Marvell acquires Lumentum: The first deep closed loop of "chip + optics"

On October 28, institutional reports said Marvell was in advanced talks to acquire optical device manufacturer Lumentum in an all-stock transaction. The report at the time put Lumentum's market value at about US$83 billion and Marvell's at about US$221 billion. The potential deal raises a noteworthy question: why would a company known for custom chips and high-speed interconnects make a major push into laser and optical device manufacturing? On October 30, semiconductor industry podcast Semi Doped invited Andrew Schmitt and Scott Wilkinson, two analysts at market research firm Cignal AI, to discuss the deal. Their judgment is that Marvell hopes to combine DSP, lasers and optical components through vertical integration to form a more complete product system. If the deal is completed, Marvell could become one of the few companies covering multiple links simultaneously: DSP, lasers and optical modules. But it would also create new competitive relationships: while supplying chips to optical module makers, it could enter those customers' markets. I. From DSP to lasers and optical modules: Marvell wants to connect the entire chain The host suggested that after acquiring Lumentum, Marvell could become a company with DSP, laser and optical module businesses all at once. Andrew Schmitt believes the closest precedent for this model is Acacia, but the Marvell-Lumentum combination could go further in business scale and coverage. He first clarified that Marvell does not lack laser supply. The company currently mainly purchases tunable lasers for ZR coherent modules and maintains good relationships with multiple suppliers. The core driver of acquiring Lumentum is vertical integration. Andrew believes that in a market where technology changes rapidly, companies need to master both chip and optical technologies to better control product development and the supply chain. Putting the two types of technology inside the same company enables collaborative design, better-performing products, and more gross profit by selling more complete solutions. Lumentum itself already has a similar business combination. Andrew cited the company's earlier acquisition of Cloud Light, which brought it into the data communications optical module market based on silicon photonics technology. Optical modules produced by Cloud Light use Lumentum's own lasers and also require PAM4 DSPs purchased from companies such as Marvell and Broadcom (AVGO.US). In an optical module, the laser and DSP account for almost the most important part of the value, with the remaining steps mainly assembly. This means Marvell and Lumentum's businesses already had upstream-downstream links. After the acquisition, Marvell would have the opportunity to bring key components that previously required external procurement into its internal system and further integrate DSP, lasers and optical engines. Andrew believes the value of this model is also reflected in profit. In the past, chip companies might have gross margins of 50%, while optical component makers had only 25% to 30%. Chip companies were naturally unwilling to acquire a company with lower gross margins, fearing it would dilute their own profitability. But he said optical component industry gross margins have now risen from about 25% to close to 50%. Lumentum has scarce technologies in products such as 200G EML, optical circuit switches (OCS) and high-power lasers, and its bargaining power has clearly strengthened. "If you can make more money, even if that money accounts for a smaller share of total revenue, what does it matter?" Andrew believes investors previously placed too much emphasis on gross margin percentage and ignored the gross profit amount a company can create. Integrating components into more complete products may create more profit even if the combined gross margin percentage changes. He also mentioned that Marvell might have been able to acquire Lumentum for only about US$3 billion six or seven years ago, while the potential deal size now could be 30 times that. This is his recollection and estimate, but it also reflects the change in the value of optical components in the AI era. II. Both "referee" and "player": Marvell may compete with customers Vertical integration would give Marvell a more complete product line and also change its relationship with customers. Scott Wilkinson pointed out that Marvell's current coherent DSPs are widely used by optical module makers. Take 400G coherent modules as an example: apart from Acacia, Ciena (CIEN.US), Nokia (NOK.US) and some Chinese manufacturers, many companies purchase Marvell's DSPs. If Marvell acquires Lumentum and further enters the optical engine and module market, it could compete directly with these existing customers. This is precisely a delicate point in the deal: Marvell is both an upstream chip supplier in the industry chain and could become a participant in downstream products. Customers will inevitably consider whether they should continue to rely on a company that competes with them to supply key chips. Host Austin raised a direct question: if customers therefore turn to other DSP suppliers such as Broadcom and Credo (CRDO.US), would Marvell be willing to sacrifice part of its chip market share in exchange for the optical module business? Andrew believes this is a business calculation that can be quantified: "Through vertical integration, I can capture more value than selling components separately, even if it means I may sell fewer components." Scott also cautioned that what really needs to be watched is how Marvell manages the business after the acquisition. Lumentum already sells lasers to external customers and also operates some optical module business. After the acquisition, Marvell may not keep all of that capacity for itself. Andrew believes whether customers leave depends on whether Marvell can give existing DSP customers and Lumentum's optical customers enough confidence that existing supply relationships will not be affected. This is also the cost of vertical integration: companies gain more control over the industry chain, but may also prompt customers to look for new suppliers. III. "Winner takes all" in optical communications: first-mover advantage is no match for million-level deliveries Beyond vertical integration, the two analysts also discussed another distinct competitive rule in the optical communications industry: the market for new technology standards is ultimately often dominated by a small number of manufacturers. Andrew calls this phenomenon the "IEEE lottery." Whenever the industry launches a new standard, usually five or six companies develop products at the same time. A few years later, often only two companies truly account for most shipments, together taking 80% to 90% of the market share. Therefore, being first to launch a product is important, but whether a company can seize the opportunity in the early stage of market formation determines whether it can become the final winner. Scott further pointed out that competition in the current AI optical communications market places particular emphasis on production capacity. "If you are first to market but can only produce 10,000 units, it actually does not mean much. If you are second to market but can produce one million units, then the orders will go to you." He cited Chinese manufacturer Eoptolink (Accelink Technologies) as an example. The company entered the 800G Ethernet optical module market slightly later than some competitors but became one of the world's major suppliers thanks to its capacity advantage. Scott believes a similar competitive situation is emerging in 1.6T products. This is also why Lumentum is drawing attention. Scott mentioned that the company has continued to expand indium phosphide (InP) manufacturing capacity over the past few years, with capacity doubling through multiple rounds, but market demand has grown even faster, making it still difficult to meet all customer demand. Lumentum's advantage also lies in having some products that competitors find difficult to replicate quickly. Andrew specifically mentioned 200G EML, OCS and high-power lasers. He cautioned that many companies can claim they can develop similar products, but the real test is whether they can scale from producing a few hundred units to 50,000 units or even one million units. "Making 100 and making one million are completely different things." In the AI era, this scale advantage is further amplified. Andrew believes the value of optical components has also changed: in traditional telecom networks, lasers mainly help operators increase transmission capacity; in AI data centers, 1.6T interconnects may increase AI cluster training speed by two to three times. When an optical technology can significantly improve the efficiency of an entire computing cluster, the price customers are willing to pay for it will naturally change. IV. Will Nvidia (NVDA.US) join the bidding? At the end of the interview, the two analysts discussed whether Nvidia could become a potential buyer. Nvidia has previously invested US$2 billion each in Lumentum and Coherent (COHR.US) and established strategic cooperation with both companies. Andrew believes Nvidia also has an incentive for vertical integration: its earlier acquisition of Mellanox was aimed at incorporating Network-1 Technologies, Inc. into its own GPU systems. Extending further into optical components and OCS would also fit that logic. Andrew also raised a possibility: if the news of Marvell's talks with Lumentum was intentionally leaked, it might be signaling other potential buyers and ultimately triggering a bidding war. However, he explicitly stated that he has no inside information on the deal and that this is only personal speculation. Nvidia's previous investments and procurement cooperation also cannot prove that it intends to participate in an acquisition. For Marvell, the core test of this deal ultimately remains how to balance two things: gaining higher overall value through integrating chip and optical technologies, while preventing existing customers from turning to other suppliers because of competitive relationships. Andrew Schmitt's judgment is that optical components have become a higher-value link in AI infrastructure, and the economic logic of chip companies acquiring optical companies is changing. Scott Wilkinson cautioned that in optical communications, a market prone to "winner takes all," capacity and scale determine final share. Marvell wants to connect the industry chain from chips to optics. Whether it can simultaneously protect customers, technology and market share will determine how much value this integration can ultimately create. This article is reposted from "Wallstreetcn," author: Pan Lingfei, GMTEight editor: Zheng Yuyang.