LightCounting复盘英伟达光互连投资，未披露相对3月官方公告的新交易事实

By Carol Cao, Bob Wheeler and Vladimir Kozlov
April 16, 2026
At GTC 2026, Nvidia unveiled its Rubin series solutions, anchored by the Vera Rubin platform. Through full-stack system-level optimization, the company further consolidated its technological leadership. Nvidia’s strategic goal for the inference era is unambiguous: to solidify its position as the “Token King” by delivering computing solutions at the lowest per-token cost, while building out a comprehensive AI factory infrastructure. During the conference, Nvidia stressed that global demand for AI computing power far exceeds supply, requiring broad capacity expansion across the entire interconnect ecosystem.
Amid tight supplies of key materials and continuous technological advancement, Nvidia has launched a series of targeted strategic investments. By locking in core manufacturing capacity and empowering its ecosystem partners, the company seeks to further consolidate its leading position in global AI computing and build a more resilient, future-proof foundation for sustained industry leadership.
As Nvidia celebrates its success in the AI boom, Apple marks its 50-year anniversary. There are many similarities between the two companies, including approaches to the supply chain management. Financial Times recently published a long article on how Apple managed their suppliers: “Apple at 50: the roots of a tech revolution.” It is a fascinating story. It starts with the quality control movement in post-war Japan led by Dr. W. Edwards Deming, followed by successes at Sony, Panasonic and Sharp, until Apple “swallowed them whole and turned them into sub-contractors.” Apple worked tirelessly with a wide range of suppliers, developing new technologies and perfecting manufacturing processes. Many suppliers were “burned” in the process. Apple challenges the not-so-glamorous details of “close partnerships with suppliers” disclosed by the Financial Times. Let us hope that an article titled “Nvidia at 50: the roots of another tech revolution” to be published in 2043 will be more glamorous.
Our industry also offers an example of Apple’s supply chain management style. Until recently, Coherent and Lumentum were leading suppliers of VCSEL arrays for 3D face recognition in Apple’s iPhones. Lumentum originally developed these products for Microsoft’s gesture recognition systems, which did not take off. Apple picked up Lumentum’s VCSEL arrays to launch 3D face recognition in the iPhone X and added Finisar (now Coherent) as a second source. In 2017, Finisar received a $390 million investment from Apple to establish a new 6” GaAs wafer production line for VCSELs in Sherman, Texas. The two suppliers competed fiercely, pushing the prices down to the delight of their customer and all iPhone users. Sony was qualified as the third supplier about 5 years ago and took most of the business away from Coherent and Lumentum.
Sales of VCSEL arrays peaked above $700 million in 2021 and declined steadily to below $450 million by 2025. Steep price declines offset modest growth in unit shipments, as detailed in our annual report on optical sensors. Has any of the suppliers made a profit? Probably not. Have they produced fantastic products in huge volumes at very low cost? Absolutely!
Coincidentally, Coherent’s fab in Sherman, Texas, is part of its new $2 billion partnership with Nvidia, including a 6” InP wafer production line for high-power edge emitting lasers. From GaAs to InP; from $390 million to $2 billion; from Apple to Nvidia; out of the frying pan and into the fire? We hope not, since we should be learning something from history.
Lumentum also received a $2 billion investment from Nvidia. These are nonexclusive arrangements with Nvidia that include multibillion-dollar purchase commitments and guaranteed future access to production capacity for advanced laser components and high-performance optical networking products.
Coherent and Lumentum accumulated a wide range of optical technologies starting from materials to devices and systems. In addition to InP lasers, Nvidia will probably also be interested in VCSELs, specialty fibers and optical switches made by the two suppliers. It is very likely that we will see new devices and systems developed in this collaboration. As Julie Eng, CTO of Coherent, mentioned in her plenary talk at OFC: “optics is inside the machine now.” Nvidia will need a wide range of optical technologies to keep “its machines” running faster than those of competitors.
In contrast with Apple, Nvidia has a very concentrated customer base: over 50% of its data-center revenue comes from the top 5 customers: Amazon, Google, Meta, Microsoft, and Oracle. All of these companies are also experts in supply chain management. Yes, Nvidia is starting to sell more AI systems to neoclouds and enterprises, but it will take time to diversify its customer base. In the meantime, Nvidia is walking along a fine line of delivering very high-performance systems based on proprietary technologies to customers prioritizing open and competitive ecosystems of suppliers. Somehow Nvidia needs to be part of this ecosystem, and it is searching for a way in.
For example, Nvidia started offering Ethernet switches in addition to InfiniBand, clearly responding to customer preferences for open source solutions. The company also recently joined OCI-MSA as a founding member, yet its initial approach to CPO is very different from the first set of specifications proposed by the new MSA. Nvidia’s recent $2 billion investment in Marvell is another step towards offering the proprietary NVLink technology via partnerships.
Full text of the research note is available to LightCounting clients at: https://www.lightcounting.com/login