The next showdown in AI optical communications? Laser capacity is running low for $83 billion giant Lument…

Is the Next Battle for AI Optical Communication Coming? Laser Capacity Is Running Low, $83 Billion Giant Lument…

On October 28 local time, Bloomberg reported, citing sources, that U.S. semiconductor company Marvell is in deep acquisition talks with optical device manufacturer Lumentum, with the parties possibly announcing an agreement as early as next week. Currently, the negotiations have not reached a final outcome.

Based on current market capitalizations, Lumentum is valued at about $83 billion and Marvell at about $221 billion. If the deal is completed, it could become one of the largest mergers and acquisitions in the history of the technology and communications sectors.

After the news broke, the two companies’ stock prices diverged sharply. Lumentum rose about 11% in after‑hours trading, while Marvell fell more than 8%, as the market expressed concerns about the financing and integration pressures of such a massive acquisition.

Behind this massive deal, one question stands out. Why would Marvell, which already has leading high‑speed optical communication chip technology, spend hundreds of billions of dollars to acquire a company that makes lasers and optical components?

Why is AI starting to compete for lasers?

In the short term, a key reason is that AI data centers are rapidly increasing their demand for optical communications, and the upstream capacity for high‑performance lasers has become a major supply bottleneck in the value chain.For Marvell, securing the manufacturing capability for these critical components has direct commercial value.

In recent years, investment in AI infrastructure has mainly focused on GPUs, HBM high‑bandwidth memory, and advanced packaging. However, as model sizes and AI cluster scales continue to grow, an increasing number of compute tasks require thousands or even tens of thousands of accelerators to work together, and the data transfer rate and latency between chips are beginning to directly affect the overall system’s computational efficiency.

This shift is driving rapid expansion in the optical communication market. In a report released by market research firm TrendForce in April this year, the global AI optical transceiver module market is projected to grow from $16.5 billion in 2025 to $26.0 billion in 2026, a year‑over‑year increase of more than 57%. High‑speed optical modules such as 800G and 1.6T are the main growth drivers. The share of 800G and higher‑rate optical modules in total shipments is expected to rise from 19.5% in 2024 to over 60% by 2026.

The demand growth is also directly reflected in Lumentum’s financial reports. The company’s FY2026 revenue reached $3.014 billion, up 83.2% year over year; Q4 revenue broke the $1 billion mark for the first time, up about 109%.

However, the issue of capacity constraints is even more noteworthy than the performance growth. According to Lumentum CEO Michael Hurlston recently, driven by AI data‑center demand, some of the company’s optical component capacity orders are already booked through early 2029, and some customers have even begun reserving capacity for 2030.On certain product lines, about 70% of market demand is expected to remain unmet through 2027, and other products may still face roughly a 30% supply gap into 2028.

This round of supply tightness is mainly concentrated in upstream high‑end optical components, with lasers being a key segment.

As high‑speed optical modules evolve toward single‑channel 200 Gbps, core light sources such as EML (electrically pumped mode‑locked lasers) and CW (continuous‑wave) lasers face higher demands on output power, signal quality, stability, and manufacturing precision.

Lumentum has long‑standing expertise in these areas, especially in indium phosphide (InP) semiconductor manufacturing. Indium phosphide is a compound semiconductor material well suited for high‑speed photonic devices and serves as a key technological foundation for many high‑performance lasers today. Unlike silicon‑based logic chips with mature foundry ecosystems, the design and fabrication processes of InP lasers are tightly coupled, and adding new capacity typically requires lengthy construction and validation cycles.

Image | Indium phosphide wafers are the core material for high‑speed lasers and one of the tightest supply points in the current optical communications industry chain. (Source: Lumentum)

To ease supply pressure, Lumentum has expanded the capacity of its key plant in the Tokyo area of Japan by about 12 times over the past two years and plans to continue investing hundreds of millions of dollars. In March this year, the company also acquired a semiconductor plant in Greensboro, North Carolina, USA, with plans to convert it into an indium phosphide optical device manufacturing base, but the new facility is not expected to ramp up production until mid‑2028.

For cloud providers and equipment vendors that are building AI clusters at scale, a shortage of key components such as lasers could directly affect equipment procurement and data‑center deployment timelines.

Therefore, securing Lumentum’s laser technology and manufacturing capacity may be a key practical motivation for Marvell’s consideration of this acquisition. However, capacity constraints only explain its current market valuation. For a chip company like Marvell, Lumentum offers another layer of attraction.

From pluggable optical modules to CPO, what changes are occurring in the supply chain?

In the long term, the deeper driver pushing Marvell upstream is the continuous evolution of optical interconnect technology.As new technologies develop, the link between optical devices and electronic chips becomes increasingly tight, requiring deeper collaboration in design, packaging, and manufacturing.

Currently, data centers mainly use pluggable optical modules for device interconnect. This solution is mature, easy to maintain, and allows optical modules and network switch chips to be designed and manufactured separately, then connected via standardized interfaces.

However, as single‑channel rates advance to 200 Gbps and even 400 Gbps, the loss and power consumption of electrical signals on printed circuit boards become increasingly problematic, and the number of optical modules that a switch panel can accommodate also limits further increases in system bandwidth density.

To address these issues, the industry has begun exploring placing optical devices closer to the switching chip.

A particularly hot direction is CPO (Co‑Packaged Optics). CPO integrates the optical engine with core chips such as the switching chip in the same package system, dramatically shortening the transmission distance of high‑speed electrical signals, reducing signal loss and associated power consumption while increasing bandwidth density.

In 2025, NVIDIA launched the Quantum‑X InfiniBand and Spectrum‑X Ethernet switches based on CPO technology, and chip companies such as Broadcom are also advancing related products. At the same time, the industry is exploring other technology routes such as NPO (Near‑Packaged Optics) and LPO (Linear Pluggable Optics).

TrendForce forecasts that the CPO and NPO market size will grow from about $100 million in 2025 to over $39 billion by 2030. At the same time, the traditional pluggable optical transceiver market is still expected to remain close to $26 billion in 2030.

This means that in the coming years, multiple optical interconnect architectures will continue to coexist for the long term, but the way optical components collaborate with electronic chips may change markedly.

Figure | In the AI photonic interconnect market, CPO/NPO and pluggable optical modules will coexist for the long term. (Source: DeepTech illustration, data from TrendForce)

In the traditional pluggable architecture, electronic chips such as DSPs and optical devices like lasers can be developed by different vendors and then integrated by optical module manufacturers.

However, in new architectures such as CPO and NPO, optical devices are placed closer to the electronic chips, requiring both sides to jointly address issues of signal integrity, power consumption, thermal management, packaging precision, and reliability. This also means that chip manufacturers and optical device suppliers need to engage in deeper collaborative design early in the R&D process.

In this process, the laser remains an indispensable core component. Especially for CPO solutions that use an external laser source, high‑power, high‑stability continuous‑wave lasers are a crucial technical foundation.

Lumentum has already begun offering related products for these new architectures. In the fourth quarter of fiscal year 2026, the company disclosed a new CPO‑related order worth several hundred million dollars, expected to be delivered in the first half of 2027, and is also advancing external laser module and NPO‑related businesses.

For Marvell, combining its high‑speed communication chips and photonic interconnect technology with Lumentum’s lasers and optical manufacturing capabilities could give it greater leverage in designing next‑generation photonic interconnect systems.This also gives a potential acquisition between the two companies strategic significance beyond short‑term capacity assurance.

Why is Marvell buying Lumentum? Chip giants start upstream integration

Viewed in the context of Marvell’s business development over the past few years, this potential acquisition is also part of the company’s ongoing effort to strengthen its photonic interconnect portfolio.

As the AI infrastructure market expands, Marvell is gradually shifting its business focus toward custom computing and high‑speed connectivity for data centers.The latest financial report shows that Marvell’s data‑center revenue reached $2.172 billion in Q2 of fiscal 2027, accounting for 79% of total revenue and representing a 46% year‑over‑year increase. The company expects total revenue for fiscal 2028 to be around $20 billion, with high‑speed interconnects, network switching, and custom AI chips driving significant growth.

To strengthen these businesses, Marvell has long supplemented its technology capabilities through acquisitions.

In 2021, Marvell completed the acquisition of Inphi, gaining technology expertise in high‑speed PAM4 DSP and electro‑optical interconnects, further expanding its market in data‑center internal network interconnects (Scale‑Out) and cross‑data‑center connections (Scale‑Across).

In February of this year, Marvell completed its acquisition of photonic interconnect startup Celestial AI. According to the company’s disclosed financial documents, the total consideration for the deal was about $3.25 billion, including $1 billion in cash and roughly $2.25 billion in stock.

Celestial AI’s Photonic Fabric interconnect platform primarily targets Scale‑Up networks, using optical technology to link compute and storage resources in large AI systems so that more accelerators can work closely together.

Through these two acquisitions, Marvell has extended its photonic interconnect business from traditional data‑center networking to areas closer to the compute chip. However, it still lacks the high‑performance laser R&D and manufacturing capabilities that Lumentum possesses.

Image | Optical interconnects in AI data centers span three distance tiers. Marvell has positioned itself in the Scale‑Out and Scale‑Across tiers and is entering the Scale‑Up domain; acquiring Lumentum’s lasers and optical switching systems would give it full coverage across all three tiers. (Source: Marvell)

If it can acquire Lumentum, Marvell will gain high‑performance lasers, indium phosphide manufacturing capacity, and mature optical components and system products, further strengthening its technology portfolio in the optical interconnect supply chain.

In fact, the two parties have already engaged in technical cooperation. In March of this year, during the Optical Fiber Communication Conference (OFC) in the United States, Marvell and Lumentum jointly demonstrated an optical circuit switching solution that combines Marvell’s 1.6‑T optical communication DSP and other products with Lumentum’s R300 optical circuit switch, achieving high‑speed connections between networks of different distances.

Optical Circuit Switches (OCS) and CPO belong to different technology directions. OCS uses micro‑mirrors in micro‑electromechanical systems (MEMS) to directly change the transmission path of optical signals between different fibers, thereby reducing the need for repeated optoelectronic conversions in intermediate stages.

These technologies have already attracted the attention of major cloud vendors. Google uses an optical circuit switching architecture in its TPU clusters to adjust network connections between different compute nodes. Lumentum previously disclosed that its OCS backlog has exceeded $400 million.

From signal‑processing chips inside high‑speed optical modules to next‑generation optical engines, lasers, and optical circuit switching systems, Marvell’s and Lumentum’s products can complement each other in many ways. If the deal closes, Marvell would have the opportunity to offer a broader portfolio of technologies and products covering more stages of optical interconnect.

Similar industry moves are also emerging among other AI infrastructure giants. In March this year, Nvidia announced $2 billion investments in both Lumentum and Marvell. The partnership with Lumentum involves advanced laser procurement, R&D, and future capacity; the partnership with Marvell involves NVLink Fusion, custom AI chips, and silicon photonics technology.

In the same month, Nvidia also announced a $2 billion investment in another optical component manufacturer, Coherent, and signed a multi‑year technology and procurement agreement.

These moves indicate that high‑speed optical interconnects have become a key investment area for AI infrastructure companies. In addition to purchasing chips and networking equipment, leading vendors are establishing early relationships with upstream critical suppliers through equity investments, long‑term orders, and R&D collaborations.

However, unlike Nvidia’s approach of securing capacity through investments and partnerships, Marvell would face greater financial and integration pressures if it opts for a full acquisition of Lumentum.

First, the acquisition cost. Based on the current market value of roughly $83 billion, this would be the largest acquisition in Marvell’s history. By comparison, the company’s cash and cash equivalents in the most recent quarter were about $3.92 billion. If the deal is ultimately completed, a substantial amount of financing would be required, likely through stock issuance, debt financing, and other means.

Bernstein analysts also note that a transaction of this magnitude could impose significant financing and integration pressures on Marvell.

There are also differences in operating models. Marvell has long focused primarily on chip design, whereas Lumentum operates multiple business units, including global manufacturing facilities, indium phosphide wafer production lines, and industrial lasers. Integrating a company with large‑scale manufacturing operations is clearly more complex than acquiring a startup that concentrates on chip design.

In addition, Lumentum’s lasers and optical components currently serve multiple chip companies, optical module manufacturers, and cloud providers. The downstream of this supply chain is filled with Chinese enterprises. According to LightCounting, seven of the world’s top ten optical module vendors are from China, including Zhongji Xuchuang, Newis, and Huagong Technology, accounting for more than 60% of total revenue share.

Among them, Zhongji Xuchuang holds the top spot with about a 21% share of the global optical interconnect market, with a global market share of over 40% for 800G optical modules and an estimated 50%–70% share for 1.6T optical modules, making it one of the largest suppliers of high‑speed optical modules for Nvidia and Google.

These companies have a pronounced dependency structure on upstream products. Taking the 1.6 T optical module as an example, DSP chips account for 30% to 40% of material costs, with supply concentrated in Broadcom and Marvell, together holding more than 90% of the market share; high‑performance EML and CW lasers mainly come from a few vendors such as Lumentum and Coherent.

If Marvell’s acquisition of Lumentum is ultimately completed, the two most critical upstream components in the global optical module market will be consolidated under a single company. This concentration could prompt downstream manufacturers to accelerate the search for alternative suppliers, reshaping the industry chain structure.

Therefore, although the two companies have clear complementary opportunities in technology and products, whether this complementarity can translate into commercial returns sufficient to justify a massive acquisition remains to be seen.

Reference content:

https://www.bloomberg.com/news/articles/2026-10-28/marvell-is-said-to-near-deal-for-lumentum-in-ai-related-megadeal

https://www.sec.gov/Archives/edgar/data/0001633978/000162828026055726/lite_ex991xq4fy26.htm

https://investor.marvell.com/news-events/press-releases/detail/1031/mar

https://www.benzinga.com/node/62198187

https://www.trendforce.com/presscenter/news/20260420-13017.html

https://siliconangle.com/?p=732754

Operations/Typesetting: He Chenlong

Note: Cover/featured image generated with AI assistance