component supplier in the "Nvidia and AMD AI GPU Ecosystem AI Computing Power Chain"—has shared the latest management outlook, stating that AI infrastructure remains in the early stages of development. The combined effect of computing power expansion and the physical limitations of copper interconnection is driving a shift towards optical interconnects for high-speed internal networks in large AI data centers. Revenue and earnings per share (EPS) are expected to maintain a strong growth trajectory into fiscal years 2027 and 2028.

The Citi analyst team summarizes their bullish investment thesis for Lumentum as stemming from three mutually reinforcing growth paths: the expansion of AI clusters is stimulating laser demand to exceed supply continuously, optical circuit switching (OCS) technology is opening up a larger system-level market, and near-packaged optics (NPO) alongside co-packaged optics (CPO) are jointly expanding demand for high-power lasers and external light sources. Citi maintains a “Buy” rating and a target price of $1,200, about 21.3% above the September 9 closing price of $988.98 used in the report, and retains a catalyst watch through October 14. Since the start of the year, Lumentum's share price has soared by approximately 170%, with a remarkable increase of 340% in 2025 alone.

Lumentum's management recently raised its earnings per share target for fiscal year 2028 significantly to $40, mainly based on increasing visibility into the growth of its OCS and NPO businesses. Citi’s latest model in the minutes still forecasts overall company revenues for fiscal years 2027 and 2028 at approximately $6.828 billion and $10.532 billion, respectively, implying year-over-year increases of 126.5% and 54.3%. Core EPS projections are $23.03 and $37.21, respectively.

The main reason why Lumentum’s stock performance in 2025 to date can be described as “exceptionally bullish” is that Lumentum can simultaneously benefit from both the Google TPU AI computing power industry chain and the Nvidia-led AI GPU computing power chain. In the future, whether it is Nvidia AI GPU-based infrastructure clusters or Google TPU clusters (TPU represents the AI ASIC technology path), both are inseparable from “the optical interconnect and CPO silicon photonics high-speed interconnect capabilities represented by Lumentum.”

Lumentum has been included in major US stock indices—the S&P 500 and the Nasdaq 100, a global technology stock benchmark. This means Lumentum now qualifies for passive capital allocation in both core US equity indices, further solidifying its institutional ownership as a core AI optical interconnect growth stock.

Being included in benchmark indices has officially moved Lumentum from a “niche supplier in the AI data center optical interconnect supply chain” into the global technology core asset pool. It also means that the AI computing power investment narrative is shifting from a “single-point AI GPU/ASIC computational race” to a “full-stack computing system driven by AI agents for AI inference.” In this new narrative, data center CPUs, storage chips, and the optical interconnect/optical communication industry chain are likely to be the biggest winners.

The recent surge in stock prices of Lumentum and other data center optical product suppliers is also attributed to market rumors that “AI chip leader” Nvidia (NVDA.US) is adjusting future AI rack computing system architectures. For example, the next-generation AI computing cluster Rubin Ultra may reduce per-rack HBM configurations in favor of optical interconnect technology to connect multiple AI racks. Due to the ongoing surge in prices of HBM/DRAM memory systems and NAND storage chips in recent years, the focus of AI infrastructure investment appears to be shifting from GPU/TPU single-point performance and HBM capacity towards overall data center architecture efficiency, where high-speed optical interconnect is potentially the most critical direction.

Shortages Extend to 2027: Lumentum Turns Near-Infinite Demand for AI Optical Interconnect into Robust Fundamentals

Lumentum's management assesses that AI infrastructure is still in its initial construction stages; computing power expansion and the physical limits of copper interconnection are collectively driving data centers’ internal high-speed interconnect networks to become optical. By setting data center interconnect (DCI) as the benchmark, management describes the interconnection intensity for scale-out networks as roughly 10 times higher. The future shift of tightly coupled interconnect (scale-up) for AI accelerator clusters towards optics is expected to open up even stronger growth opportunities.

Citi’s forecast data in the minutes predict that shipments of related modules will roughly double from 2026 to 2027, with further growth into 2028. Shortages of electro-absorption modulated lasers (EML) and continuous-wave (CW) lasers are expected to persist through 2027. New contract prices remain at or above previously raised levels, indicating that demand has not meaningfully retreated due to price hikes. In terms of product mix, 800G EML remains dominant, and the share of 1.6T is expected to increase. CW capacity is utilized for filling customer gaps, securing future design wins, and supporting in-house coherent modules, with yield and profit margins improving but not yet at the EML level.

OCS and pump lasers constitute two incremental business lines worth noting. As the largest OCS customer expands its own hardware business, the required number of switching ports increases. New uses, such as dynamic partitioning of accelerator resources and bypassing performance anomaly nodes, lead management to believe the previous $8 billion estimate for the total OCS addressable market is low and that business visibility may extend to 2028 and early 2029. Management uses a rule of thumb: about 1.5 ports correspond to one accelerator, and emphasizes that fiscal year 2028 profit growth will materialize gradually, not just at year-end.

The Citi analyst team notes that cross-cluster, cross-campus connectivity-related pump laser business is also strong: it has grown for ten consecutive quarters, has long-term agreements with all major networking equipment vendors, and volume is expected to quadruple over the next five quarters. These data and outlooks indicate that Lumentum’s growth sources now cover laser chips, optical modules, and, more broadly, data center optical switching systems.

Whether supply growth can be realized depends on high-performance manufacturing and customer qualification. Ultra-high-power laser capacity will be advanced in three stages: Lumentum's existing US plant will continue to ramp through 2027; the UK facility will add capacity from the second quarter of 2027, extending into 2028; a third facility aims to launch in mid-2028, with early ramp-up constrained primarily by customer qualification cycles.

Lumentum’s management believes the combination of high power, low relative intensity noise, narrow linewidth, and mode-hop-free performance constitutes a manufacturing scale barrier, with Coherent considered the only comparable competitor in ultra-high-power lasers currently. Regarding Chinese competitors, management says there have not yet been any new EML external suppliers with substantial impact, but acknowledges the industry will need a second source of supply to support 2028 demand.

The More Capable AI Agents Become, the Less Tolerant Network and Data Transmission Can Be of Bottlenecks: “The Next HBM System” Is AI Optical Interconnect

NPO’s early market expansion brings Lumentum the opportunity to benefit across architectures. Another optical technology leader, Lightmatter, explained to Citi in an interview that it expects NPO to enter the market in 2027 and see broad deployment in 2028, with the window for scale-up CPO coming in 2028–2029. Lumentum expects that by 2028, NPO and CPO will each account for about half of the optical scale-up market, with approximately three-quarters of NPO using external light sources to avoid thermal environments near accelerators.

Even when using integrated light sources, there is a preference for fewer, higher-powered lasers in the range of about 120–150 mW and above. Lumentum has secured its first CPO customer for the external laser source (ELS) project, with deliveries expected to start in the second half of 2027, covering about 15%–20% of that customer’s related equipment deployments. NPO customers have shown broad interest but no formal project wins have been announced. Meanwhile, reinforced CPO order demand has not yet altered certification schedules.

Lightmatter’s noted testing capacity bottleneck indicates that commercialization is still subject to mass production infrastructure; its L20 chip’s single-direction 6.4Tbps bandwidth and single-fiber bidirectional transmission likewise reflect the industry’s focus on raising bandwidth while reducing wiring costs.

From an engineering perspective, the growth in optical interconnect demand comes from more frequent and stringent data exchanges among a greater number of compute nodes: Mixture-of-Experts (MoE) model parallelism requires data distribution and aggregation across GPUs; Prefill–Decode Disaggregation requires transmitting key value caches (KV Cache); complex agent workflows increase the concurrency of model calls, tool executions, and data accesses. When clusters expand, network congestion can result in expensive GPUs idling for data, directly reducing the effective task throughput per dollar and per watt. The limitations on reach, loss, and power consumption of high-speed electrical connections are therefore driving broader adoption of optical connectivity. NPO/CPO shortens electric signal paths, EML, CW, and ELS provide the requisite light sources for optical communication, and OCS supports workload-driven reconnection.

The growth logic around Lumentum’s AI optical interconnect is based on the scale of deployment and expanding communication needs, rather than “the stronger the model, the more tokens every task must consume”—as OpenAI specifically emphasizes Astra’s reduction of retries and single-task token consumption. Compared with Goldman Sachs’s expectations for SanDisk, NAND will in the future mainly handle massive-scale data persistence and some cache layering, while optical interconnects will handle cross-node transport, each benefited by the growth in data capacity and communication needs of AI inference systems coping with trillion-level workloads. How much profit Lumentum can capture remains dependent on order share, qualification progress, and production ramp-up; the Citi analyst team also points to potential risks, including the slowing of AI buildout, intensified competition, and CPO/OCS adoption falling short of expectations.