Marvell Technology (MRVL.US) shares surged nearly 5% by the close of trading on Thursday, extending the chip giant's year-to-date gains to 40%. This performance significantly outpaces both the S&P 500 and the tech-heavy Nasdaq 100 index. The company, a key partner for Amazon's AWS Trainium AI ASIC series, is experiencing a fundamental re-rating driven by the powerful combination of its custom AI chip (AI ASIC) business and its leadership in optical interconnect technology, rather than by chip demand alone.
Also on Thursday, Barclays upgraded Marvell's stock rating from "Equal Weight" to "Overweight" and raised its price target substantially from $105 to $150, citing strong growth prospects for its optical interconnect and silicon photonics products. The stock closed at $119.93.
AI ASICs represent the first core pillar supporting the upward revision of Marvell's valuation and fundamentals. Recent strong earnings reports from both Marvell and Broadcom provide concrete evidence of the unprecedented growth trajectory for custom AI chips. The global generative AI boom is accelerating the development of AI infrastructure, with cloud and chip giants racing to build the fastest and most energy-efficient computing clusters for advanced data centers. Marvell and Broadcom leverage their expertise in high-speed interconnect and chip IP to collaborate with cloud leaders like Amazon, Google, and Microsoft on creating bespoke AI ASIC clusters tailored to specific data center needs. For instance, Google's TPU clusters, developed with Broadcom, exemplify this AI ASIC approach, a business that has become critically important for both companies.
Optical interconnect technology forms the second core pillar of the bullish thesis on Marvell, and is also a key reason behind NVIDIA's recent $2 billion investment in the company. For connections exceeding approximately 10 meters, traditional copper cables cannot meet the bandwidth and distance requirements of large-scale AI data centers, necessitating a shift to optical systems. Marvell's strength lies in its portfolio of optical DSPs and interconnect products, which are essential hardware for high-speed links between servers, switches, racks, and data centers. Looking further ahead, Wall Street is particularly optimistic about Marvell's leading position in silicon photonics. This technology is not a parallel alternative to optical interconnect, but rather a key enabler that pushes optical signaling from between racks and cabinets to within racks, systems, and even into chip packages themselves. More precisely, silicon photonics advances optical transmission closer to the chip, package, and I/O interfaces, which explains Marvell's strategic acquisition of Celestial AI.
Analysts are also growing increasingly optimistic about the revenue prospects for Marvell's SSD storage controller business, which serves as a third pillar for its fundamental expansion. Within AI training and inference systems, I/O bandwidth, persistent storage access efficiency, and memory pool interconnect efficiency are critical constraints on overall performance and cost. Marvell's SSD controller chips, NVMe/CXL cache controllers, and high-bandwidth memory interconnect products are vital components addressing the strong growth in storage demand. While these highly specialized control ASICs may be less conspicuous than the exponentially expanding AI ASIC business, they are crucial for data processing in massive AI models, directly enhancing system-level efficiency and service quality in data centers.
Recent Wall Street commentary indicates a shift in how the market perceives Marvell. It is no longer viewed merely as a generic data center AI chip company, but as a "dual-engine asset" that is simultaneously capitalizing on the wave of custom XPU/ASIC development by cloud giants and leading the large-scale upgrade cycle driven by exploding bandwidth demands in AI clusters, fueled by optical interconnect and silicon photonics.
Barclays analysts, led by Tom O'Malley, stated that Marvell is first and foremost a leading-edge optics company. They noted that the growth in the optics market, driven by a rapid increase in the number of data center ports, is sufficient on its own to support strong revenue expectations. Industry checks suggest the number of optical ports could at least double by 2026 and double again by 2027, they added.
Around the same time, analysts at Citigroup suggested that Marvell could expand its role in larger AI compute clusters by enabling tighter integration of core chips within a single system. The Barclays team presented calculations showing that even if some optical market share shifts to Broadcom, Marvell's overall optics-related business could still deliver robust growth of approximately 90% this year and next. They outlined a scenario suggesting earnings power of around $5 per share, even after excluding Microsoft orders, assuming no unit growth from Amazon, and applying a discount to XPU adoption rates. However, the analysts expressed greater confidence in an acceleration from Microsoft and Amazon following NVIDIA's NVLink Fusion announcement and its investment in Marvell.
These analysts concluded that the investment thesis for Marvell hinges on its ability to execute on a well-understood and highly optimistic earnings forecast, and they observe that the market's bullish narrative is increasingly shifting focus toward its optics business.
It is undeniable that future AI infrastructure, whether based on NVIDIA's GPU clusters or Google's TPU clusters (representing the AI ASIC path), will rely heavily on the optical interconnect and advanced silicon photonics capabilities that Marvell provides. The company's potential for long-term valuation premium increasingly stems from the growth logic of optical interconnect and the deeper penetration of silicon photonics technology into global data centers. This evolution would see Marvell's optical components move from board-level peripherals to becoming core interconnect elements at the rack, system, and even within-package levels. In its statement regarding the acquisition of Celestial AI, Marvell identified the next inflection point for optics as the "migration of electrical connectivity to optical, within the rack, within the system, and even within the package." Celestial's Photonic Fabric technology enables optical I/O at the package, system, and rack levels, offering higher bandwidth, lower power consumption, and longer distances compared to copper-based systems.
Trends observed at OFC 2026 indicate that hyperscale AI data centers are increasingly believing in the "power of light." The main theme in the AI compute supply chain is clear: the bottleneck for hyperscale AI data centers is shifting from single-chip compute power to "cluster-level optical interconnect bandwidth, power efficiency, and scalability." This explains the recent wave of bullish analyst reports on Marvell, with the most optimistic price target on Wall Street reaching as high as $160.
AI superclusters, whether TPU or GPU-based, share a common requirement: the need for extremely high-bandwidth, low-latency, and high-efficiency internal data center interconnects. Traditional copper cables or electronic switches become impractical when scaling to thousands or tens of thousands of chips due to exploding power consumption and heat dissipation. Optical interconnect technologies—including Co-Packaged Optics (CPO), silicon photonic switches, and Optical Circuit Switching (OCS)—along with more advanced silicon photonics-based optical I/O, use light signals instead of electrical ones. This significantly enhances bandwidth density and energy efficiency while reducing latency and power consumption in large-scale AI training and inference networks. This demand for greater optical interconnect capacity is common to both GPU and TPU clusters.
As NVIDIA's AI cabinet clusters evolve alongside scale-up/scale-out high-performance networks, 1.6T technologies, and CPO, the hardware bottleneck is increasingly shifting from the "compute chip itself" to "optical-level bandwidth, power consumption, and signal integrity." NVIDIA's announcement in March of a $2 billion investment in Marvell and a strategic partnership aimed at connecting Marvell into NVIDIA's AI factories and AI-RAN ecosystem via NVLink Fusion underscores a joint commitment to accelerating the adoption of optical interconnect and silicon photonics technologies in data centers.