Optical Modules: The “Data Couriers” of the AI Era
2026-07-14Beginner
2026-07-14Beginner
Optical modules were once among the least noticed parts in communication networks, but by 2026 they have become an unavoidable “critical chokepoint” in the AI computing race. Swelling demand, tight supply, and rapid technology iteration—these three threads intertwine to form the most realistic picture of the optical module industry today. As of July 2026, the industry stands at a key juncture where “orders are booked through 2028” and “technology roadmaps are undergoing intense iteration” in parallel.
According to Yole Group, the global optical module market is set to grow from $23.4 billion in 2025 to $112.3 billion in 2031, at a 30% CAGR, with AI training and inference infrastructure as the core driver.【1】LightCounting further predicts that the 2026 global datacom optical module market is expected to reach approximately $22.8 billion, of which 800G and 1.6T together will account for about 64%.【2】
What Are Optical Modules
An optical module is essentially an optoelectronic converter. On the transmitting side, it converts electrical signals into optical signals and sends them through optical fibers; on the receiving side, it converts optical signals back into electrical signals. It sounds simple, but it is an indispensable part of fiber-optic communication—routers, switches, servers, and storage devices all rely on it for interconnection.
Structurally, an optical module consists of a transmitter optical subassembly, receiver optical subassembly, optical interface, circuit board, and other components. By transmission rate, they range from 1G, 10G, 100G, 400G, up to today’s 800G and 1.6T; by packaging, there are SFP, QSFP, QSFP-DD, and other types. The higher the rate and the more complex the packaging, the higher the technical barrier.
For decades, optical modules played a “behind‑the‑scenes role”—they quietly existed in data centers and communication networks, moving data from point A to point B, with few people paying them special attention. But in the last two years, that has completely changed.
Why the Market Is Suddenly Fixated on Optical Modules
The answer is simple: AI.
AI model training and inference are essentially a process of “stacking compute power.” Thousands of GPUs are connected together to form massive computing clusters. These GPUs need to constantly exchange data—training a trillion‑parameter model involves astronomical data transfer volumes. Copper cables suffer from too much signal attenuation and too short a distance at high speeds, so optical fiber becomes the only choice. Optical modules are the “interfaces” for fiber‑optic communication.
Different Approaches of Six Companies
Yole Group data shows the global optical module market growing from $23.4 billion in 2025 to $112.3 billion in 2031, at a 30% CAGR, with AI training and inference infrastructure as the core driver. LightCounting’s forecast is more specific: the 2026 global datacom optical module market is expected to reach $22.8 billion, with 800G and 1.6T together accounting for roughly 64%. TrendForce adds that, driven by Google’s high‑speed interconnect architecture, the global shipment share of optical modules above 800G will exceed 60% in 2026.
Demand‑side signals are also clear. Alphabet’s 2026 capex plan is as high as $175‑185 billion; Amazon is projected at about $200 billion; Microsoft’s single‑quarter capex has already reached $34.9 billion. Much of this money flows into data centers, servers, and networking equipment—and one of the core components of networking equipment is optical modules.
There are many players in the optical module industry chain, but each occupies a different position and follows a different logic.
Marvell (MRVL): Sells “connectivity” itself. Marvell does not make GPUs or CPUs; it makes high‑speed optical DSPs—the signal‑processing chips inside optical modules. In the data‑center optical module market for 400G and above, Marvell’s optical DSP holds roughly a 70% share. In March 2026, Marvell launched the industry’s first 1.6T ZR/ZR+ pluggable optical module and 2nm coherent DSP. NVIDIA invested $2 billion in Marvell in March 2026, incorporating it into the NVLink Fusion ecosystem. Marvell’s logic is straightforward: the more AI data centers are built and the larger they become, the stronger the demand for data flow, and any data flow cannot bypass optical DSP.
Lumentum (LITE): Stands further upstream—optical chips. In the third quarter of fiscal 2026, Lumentum reported revenue of $808 million, up about 90% year‑over‑year. Growth came mainly from two things: record shipments of 200G EML laser chips, and optical circuit switch (OCS) orders beginning to ramp, with order volume exceeding $400 million. More notably, Lumentum is a supplier of NVIDIA’s UHP lasers, and its CPO (co‑packaged optics) products are expected to begin volume shipments in the second half of 2026. Optical chips represent the highest technical barrier in optical modules, and NVIDIA’s simultaneous $2 billion investments in both Lumentum and Coherent in March 2026 were essentially aimed at securing upstream supply.
Coherent (COHR): Takes a “full‑chain” approach. Coherent’s business covers the entire chain from materials to complete optical systems. In the third quarter of fiscal 2026, the company reported revenue of $1.58 billion, with data center and communications business accounting for as much as 75.4%. In the CPO space, Coherent’s management emphasizes that its capabilities go far beyond CW lasers and ELS modules, including VCSELs, isolators, thermoelectric coolers, PICs, photodetectors, and a range of other components—meaning its participation depth in the CPO value chain far exceeds market expectations. On the capacity front, Coherent plans to double its indium phosphide device production by the third quarter of 2026, and double it again by 2027.
Applied Optoelectronics (AAOI): Smaller in scale but growing fast. In the first quarter of 2026, AAOI reported total revenue of $151 million, up 51.4% year‑over‑year. Data center revenue reached $81.4 million, more than doubling year‑over‑year. The company is actively expanding capacity: its first 100,000‑unit‑per‑month 800G/1.6T production lines have already achieved volume shipments, and it expects to boost capacity to 650,000 units per month by the end of 2026. AAOI management expects 800G to become the largest revenue source for its data center business, and demand is projected to continue exceeding capacity until mid‑2027.
Ciena (CIEN): Unlike the previous players, Ciena is more system‑oriented—optical networking equipment and data‑center interconnect solutions. In the first quarter of 2026, revenue was $1.43 billion, up 33% year‑over‑year, a record high. The company has raised its fiscal 2026 revenue guidance to $5.9‑6.3 billion. Ciena’s highlight lies in distributed AI training scenarios—hyperscale customers are spreading compute power across multiple sites to overcome power and space limitations, requiring high‑speed optical networks to connect cross‑region AI training clusters. The company expects its hyper‑rail standardization to start by the end of 2026 and roll out more broadly in 2027.
Lumentum and Coherent hold positions upstream in optical chips and lasers; Marvell dominates the mid‑stream optical DSP space; AAOI is rapidly catching up in module manufacturing; and Ciena focuses on system‑level interconnect solutions. These six companies represent different slices of the optical module industry chain, from materials to systems.
“White‑Haired Stock God” Serenity’s Investment Logic in Optical Modules
In discussions of this optical module rally, one cannot avoid the anonymous researcher named Serenity. He has over 700,000 followers on X, uses a white‑haired anime avatar, and is dubbed the “White‑Haired Stock God” in Chinese investment circles. His approach is not to bet on core compute players like NVIDIA, but to extend along the AI industry chain both upstream and downstream, looking for overlooked “pick‑and‑shovel” plays—optical modules, liquid cooling, power, copper interconnects. In less than half of 2026, his stock investment returns are said to have reached 45‑fold.
Serenity’s core methodology is called the “bottleneck theory” by his followers. In his view, in every wave of AI compute expansion, true excess returns do not come from the headline leaders in the spotlight, but from peripheral players that control the “allocation rights of physical constraints.” His investment logic can be summed up in one sentence: follow the compute industry chain, find the tightest supply bottleneck at the moment, and then overweight that position.
This methodology has been fully validated in the optical communications sector. In early 2025, Serenity focused on AAOI, LITE, and AXTI, when their market capitalizations were roughly $2 billion, $26 billion, and $500 million, respectively. By mid‑2026, their market caps had soared to $15.37 billion, $74.47 billion, and $7.24 billion, respectively. Among them, AXTI rose from $12 to over $70—a classic case of the bottleneck theory: indium phosphide substrates are the underlying material for optical chips, and optical chips are the core of optical modules; Serenity captured the most upstream and scarcest link in this chain.
In terms of specific names, Serenity’s coverage heavily overlaps with the six companies above. His earliest research that gained wide attention focused on COHR, LITE, Eoptolink, and Zhongji InnoLight (a Chinese optical module maker); ALAB is also one of his representative holdings, having recorded multi‑fold gains from 2024 to 2025. For MRVL, Serenity lists it as a long‑term favorite but also admits market concerns about MRVL’s ASIC share being squeezed by Broadcom. For AAOI, Serenity maintains a strong bullish stance, believing the market’s pessimistic expectations are misguided—the core logic revolves around structural supply‑demand gaps in the AI infrastructure chain: high‑end laser and optical module capacity is constrained, while AMD and large cloud vendors continue to see rising demand for high‑speed optical interconnects. He expects AAOI could reach approximately $560 million in annualized revenue by 2027, with some ramp‑up cycles potentially extending to 2028. For LITE, Serenity notes that its market cap grew from about $3 billion in 2024 to over $65 billion, driven primarily by NVIDIA‑induced EML laser supply bottlenecks and optical network architecture changes—a classic demonstration of his bottleneck theory.
Serenity also played a significant role in the CPO technology route debate. In June 2026, research firm SemiAnalysis released a report that triggered a sell‑off in CPO concept stocks, with AAOI, LITE, COHR, MRVL and others pulling back by high‑single‑digit to double‑digit percentages. Serenity subsequently pushed back publicly, arguing that SemiAnalysis relied too heavily on conservative engineering models and underestimated NVIDIA’s ability to compress hardware cycles. He cited industry chain data showing that Foxconn had raised its CPO‑related product registrations, that NVIDIA’s optical switches had already shipped ahead of schedule, and that CPO scale‑up would accelerate in the second half of 2026. In his view, rather than relying on some institutions’ pessimistic judgments, one should trust the accelerated mass‑production timelines repeatedly reiterated by industry players. After these views were published, the related sector rebounded.
Serenity’s influence has already spread from overseas social platforms to the A‑share market. Public data shows that his calls have directly triggered 20% rallies in A‑share companies, and his research framework has attracted attention and tracking from some quantitative funds. Whether or not the market agrees with his views, Serenity has become an undeniable “variable” in this optical module rally—he has used the bottleneck theory to push optical modules from behind the scenes to center stage, and has helped more people understand that AI compute competition ultimately comes down to breaking through one physical limit after another.
Several Trends Worth Watching
The optical module industry is undergoing several structural changes that affect the competitive dynamics of each of the above companies.
First, product iteration is accelerating. The ramp‑up of 1.6T optical modules is occurring even before 800G has reached its peak shipment volume. This means product life cycles are shortening, and the contribution of high‑end products to revenue and profit is being realized earlier. For upstream optical chip and DSP suppliers, this is a logic of both volume and price growth; for module makers, it means ongoing R&D pressure and capacity challenges.
Second, CPO is moving from concept to reality. At GTC Taipei in May 2026, NVIDIA for the first time pushed its CPO‑based Spectrum‑X Ethernet Photonics switch into production. CPO will not completely replace traditional pluggable optical modules—at least in “scale‑out” deployment scenarios, transceivers will remain the preferred solution—but it is indeed changing the value distribution of optical communications. The value of optical communications is expanding from single modules to the entire AI network.
Third, upstream supply is becoming a bottleneck. The certainty of optical chip supply directly determines optical module share. EML lasers, DSPs, indium phosphide substrates, and other segments all face varying degrees of supply tightness. NVIDIA’s successive investments in Lumentum and Coherent, Marvell’s accelerated capacity expansion, and AAOI’s plan to increase indium phosphide laser capacity by 350%—all these moves share the same logic: whoever secures upstream supply will have the upper hand in the next phase of competition.
Optical modules were once among the least noticeable parts in communication networks. But in the AI era, they have become a “toll station under construction” on the “computing highway.” This transformation has come quickly, and it is far from over.
References
-
Optical Transceivers for Datacom & Telecom 2026, https://www.dqindia.com/semiconductors/from-23-billion-to-112-billion-in-six-years-ai-is-driving-a-new-growth-cycle-in-optical-connectivity-12123090
-
Zhongji InnoLight (300308) 2025 Annual Report, https://s.askci.com/stock/financialreport/300308/financefile/
You May Also Like
From "Shiso Leaf" to 45‑Fold Myth: Decrypting the Alternative Investment Philosophy of Serenity
In May 2026, the anonymous account "Serenity" posted a 4502.45% annual return, earning the title "White‑Haired Stock God" and rapidly surpassing 750,000 followers on X. His core investment philosophy can be summarised as the "Shiso Leaf" theory and the "Chokepoint" theory – not chasing giants, but deeply cultivating irreplaceable "bottleneck" links in the industry chain, using public information to uncover undervalued assets. His holdings are concentrated in global small‑ to mid‑cap tech stocks in photonics, semiconductor substrates, and power semiconductors. CoinW has listed AI‑theme tokens such as TAO, RENDER, and FET, but no token exclusive to him. Risks to note include his unverified identity, post‑surge pullbacks, and high volatility in crypto assets.
Analysis of the U.S. Equity AI Industry Chain Investment Logic
In 2026, the U.S. equity AI investment logic is shifting from concept speculation to earnings delivery. A capital expenditure super-cycle, led by hyperscale cloud providers, has taken shape, with total annual CapEx expected to exceed $700 billion, securing order visibility for the industry chain over the next 12–24 months. Within the three‑tier structure of the industry chain, compute infrastructure (Nvidia, Broadcom, etc.) offers the highest certainty; the foundation model layer still faces unclear profitability paths; and the application software layer benefits from dual optimization of revenue and costs. Investment opportunities are spreading sequentially across compute, storage, optical communications, and power supply. CoinW has launched its TradFi zone, supporting trading in U.S. equities such as Nvidia and Google, as well as AI‑theme tokens including TAO, RENDER, and FET. Risks to watch include elevated valuations, slowing CapEx growth, and geopolitical factors.
Gold Plummets 27% in Record Drop: Constructing All-Weather Hedging Strategies with Crypto and Traditional Assets Amid Geopolitical Conflict
Extreme liquidity crises shatter the myth of absolute safe havens in single assets. As geopolitical conflicts breach gold's traditional defenses, embracing the agility of asset tokenization (RWA) and the non-correlation of native crypto networks has become the ultimate margin of safety for building a cross-cyclical, all-weather hedging portfolio.
All
Articles