As AI compute clusters scale from tens of thousands to hundreds of thousands of GPUs, the bottleneck limiting training efficiency is shifting from GPU supply to data movement capabilities. Optical interconnects—long considered a niche, foundational hardware segment—have suddenly moved to the core of AI infrastructure investment. A Lumentum earnings report showing a staggering 90% revenue surge has revealed just the tip of the supply-demand imbalance in photonic components, forcing the market to re-examine where hyperscale cloud providers are truly directing their capital expenditures.
From "Copper Walls" to Optical Interconnects: What’s Transforming AI Data Centers?
The internal data traffic patterns of AI training clusters are fundamentally reshaping the design logic of data center network architectures. Traditional data centers primarily handle north-south traffic between users and servers. In today’s AI clusters—scaling to tens or even hundreds of thousands of GPUs—over 80% of data exchanges occur as east-west communication between xPU chips. This traffic pattern demands bandwidth and latency performance far beyond any previous data center generation.
Within the industry, this physical constraint is often called the "copper wall"—once single-channel data rates exceed 800 Gbps, copper connections hit their ceiling in terms of bandwidth, signal integrity, and power consumption. NVIDIA CEO Jensen Huang has repeatedly stated that next-generation AI infrastructure will rely heavily on optical connectivity, as traditional electrical interconnects can no longer meet the demands. This view is being validated by upstream order data and capital spending trends across the supply chain.
While optical interconnects themselves aren’t new, their role inside data centers is undergoing a dramatic transformation. Previously, optical modules mainly handled long-haul links between data centers. Now, optical interconnects are penetrating inside server racks, between switches, and even down to the chip packaging level. The technology roadmap—from pluggable optical transceivers to optical circuit switches (OCS), and on to co-packaged optics (CPO)—is evolving much faster than the industry anticipated. Lumentum’s 90% revenue growth in its fiscal Q3 2026 is a direct reflection of this structural shift.
What’s Driving Lumentum’s 90% Growth? Data Breakdown and Structural Drivers
On May 5, 2026, Lumentum reported its fiscal Q3 2026 results for the quarter ending March 28. Quarterly revenue reached $808 million, up 90.1% year-over-year and 21.5% sequentially—a company record for single-quarter revenue. Non-GAAP gross margin jumped from 35.2% a year ago to 47.9%, an increase of 1,270 basis points. Non-GAAP operating margin hit 32.2%, up 2,140 basis points year-over-year. Earnings per share came in at $2.37.
| Metric | Value | YoY Change |
|---|---|---|
| Revenue (USD millions) | 808.4 | +90.1% |
| Non-GAAP Gross Margin | 47.9% | +1,270 bps |
| Non-GAAP Operating Margin | 32.2% | +2,140 bps |
This margin expansion is no accident; it’s driven by three structural factors: a rising share of high-margin laser chips and optical circuit switch revenue; indium phosphide production lines running at full capacity, diluting unit fixed costs; and surging AI data center demand granting suppliers greater pricing power.
Breaking down by business segment, optical components generated $533 million in revenue, up 77.3% year-over-year; modules and systems brought in $275 million, up 121.1%. In components, shipments of 200 G EML laser chips hit record highs, and narrow linewidth lasers for data communications grew over 120% year-over-year, marking nine consecutive quarters of sequential growth. On the systems side, cloud transceiver shipments grew 40% sequentially, and OCS orders entered a ramp-up phase.
It’s worth noting that despite these blowout results, Lumentum’s stock saw a moderate pullback in after-hours trading following the earnings release. This isn’t uncommon for high-growth tech stocks at earnings season, typically reflecting profit-taking after a strong run, valuation concerns, or cautious interpretation of next-quarter guidance. The company’s Q4 2026 revenue guidance is $960 million to $1.01 billion, with the midpoint implying about 22% sequential growth. Whether this guidance is met will be a key indicator of sustained optical interconnect demand.
Value Redistribution in the Optical Interconnect Supply Chain: The Competitive Landscape of EML, OCS, and CPO
Value in the optical interconnect supply chain is not distributed evenly along a "smile curve"—it’s highly concentrated in upstream core photonic components. To understand the structural forces behind Lumentum’s results, we need to trace profit flows from raw materials to final systems.
EMLs (Electro-Absorption Modulated Lasers) are indispensable in all high-speed optical transceivers, converting electrical signals into stable, modulated light for fiber transmission. In hyperscale data centers deploying 800 G and 1.6 T transceivers, Lumentum holds about 50–60% of the global EML market. In the most technically demanding segment—200 G EML lasers—Lumentum’s global market share is around 90%.
Such dominance brings not only scale, but pricing power. There’s a global shortfall of about 25–30% in high-end optical chip capacity, with no quick fix in sight. All Lumentum product lines are supply constrained. The supply gap for indium phosphide substrates exceeds 70%, and prices have soared from $800 to $2,500 per wafer.
| Photonic Component | Lumentum Market Position | Core Application |
|---|---|---|
| 200 G EML Laser Chip | ~90% global share | 800 G/1.6 T AI data center transceivers |
| 400 G/800 G EML | ~60% global share | High-speed data center interconnect |
| OCS Optical Circuit Switch | NVIDIA exclusive certification | AI cluster cross-connects |
| Narrow Linewidth Laser | 9 consecutive quarters of growth | Data Center Interconnect (DCI) |
| CPO Laser Component | NVIDIA UHP supplier | Next-gen CPO optical engines |
OCS (Optical Circuit Switch) is another growth engine for Lumentum. Unlike traditional electrical switches that require "optical-electrical-optical" conversion, OCS routes signals entirely in the optical domain, eliminating the latency and power draw of conversions. Lumentum’s OCS uses MEMS technology, consuming less than 10% of the power of traditional packet switches and with latency measured in tens of nanoseconds. The company’s OCS backlog exceeds $400 million, and single-quarter revenue could reach $100 million by the end of 2026.
CPO (Co-Packaged Optics) is widely viewed as the next evolution in optical interconnects post-2026. By directly packaging the optical engine with the switch chip, CPO shortens signal transmission distance by about 70% and can cut power consumption by roughly 30%. Lumentum is a supplier for NVIDIA’s UHP lasers and will begin volume shipments of CPO products in the second half of 2026. While CPO’s penetration in overall optical transceivers remains low for 2026 (industry estimates around 0.5%), it represents a clear long-term growth trajectory.
Market Consensus and Debate: The Real Story and Controversies in Optical Communications
Market discussions around the optical interconnect sector reveal both clear consensus and sharp disagreements.
On consensus, industry and research institutions broadly agree that optical interconnects are becoming the core bottleneck for AI infrastructure. At Photonics West 2026, Yole Group noted, "The limiting factor for AI scale-out is data movement, not compute." Silicon photonics, CPO, and advanced photonic packaging are shifting from "emerging" to "strategic" technologies. LightCounting forecasts a 65% year-over-year increase in the Ethernet optical module market in 2026, and Lumentum itself projects the total addressable AI optical communications market to soar from about $18 billion in 2025 to $90 billion by 2030, a CAGR of around 40%.
On disagreements, two main threads emerge. First is the debate between CPO and pluggable transceiver technology roadmaps. While CPO is viewed as the long-term direction, Tower Semiconductor’s Q1 2026 earnings call suggested pluggable optical transceivers will continue to dominate data center interconnects at least through 2030. LPO (Linear-Drive Pluggable Optics) technology poses a potential substitute for CPO in terms of power and cost, and its commercialization is progressing faster than previously expected.
Second is the issue of valuation digestion and growth sustainability. Despite Lumentum’s outstanding results in both absolute terms and growth rate, its stock price reaction post-earnings was cautious, even signaling some profit-taking. This reflects the market’s need to digest earlier rapid gains in the optical sector, and ongoing questions about whether growth rates can be maintained. The key issue: As the upgrade cycle shifts from 800 G to 1.6 T, and as NVIDIA’s large-scale procurement peaks, how quickly can Lumentum’s next growth drivers take over?
How Hyperscale CapEx Is Reshaping Demand for Photonic Components
To understand the sustainability of optical interconnect demand, one must track the capital expenditure trajectories of hyperscale cloud service providers. Since 2026, major North American cloud providers have repeatedly raised their CapEx guidance. According to TrendForce, the world’s nine largest CSPs now expect combined 2026 CapEx of about $830 billion, with annual growth accelerating from 61% to 79%. Amazon plans to invest roughly $200 billion in data center expansion and AI chip deployment, while Microsoft has raised its 2026 CapEx target to $190 billion, up about 130% year-over-year.
Morgan Stanley further estimates that the top five hyperscale cloud providers will spend about $800 billion on CapEx in 2026, rising to $1.16 trillion in 2027—ushering in a "trillion-dollar era" for AI infrastructure. Within this massive spend, the share allocated to optical interconnect equipment is steadily rising. The logic: As AI clusters scale from tens of thousands to hundreds of thousands of GPUs, the number of interconnect endpoints grows much faster than compute nodes, naturally increasing optical interconnect’s share of total network costs.
NVIDIA’s strategic moves further validate this trend. In March 2026, NVIDIA invested $2 billion each in Coherent and Lumentum—a total of $4 billion—to secure advanced laser component and optical network product capacity. During the same period, NVIDIA committed up to $3.2 billion to Corning to build three new optical manufacturing plants in the US. These investments are not just financial partnerships; they reflect leading AI chipmakers’ intent to lock in optical interconnect capacity—supplier manufacturing capability is now a key bottleneck for AI cluster expansion.
Conclusion
Photonic components are shifting from peripheral hardware in AI data centers to a core bottleneck. The logic is clear: demand for data movement is outpacing compute growth by a wide margin. As cluster sizes grow from thousands to tens or hundreds of thousands, compute performance can still ride Moore’s Law and advanced packaging, but the physical limits of copper—bandwidth ceilings and power constraints—are far more rigid. This "compute is stackable, data is hard to move" asymmetry cements photonic components as a strategic, irreplaceable asset in AI infrastructure.
Looking ahead, the second half of 2026 through 2027 will be a pivotal window as the optical interconnect sector transitions from 800 G to 1.6 T. The pace of cloud CapEx deployment, upstream optical chip capacity expansion, and CPO industrialization will jointly shape value distribution across the sector. For readers focused on structural opportunities in AI infrastructure, key variables to watch include: the rollout of NVIDIA’s Rubin platform, TSMC’s COUPE platform mass production, and actual CapEx execution by North American cloud providers.
FAQ
What role do photonic components play in AI data centers?
Photonic components convert electrical signals into optical signals and transmit data between chips, directly determining the compute utilization efficiency of AI clusters.
What’s the core driver behind Lumentum’s 90% revenue growth?
Explosive demand from AI data centers for 200 G EML laser chips and OCS optical circuit switches, coupled with sharply increased CapEx from hyperscale cloud providers.
How severe is the supply-demand gap for EML laser chips?
There’s a global shortfall of about 25–30% for high-end EML chips. In the 200 G EML segment, where Lumentum holds about 90% market share, the shortage is even more acute.
When will CPO replace pluggable optical modules at scale?
The industry consensus points to 2027–2028, but competition from LPO solutions and uncertainties in yield ramp-up could affect the timeline.
Why is NVIDIA investing in optical interconnect suppliers?
NVIDIA needs to secure capacity for advanced laser components and optical networking products, as optical interconnect is becoming a critical bottleneck for AI cluster expansion.
What is the projected CapEx for hyperscale cloud providers in 2026?
The top five hyperscale cloud providers are expected to spend about $800 billion in 2026, rising to $1.16 trillion in 2027.
What factors will determine the pace of the 800 G to 1.6 T transition?
The speed of upstream optical chip capacity expansion, inventory management strategies of major end customers, and actual growth in AI training workloads will all play a role.
What’s the biggest investment risk in the optical interconnect sector?
A slowdown in compute demand growth could lead to a CapEx plateau, or technology roadmap competition could dramatically reshape upstream value distribution.
