Lumentum Orders Booked Through 2028: Can Optical Communications Replace Storage as the New AI Computing Engine?

TradingKey - April 10, 2026, optical communications leader Lumentum ( LITE) released news that ignited the market: Driven by accelerating demand from U.S. mega AI data centers, the company's order book is expected to be full through 2028, with 2028 capacity projected to be sold out within two quarters. Despite expanding production as quickly as possible, it is still struggling to meet demand. More notably, Nvidia ( NVDA) had previously made a strategic investment of $2 billion in Lumentum, coupled with billions of dollars in capacity procurement commitments. On the same day, Lumentum's shares rose 4.8% pre-market as J.P. Morgan sharply raised its price target from $565 to $950. Buoyed by this, the 'Big Three' of China's A-share optical module sector—Zhongji Innolight, Eoptolink, and TFC Optical Communication—all traded higher. Zhongji Innolight's market capitalization surpassed 800 billion yuan and Eoptolink's topped 500 billion yuan, both hitting new all-time highs.

All signs indicate that optical components are replicating the explosive growth trajectory of HBM, poised to become the next frontier in AI computing infrastructure.

I. Why have optical interconnects become the chokepoint for AI computing power expansion?

To understand why optical communication is becoming the next major investment theme, one must first recognize that the physical bottleneck of AI computing clusters is shifting from 'computing power' to 'transmission capacity.' Currently, GPUs still command the highest value proportion in AI servers, accounting for approximately 80% to 90% of the total system cost; however, as GPU computing power grows exponentially, data interconnection between GPUs is becoming the new bottleneck.

As data transmission rates evolve toward 100G/200G per lane, traditional copper interconnects have hit their physical limits in bandwidth, power consumption, and effective transmission distance, making optical interconnects a vital evolutionary direction for raising the ceiling of AI computing power. A research report from Shenwan Hongyuan explicitly points out that network equipment is leaping from traditional 800G to 1.6T speeds, with optical interconnects becoming the critical path for enhancing AI performance limits. Michael Hurlston, CEO of Lumentum, stated bluntly: 'Capital expenditures by U.S. hyperscale cloud providers are massive and show no signs of ending. We are increasingly falling behind demand.'

This assessment has been fully corroborated across the industry chain. In the first quarter of 2026, the optical module sector saw an overall increase of nearly 30%, with several leading stocks doubling in value, establishing it as a clear 'alpha' sector alongside memory chips. From an earnings perspective, companies in computing infrastructure segments such as optical modules, PCBs, and probe cards generally achieved growth of over 100%, which, together with memory chips, validates the accelerated transmission of AI computing demand through the hardware supply chain.

II. Replicating the Storage Sector Logic: Optical Communication is Undergoing a "Prosperity Spillover" Scenario.

The core logic behind the 2025 memory sector super cycle is "high prosperity spillover": AI training drives the priority allocation of capacity to high-margin products like HBM, crowding out traditional DRAM/NAND. This, combined with rising broad-spectrum demand from inference, creates a supply-demand gap that triggers a double rerating of both price and valuation.

Optical communications is replicating a similar logic, comprising four core components:

1. "Structural surge" on the demand side

To support AI training and inference, North American cloud service providers are maintaining high capital expenditures. East-west traffic between GPUs within data centers is exploding, causing demand for 800G and 1.6T optical modules to grow exponentially. Meanwhile, the evolution of large AI models from training to inference is broadening computing power demand, driving optical communication needs to spread from leading cloud providers to the entire industry. As of March 2026, the daily average token call volume for large AI models in China has increased more than a thousand-fold compared to the beginning of 2024, igniting comprehensive demand ranging from computing chips to optical modules.

2. "Rigid constraints" on the supply side

The bottleneck in the optical module supply chain lies in the long capacity expansion cycles for high-end optical chips (EML, VCSEL). Lumentum possesses scarce technological barriers in fields such as Indium Phosphide (InP) components—200G EML is a core component of 1.6T optical modules, and Lumentum is one of the few manufacturers capable of reliable mass delivery. To meet demand, Lumentum increased its Tokyo plant capacity 12-fold within two years and plans to invest at least another $100 million; the CEO admitted to immense pressure regarding expansion. This supply rigidity is the underlying logic of the memory sector's "price and valuation double rerating" seen last year.

3. "Lock-in effect" with major customers

Nvidia's $2 billion investment in Lumentum, coupled with billions of dollars in capacity commitments, is essentially a strategic move to secure positions in bottleneck segments. This is highly similar to how SK Hynix's HBM capacity was locked in by major customers—the scarcest segments gain the highest pricing power and earnings elasticity.

4. Dual drivers of rising volume and price

Compared to other segments, optical communications possesses a steeper growth curve for both volume and price. In the future: first, the performance improvement of individual AI chips will double interconnection requirements, multiplying the number of optical components per chip; second, higher performance requirements for optical chips will lead to a continuous increase in the unit prices of high-performance products.

III. Accelerating Technological Iteration: From 800G to 1.6T, and the Re-rating of CPO

Optical communication has emerged as a major industry tailwind, driven by an unprecedented shortening of technology iteration cycles. According to the China Academy of Information and Communications Technology (CAICT), AI has reduced the iteration cycle for optical modules from 3–4 years to approximately 2 years. Currently, 800G has become the mainstream for data centers, 1.6T is entering commercial volume production, and 3.2T is expected to undergo validation in 2027 and 2028.

More significant is the revolutionary upgrade in technology roadmaps. To address heat dissipation and power consumption pressures in AI clusters, the industry is actively exploring Co-Packaged Optics (CPO) and Optical Circuit Switches (OCS). CPO packages optical engines with computing chips to significantly reduce power consumption and latency, while OCS provides efficient interconnects for large-scale GPU clusters through dynamic routing in the pure optical domain. The trend toward all-optical switching has been validated by leading cloud service providers; Lumentum disclosed in FY2026Q2 that its OCS backlog exceeds $400 million, with a projected shipment CAGR of over 150% from 2025 to 2028.

The acceleration of technology iteration is directly reflected in market forecasts. Lumentum projected at OFC 2026 that the Total Addressable Market (TAM) for AI optical communications—covering pluggable optical modules, CPO, and OCS—will surge from $18 billion in 2025 to $90 billion by 2030, a CAGR of 40%. LightCounting also forecasts that the global optical module market will maintain a 60% growth rate in 2026, reaching a market size of nearly $60 billion by 2031.

IV. What are the key risks to monitor at present?

Optical communications have broad prospects, but directly benchmarking the sector against the storage segment still carries non-negligible risks.

Uncertainty in technological pathways: Whether CPO will completely replace pluggable optical modules remains the primary point of contention. Some argue that the two can coexist—with CPO focusing on the growth of intra-rack interconnects, while pluggable modules remain the mainstream for inter-rack connections over the next two to three years. However, the rapid iteration of technological routes means the competitive landscape can shift at any time, and the risk of ecological exclusion is real.

Complexity of the competitive landscape: The global optical module market follows a "dual-center" pattern between China and the U.S. American firms Lumentum and Coherent maintain a technological edge in high-end optical chips, while Chinese companies such as Zhongji Innolight, Eoptolink, and TFC continue to gain market share through their supply chain advantages and capacity expansion capabilities.

Valuation and positioning risks: After two years of gains, leading optical module stocks are trading at historically high valuations. Zhongji Innolight's 2025 net profit is projected at approximately 10.8 billion yuan, with a P/E ratio of about 60x; Eoptolink's P/E is around 40x; and TFC's P/E exceeds 100x. Simultaneously, shareholding concentration is high, and the stability of short-term profit-taking and trend-following positions is weak, making them prone to intensified volatility due to shifts in market style or negative news. Finding a balance between long-term AI prosperity and short-term valuation pressure is currently the central theme.

V. Conclusion

Optical communications indeed possess a solid foundation to become the next "super trend" in AI computing: on the demand side, interconnection bottlenecks in AI computing clusters are driving exponential growth in optical communication requirements; on the supply side, the capacity constraints of high-end optical chips are replicating the "spillover prosperity" pricing logic of the memory sector; and technical iteration—from 800G to 1.6T to CPO—is accelerating, reshaping value distribution.

However, unlike the "price elasticity" of the memory sector, the growth logic of optical communications is more driven by a "volume-price surge"—the core being the doubling of optical interconnect demand per unit of computing power and higher unit prices for high-performance products, rather than simple price fluctuations. Rapid technical path iterations also introduce greater uncertainty regarding the competitive landscape and investment returns.

Lumentum's order book being filled through 2028 is a landmark signal that AI computing infrastructure has reached the "interconnect bottleneck" stage. As cluster expansion becomes increasingly dependent on breakthroughs in optical interconnect technology, the upcycle for the optical communications supply chain is lengthening rather than shortening. Yet, AI infrastructure investment requires long-term validation of returns; whether optical communications can truly succeed memory depends on the delivery of technical evolution, capacity matching demand, and a benign competitive environment. For investors, balancing high growth prospects with high valuations is more critical than merely chasing the boom.