
Lumentum Holdings more than doubled quarterly revenue as spending on artificial intelligence infrastructure translated into sharply higher demand for the optical components and networking systems that move data through large GPU clusters. Net revenue reached $1.006 billion in the fiscal fourth quarter ended June 27, up 109.3% from $480.7 million a year earlier and 24.5% from the preceding quarter.
The result highlights a less visible constraint in the AI buildout. Adding more accelerators does not help much if thousands of processors cannot exchange data quickly enough. As clusters become larger and denser, bandwidth, latency and power consumption inside the network become part of the computing problem itself. Lumentum says that shift is pulling optical links closer to the GPUs and switches, expanding demand beyond the traditional connections between rows of servers and data centers.
In its fourth-quarter earnings release, Lumentum said data-center architects are increasingly turning to optical links as a primary means of connectivity. The company pointed to optical circuit switching, its cloud module business and early adoption of 1.6-terabit products as growth drivers that are beginning to contribute, while demand for ultra-high-power lasers and newer co-packaged and near-packaged optical architectures suggests optics is starting to move into in-rack connectivity.
AI clusters are making the network part of the compute bottleneck
A large AI cluster is designed to operate less like a collection of independent servers and more like a tightly coordinated computing system. GPUs working on the same training or inference workload have to move large amounts of data among one another, often with very low latency. As the number of accelerators rises, the volume of traffic across the fabric grows with it, making the interconnect increasingly important to how efficiently the expensive compute hardware can be used.
Lumentum describes three broad layers of that connectivity. Scale-up networks link accelerators that need very tight synchronization, scale-out networks connect much larger numbers of endpoints across a data center, and scale-across links extend between buildings or campuses. The technical requirements differ, but the common pressure is rising bandwidth density. Electrical connections remain important, yet the distance they can carry very high data rates efficiently becomes more restrictive as speeds rise, which is one reason optical technology is moving closer to switch chips and accelerators.
That transition shows up in the products Lumentum is emphasizing. High-speed pluggable transceivers use lasers to send data over fiber, and the move toward 200-gigabit-per-lane technology supports 1.6-terabit modules. Co-packaged optics, or CPO, and near-packaged optics, or NPO, shorten the electrical path by placing optical engines closer to the switching silicon. Optical circuit switches can create direct light paths through a network, reducing the need for some intermediate electrical conversions and helping large systems manage bandwidth and power more efficiently.
The opportunity is therefore broader than selling more conventional transceivers as hyperscalers add data centers. Lumentum is positioning lasers, modules and optical switching as parts of the internal fabric of future AI systems. Management said the first signs of that change are already visible in demand for ultra-high-power CPO lasers, an initial external-laser-source module order and a wider set of NPO engagements. Those are still emerging product areas, so the size and timing of their contribution remain uncertain, but they explain why the company sees the addressable optical market expanding as GPU clusters scale.
Revenue growth came with a sharp improvement in operating margins
The demand increase was broad across Lumentum’s product mix. Components revenue rose 102.7% from a year earlier to $649.4 million and represented 64.5% of quarterly sales. Systems revenue climbed 122.6% to $356.9 million. For the full fiscal year, total revenue increased 83.2% to $3.014 billion, with components up 79.7% and systems up 90.7%.
Margins improved at the same time. GAAP gross margin reached 47.4%, compared with 33.3% a year earlier, while GAAP operating margin rose to 27.8% from a 1.7% operating loss. On the company’s adjusted basis, gross margin increased to 50.4% from 37.8% and operating margin rose to 36.6% from 15.0%. Non-GAAP net income was $326.3 million, or $3.23 per diluted share, compared with $63.3 million, or $0.88 per share, in the year-earlier quarter.
The quarter also produced an unusually large GAAP net loss that needs to be separated from the operating trend. Lumentum reported a $7.2 billion GAAP loss, or $84.65 per diluted share, after recording a one-time, noncash $7.8 billion loss on debt extinguishment tied to the equitization of certain convertible notes. The accounting charge also drove a $6.9 billion GAAP loss for the full fiscal year. It does not change the revenue or operating-margin expansion, but it makes the GAAP bottom line a poor shorthand for the quarter’s underlying operating performance.
Lumentum ended the fiscal year with $2.7 billion of cash, cash equivalents and short-term investments. That was $433.9 million lower than at the end of the third quarter, but $1.9 billion higher than a year earlier. The stronger balance of liquidity comes as the company prepares to spend more on the manufacturing capacity needed for the next stage of optical demand.
Capacity is becoming strategic as Lumentum forecasts another revenue step-up
The supply side of the optical market is becoming strategically important enough that major AI companies are helping secure future production. In March, NVIDIA announced a multiyear agreement with Lumentum that includes a $2 billion investment, a multibillion-dollar purchase commitment and future capacity access rights for advanced laser components. NVIDIA said optical interconnect technology and package integration are critical to scaling AI factories because they can improve the energy efficiency and resiliency of large networks.
Lumentum is also expanding its own manufacturing footprint. The company acquired a 240,000-square-foot facility in Greensboro, North Carolina, that it plans to retrofit for production of indium-phosphide devices, including continuous-wave and ultra-high-power lasers used in advanced optical systems. It expects the facility to begin ramping production in mid-2028 and has said it plans to invest hundreds of millions of dollars there over several years. NVIDIA is expected to be a customer, while Lumentum also intends to supply other AI-infrastructure customers.
The long lead time for that factory illustrates why the networking bottleneck is not simply a question of product design. Optical demand ultimately depends on wafer capacity, manufacturing yields, qualification and the ability to produce lasers and related components at high volume. Lumentum has said it expects to add more than 50% to its EML laser unit capacity by the end of calendar 2026 compared with the end of 2025, while other laser capacity is scheduled to arrive later. If AI network architectures adopt more optics per accelerator, manufacturing execution will determine how much of that demand can be converted into revenue.
For the first quarter of fiscal 2027, Lumentum expects revenue of $1.225 billion to $1.275 billion. The midpoint of $1.25 billion would be about 24% above the fiscal fourth-quarter level, even after the latest quarter’s 24.5% sequential increase. The company is also forecasting a non-GAAP operating margin of 39.5% to 40.5% and adjusted diluted earnings of $4.05 to $4.35 per share.
That guidance makes the September quarter the next concrete test of whether the optical networking cycle can keep accelerating at the pace suggested by Lumentum’s fourth-quarter results. The AI infrastructure story is still led by GPUs, but the company’s numbers show that the network connecting those processors is becoming a larger part of the spending equation. As optical links move deeper into racks and closer to the compute itself, Lumentum’s growth increasingly depends on both demand for those architectures and its ability to manufacture enough of the components they require.
Latest News
View all news- U.S. Existing-Home Sales Fall 1.7% in July as Median Price Reaches $434,100
- U.S. Auto Loan Balances Rise to $1.71 Trillion as Consumer Debt Pressures Persist
- Bank of America Launches $250 Billion U.S. Infrastructure Finance Initiative
- Super Micro Forecasts Up to $72 Billion in FY2027 Sales as AI Demand Accelerates
- CoreWeave Revenue More Than Doubles as AI Backlog Hits $104 Billion