Optical module leader Lumentum (LITE.US) has announced that its order book could be fully committed through 2028, with Nvidia securing capacity by investing $2 billion directly. These developments suggest that optical components are following a similar surge pattern to HBM, positioning themselves to become the next "memory-level" growth area in AI computing infrastructure.
Lumentum's CEO Michael Hurlston stated that demand from top U.S. technology companies for its optical components is accelerating and expected to fill its production schedule through 2028. "The capital expenditures from U.S. hyperscale cloud providers are enormous and show no signs of stopping," Hurlston commented during an interview in Tokyo. "We are increasingly falling behind demand. We anticipate that within two quarters, our capacity through the end of 2028 will be completely sold out."
This forecast extends the company's previous disclosure that its capacity was sold out through at least the end of 2027. It underscores market expectations that demand for components used in AI data center equipment will remain resilient, even amid global economic disruptions.
Last month, Lumentum and competitor Coherent received a combined $2 billion investment from Nvidia. Lumentum supplies advanced indium phosphide devices designed for high-speed data transmission. As photonics technology becomes more prevalent in cloud computing clusters, its Nasdaq-traded stock has surged over 1,500% in the past year, marking a significant rebound after years of underperformance.
To address overwhelming demand, Lumentum has increased capacity at its primary factory in the Tokyo metropolitan area by twelvefold over the past two years. The company now plans to invest at least $100 million in this facility and adjacent sites. "We might even see the investment rise to $250 million," the CEO noted. "The pressure on our team to expand production is unimaginable. All U.S. hyperscale cloud providers want to know how many units we can produce this quarter, next quarter, and beyond—this is currently the industry's bottleneck."
As advanced AI hardware requires faster transmission speeds, silicon photonics technology, which uses light to transmit data, is expected to replace copper-based interconnects in data center clusters. Optical networks also help reduce heat generation and power consumption. Japan hosts several companies with expertise in indium phosphide, optical fiber materials, and silicon photonics, including Fujikura, Sumitomo Electric, and Hamamatsu Photonics.
Lumentum acquired some of its Japanese technology through the 2018 purchase of Oclaro Inc., which had inherited optical research from Hitachi. Today, its Sagamihara plant is one of the world's most advanced indium phosphide device production bases, known for stable, high-quality output. With limited space for expansion at the Sagamihara and adjacent Takao facilities, Lumentum is seeking new sites in Japan to increase production. The company aims to repurpose existing electronics manufacturing facilities to shorten the timeline for bringing new capacity online.
Hurlston mentioned that the company considered acquiring such a site in Japan last year. Moving production outside Japan is not feasible, as manufacturing indium phosphide requires precise control over variables like temperature, humidity, mixing speed, and water quality. A robust supplier network is also critical; Lumentum has signed a seven-year agreement with a Japanese supplier described as "extremely vital."
Despite heavy investments in factories, long-term supplier contracts, and hiring more engineers, Lumentum expects its order backlog to continue growing. "This momentum can't last forever—that's unrealistic—but for at least the next five years, the cycle appears sustainable," Hurlston said. "When we say 'sold out,' we mean non-cancelable agreements. This is a big deal."
A paradigm shift is occurring from "compute-centric" to "connectivity-centric" infrastructure. Understanding why optical components are crucial requires examining the evolution of AI cluster scale. Early AI tasks might have required single cards or machines, but today's trillion-parameter models demand tens of thousands of GPUs working together. In this context, intra-node computation is no longer the sole challenge; inter-node communication has become critical.
Traditional copper interconnects suffer significant signal loss and generate excessive heat at high speeds. As computing clusters scale from thousands to hundreds of thousands of cards, optical communication is no longer an optional accessory but an essential artery. Optical components—including lasers, optical chips, and modules—convert electronic signals into light signals instantly, enabling lossless, low-latency transmission across complex computing networks.
Just as HBM addresses the "memory wall," optical components break through the "communication wall." When Nvidia's chip processing speeds multiply every two years, insufficient connection bandwidth would leave expensive GPUs idle. This elevated strategic importance is the core logic behind optical components becoming a growth hotspot.
If HBM's evolution from HBM2e to HBM3e fueled its growth, optical modules advancing from 800G to 1.6T serve as an accelerator for the optical sector. As AI chip power consumption exceeds 1,000W, traditional pluggable optical modules face thermal and efficiency limits. Silicon photonics integrates photoelectric conversion directly into chip packaging, representing not just product iteration but a fundamental architectural revolution. This integrates optical component manufacturers deeply into the semiconductor ecosystem, moving beyond their traditional peripheral supplier role.
U.S. hyperscale cloud providers are engaged in a "capacity arms race." Lumentum's orders through 2028, all non-cancelable, mirror the scenario where Nvidia previously reserved HBM capacity from Micron and Samsung. When a component shifts from "procurement as needed" to "strategic reserve," its valuation premium undergoes a qualitative change.
Nvidia's strategic investment in Lumentum and Coherent is particularly noteworthy. As the "shovel seller" of the AI era, Nvidia's supply chain moves carry significant indicative value. By investing, Nvidia is essentially buying "insurance" for its GPU clusters. This sends a clear message: without light, there is no computing power.
Endorsement from top downstream customers provides optical components with high certainty. In capital markets, certainty often attracts more interest than pure growth. Optical components have evolved from peripheral communication parts into indispensable foundational pillars of the AI computing empire. For investors and industry observers, if memory was the first wave of AI computing, optical components are poised to take the baton and dominate the next cycle as a "super growth area."
Comments