As AI data centers increasingly believe in the power of light, the "optical interconnection + silicon photonics" engine ignites a new bull market for Marvell (MRVL.US).

Focus on the customization of AI chips (AI ASIC) and optical interconnection technology for large AI data centers, as well as Marvell Technology, Inc.(MRVL.US), one of the largest partners for Amazon.com, Inc.'s AWS Trainium series AI ASICs, saw its stock price rise nearly 5% as of the Thursday close of the US stock market. This chip giant's stock price has surged by as much as 40% this year, significantly outperforming the S&P 500 index and the Nasdaq 100 index, known as the "tech stock barometer". Marvell Technology, Inc. is driven not by a singular chip demand theme, but by the resonance of two hardcore logics: "AI ASIC/XPU + optical technology". Also on Thursday in the US Eastern Time, Wall Street financial giant Barclays PLC Sponsored ADR raised its rating on the stock due to the growth prospects of optical interconnection/silicon photonics products from hold to buy, while also significantly raising the target price of the stock from 105 USD to 150 USD. As of the Thursday close of the US stock market, Marvell Technology, Inc.'s stock price closed at 119.930 USD. AI ASIC is undoubtedly the first core pillar of valuation and fundamental growth for Marvell Technology, Inc., and the strong financial reports from Broadcom Inc. and Marvell highlight the unprecedented growth logic of AI ASICs being rapidly confirmed by "financial evidence". The global wave of generative AI has accelerated the development of AI chips for cloud computing and chip giants as they strive to design the fastest and most efficient AI computing infrastructure clusters for advanced large-scale AI data centers. Broadcom Inc. and its main competitor Marvell Technology, Inc. mainly focus on leveraging their absolute advantages in high-speed interconnection and chip IP domains to collaborate with cloud computing giants such as Amazon.com, Inc., Alphabet Inc. Class C, and Microsoft Corporation to jointly create AI ASIC computing clusters tailored to the specific needs of their AI data centers. This ASIC business has become a very important business for both companies, with Broadcom Inc.'s TPU AI computing cluster in collaboration with Alphabet Inc. Class C being a typical line of AI ASIC technology. High-growth logic of ASIC + optical business, bullish research reports from Wall Street towards Marvell Optical interconnection is another core pillar of the bullish outlook for Marvell Technology, Inc. stock and its expanded fundamentals, as well as a key factor in NVIDIA Corporation's 20 billion USD investment in the company. When the connection distance exceeds about 10 meters, copper cables are no longer able to meet the bandwidth and distance requirements of large AI data centers; at this point, it is necessary to turn to optical interconnection systems, with the core hardware in optical devices being a series of optical DSPs and optical interconnection products that Marvell excels at. Optical interconnection focuses on high-speed links between servers, switches, racks, and data centers. Looking ahead, Marvell's leading silicon photonics technology is also a key factor in Wall Street's positive outlook on the company's stock price, and silicon photonics is not an alternative concept parallel to optical interconnection, but more like a crucial technology that pushes optics from cabinet exteriors and rack interconnections further into rack interiors, system interiors, and even package interiors. Specifically, silicon photonics technology leans towards pushing optical transmission towards the chip/packaging/I/O side, which is why Marvell chose to acquire Celestial at a high cost. Additionally, analysts are increasingly optimistic about Marvell's SSD storage chip controller business revenue prospects, which can be considered the third pillar of significant expansion expectations for the company's fundamentals. In large-scale training/inference systems, I/O bandwidth, persistent storage access efficiency, and interconnect efficiency within memory pools also constrain overall training costs and performance. Marvell Technology, Inc.'s SSD controller chips, NVMe/CXL cache controllers, and high-bandwidth storage interconnect product lines are important components of the strong growth in storage demand. Though these highly specialized control ASICs do not grab as much attention as the exponential expansion of the AI ASIC business, they are crucial for the data-centric processing of ultra-large parameter AI models, directly driving system efficiency and service quality at the data center level. Recent sentiments from Wall Street indicate that the market no longer sees Marvell Technology, Inc. as just a general data center AI chip company but rather as a key player in the wave of self-developed XPU/ASICs by major cloud computing giants while also leading in the explosive expansion of AI computing clusters based on optical interconnection and silicon photonics development. Lead by analyst Tom O'Malley, Barclays PLC Sponsored ADR analysts have expressed that Marvell is primarily a cutting-edge optical technology company, and with the rapid growth in data center port numbers, the growth in the optical market itself is enough to support the company's strong revenue expectations. Analysts further stated that industry research indicates that the number of optical level ports is expected to at least double by 2026 and double again by 2027. Almost simultaneously, analysts from another major Wall Street firm, Citi, indicated that Marvell Technology, Inc. could expand its role in larger-scale AI computing clusters by "connecting core chips more closely within a single system". Our calculations show that even if some of the market share in the optical market transitions to Broadcom Inc., the overall generic optical business this year and next could still bring Marvell about 90% strong growth. We have listed a scenario below: even if Microsoft Corporation orders are completely removed, assume Amazon.com, Inc. does not experience any unit growth, and discount the XPU utilization rate, we can still achieve an earnings per share of about 5 USD. However, we do not agree with these assumptions. In fact, after NVIDIA Corporation announced the NVLink Fusion and the latest decision to invest in Marvell, we are more confident in the accelerated growth of Microsoft Corporation and Amazon.com, Inc., stated the Barclays analyst team led by O'Malley. These analysts point out that the market's investment logic for Marvell will boil down to whether they can execute a logic that is already well understood and extremely optimistic in performance predictions. They believe that the market's bullish narrative is increasingly shifting towards the optical business on a larger scale. Data centers are increasingly believing in the power of light! Marvell, with "optical interconnection + silicon photonics" in hand, is on its way to a new bull market trajectory. Undeniably, whether it be the AI GPU computing infrastructure clusters led by NVIDIA Corporation or the TPU clusters from Alphabet Inc. Class C (TPU representing the AI ASIC technology line), both rely on Marvell Technology, Inc.'s optical interconnection capabilities and cutting-edge silicon photonics abilities. Marvell's medium-to-long-term valuation premium increasingly comes from the growth logic at the optical interconnection level and the deeper penetration of silicon photonics technology into major data centers globally - as it means Marvell's optical can migrate from board edge devices to rack level, system level, and even chip internal packaging level interconnect core components. Marvell Technology, Inc. directly stated in its statement when it acquired Celestial AI in late 2025: the next milestone for optical technology is the "transition of electrical connections to optics within racks, systems, and even packaging", and Celestial's Photonic Fabric can bring optical I/O technology to the package, system, and even rack level, providing higher bandwidth, lower power consumption, and longer distance connections based on optical interconnection systems compared to copper. From the overall data center optical interconnection development trends at OFC 2026, large-scale AI data centers are increasingly convinced of "the power of light". The main line of the AI computing industry chain is very clear: the bottleneck of large-scale AI data centers is shifting from single-chip computing power to "cluster-level optical interconnection bandwidth, energy consumption, and scalability". This is why Wall Street analysts have collectively issued bullish research reports on Marvell Technology, Inc. lately, with the most optimistic target price on Wall Street even reaching 160 USD, higher than Barclays' latest expectations. AI super clusters like TPU and GPU compute clusters share a common core characteristic that they require extremely high bandwidth, low latency, and high energy efficiency internal interconnect within data centers. Traditional copper cables or electronic switching solutions explode in power consumption and heat dissipation when scaled to thousands or even tens of thousands of chips; optical interconnection technologies (primarily inclusive of co-packaged optical CPO, silicon photonics switches, and optical circuit switches OCS) as well as more advanced silicon photonics-level optical I/O technologies can use light signals to replace electrical signals, significantly improving bandwidth density and energy efficiency while reducing latency and power consumption in large-scale AI training/inference networks, which is the common requirement for both GPU and TPU clusters. With NVIDIA Corporation's dominant AI cabinet computing clusters and the development of scale-up/scale-out high-performance networks, 1.6T and even CPO technologies, hardware bottlenecks are increasingly shifting from "core chip computing power itself" to "optical-level bandwidth, power consumption, and signal integrity." NVIDIA Corporation announced its 20 billion USD investment in Marvell Technology, Inc in March, unveiling a strategic partnership aimed at linking Marvell to NVIDIA Corporation's AI factories and AI-RAN ecosystem via NVLink Fusion, highlighting NVIDIA Corporation's intention to work hand in hand with Marvell Technology, Inc. in advancing data center optical interconnection and silicon photonics technologies.