Citi's Global TMT Conference notes reveal that Lumentum, a core participant in the Google TPU AI computing chain and an indispensable optical module supplier for Nvidia and AMD AI GPU systems, sees AI infrastructure still in its early construction phase. The combination of computing power expansion and copper interconnect physical limitations is driving large AI data centers toward optical high-speed interconnect networks, with management projecting strong revenue and EPS growth trajectories for fiscal 2027 and 2028.

Citi's bullish thesis for Lumentum rests on three mutually reinforcing growth paths: AI cluster expansion is driving laser demand to persistently outpace supply, optical circuit switching (OCS) technology opens a larger system-level market, and near-package optics (NPO) alongside co-packaged optics (CPO) expand demand for high-power lasers and external light sources. The firm maintains a Buy rating with a $1,200 price target, implying roughly 21.3% upside from the September 9 close of $988.98 used in the report, while keeping an upside catalyst watch through October 14.

Lumentum shares have surged about 170% year-to-date, with 2025 gains reaching as high as 340%. Management recently raised its fiscal 2028 EPS target substantially to $40, driven by increasing visibility into OCS and NPO growth. Citi's latest model estimates fiscal 2027 and 2028 revenue at approximately $6.828 billion and $10.532 billion, representing year-over-year jumps of 126.5% and 54.3%, with core EPS expectations of $23.03 and $37.21 respectively.

Where the opportunity lies

Lumentum's exceptionally strong 2025 stock performance stems from its dual exposure to both the Google TPU AI computing chain and Nvidia-led AI GPU infrastructure. Whether it's Nvidia's AI GPU clusters or Google's TPU clusters—representing the AI ASIC roadmap—both depend on the optical interconnect and CPO silicon photonics high-speed capabilities that Lumentum exemplifies.

Lumentum has joined the S&P 500 and the Nasdaq-100, two core US benchmark indices, granting it passive fund allocation eligibility. This index inclusion has formally elevated Lumentum from a niche optical interconnect supply chain player to a core global tech asset, reflecting a broader AI narrative shift from single-point GPU/ASIC computing competition toward full-stack AI inference systems driven by intelligent agents. Within this transformation, data center CPUs, memory chips, and optical interconnect/communication chains may emerge as the biggest beneficiaries.

Recent strength across optical product suppliers also reflects market reports that Nvidia is adjusting its next-generation AI rack architecture. The upcoming Rubin Ultra cluster may reduce HBM configuration per rack in favor of optical interconnect architecture to link multiple AI racks. With HBM/DRAM and NAND storage prices surging, AI infrastructure investment focus appears to be shifting from single-point GPU/TPU performance and HBM capacity toward overall data center architecture efficiency, with high-speed optical interconnect emerging as a critical direction.

Shortages extending into 2027

Lumentum management views AI infrastructure as still early-stage, with computing expansion and copper interconnect physical limits jointly pushing data center high-speed networks toward optical solutions. Using data center interconnect (DCI) as a baseline, management describes scale-out network interconnect demand intensity at roughly 10x, with the migration of tightly coupled accelerator cluster interconnects (scale-up) to optical opening even stronger growth potential.

Citi forecasts module shipments roughly doubling from 2026 to 2027, with further growth in 2028. Supply shortages for electro-absorption modulated lasers (EML) and continuous wave (CW) lasers are expected to persist through 2027. New contract prices remain at previously raised levels or higher, indicating demand has not significantly retreated in response to price increases.

On the product mix front, 800G EML remains dominant, with 1.6T share expected to rise. CW capacity serves multiple purposes—filling customer gaps, securing future design wins, and supporting proprietary coherent modules—with improving yields and margins, though not yet reaching EML levels.

OCS and pump lasers represent two incremental business lines worth attention. As the largest OCS customer expands its own hardware business, required switch port counts increase. New use cases like dynamically partitioning accelerator resources and bypassing underperforming nodes have led management to view the roughly $8 billion OCS total addressable market estimate as conservative, potentially extending business visibility into 2028 and early 2029. Management uses approximately 1.5 ports per accelerator as an empirical estimate, emphasizing that fiscal 2028 earnings growth will materialize gradually rather than concentrating at year-end.

Citi analysts note pump laser demand linked to cross-cluster and cross-campus connectivity remains robust: ten consecutive quarters of growth, long-term agreements with all major network equipment manufacturers, and business volume expected to expand approximately fourfold over the next five quarters. These data points suggest Lumentum's growth sources now span laser chips, optical modules, and broader data center optical switching systems.

Supply capacity and manufacturing hurdles

Sustaining growth depends on high-performance manufacturing and customer certification. Ultra-high-power laser capacity advances in three phases: Lumentum's existing US facility continues ramping through 2027; the UK facility adds capacity from Q2 2027 extending into 2028; and a third facility targets mid-2028 startup, with customer certification cycles being the key constraint on accelerated ramp-up.

Management believes the combination of high power, low relative intensity noise, narrow linewidth, and mode-hop-free performance constitutes a scale manufacturing barrier, identifying Coherent as the only currently comparable ultra-high-power laser competitor. Regarding Chinese competitors, management states no new EML external supplier with material impact has emerged, while acknowledging the industry needs a second supply source to support 2028 demand.

The next HBM system: AI optical interconnect

NPO's expanding market presents Lumentum with cross-architecture opportunities. Optical technology leader Lightmatter projects NPO entering the market in 2027 with broad deployment in 2028, while CPO's scale-up window spans 2028–2029. Lumentum expects NPO and CPO to each account for roughly half of the optical scale-up market by 2028, with approximately three-quarters of NPO adopting external light sources to avoid the thermal environment near accelerators. Even with integrated sources, the trend favors fewer, higher-power lasers in the 120–150 milliwatt range and above.

Lumentum has secured its first CPO customer external laser source (ELS) project, with deliveries expected to begin in the second half of 2027, covering 15%–20% of that customer's related equipment deployments. NPO customer interest is broad, though no formal project wins have been announced. Meanwhile, strengthening CPO order demand has not altered certification timelines.

The testing capacity shortage Lightmatter identifies indicates commercialization still depends on volume manufacturing infrastructure. Its L20's 6.4Tbps unidirectional bandwidth and single-fiber bidirectional transmission reflect the industry's push to raise bandwidth while lowering cabling costs.

At the engineering level, optical interconnect demand growth stems from more frequent and demanding data exchange among more compute nodes: mixture-of-experts (MoE) models require cross-GPU distribution and aggregation of data; prefill-decode disaggregation requires transferring key-value caches; and complex agent workflows increase concurrency across model calls, tool execution, and data access. As clusters expand, network congestion forces expensive GPUs to wait for data, directly reducing effective tasks delivered per dollar and per watt.

High-speed electrical connection distance, loss, and power constraints therefore drive broader optical adoption. NPO/CPO shorten electrical signal paths, EML/CW/ELS provide the light sources for optical communication, and OCS supports workload-based connection reconfiguration.

The AI optical interconnect thesis centers on deployment scale and communication demand growth rather than "stronger models necessarily consuming more tokens per task"—OpenAI notably emphasizes Astra's reduced retries and lower per-task token consumption. Combined with Goldman Sachs' growth expectations for SanDisk, NAND is increasingly positioned for mass data persistence and partial cache tiering, while optical interconnect handles cross-node transport—both benefiting from AI inference systems processing massive workloads with surging data capacity and communication demands.

Ultimately, how much Lumentum delivers depends on order share, certification progress, and capacity ramp-up. Citi analysts also flag AI construction slowdowns, intensifying competition, and lower-than-expected CPO/OCS adoption as primary risks.