Artificial intelligence (AI) has been transforming the business landscape at an unprecedented rate. The market researchers at Dell'Oro Group expect global capital spending on AI data centers to surpass $1 trillion by 2029. Nvidia (NVDA 2.02%) is all set to be one of the key beneficiaries of this trend. The company has generated almost $146.6 billion in data center revenues over the past four quarters. According to IoT Analytics, Nvidia accounts for nearly 92% of the data center GPU market.
Although Nvidia's AI chips are the backbone of the global AI infrastructure buildout, they also require custom chips and high-speed networking infrastructure to run complex high-performance computing and AI workloads.

Image source: Getty Images.
That's where Marvell Technology (MRVL -1.53%) comes in. This semiconductor company builds custom silicon (XPUs, controllers), optical interconnects, and networking platforms that link thousands of GPUs inside large-scale AI clusters. As demand for faster and more efficient AI data centers accelerates, Marvell is quietly becoming one of the most critical players in this AI revolution.
Custom silicon business
With AI data centers upgrading network bandwidth from 400-gigabits (400G) to 800-gigabits (800G) or 1.6-terabits (1.6T), demand for Marvell's high-speed optical interconnects, custom switches, and digital signal processors (DSPs) has accelerated dramatically. The business momentum is evident, as the company's data center revenue surged 69% year over year to $1.49 billion in the second quarter of fiscal 2026 (ended Aug. 2). Data center business makes up nearly 74% of Marvell's total revenue.
Marvell's growth is being driven by the company's expanding footprint in custom AI chips and optical connectivity solutions. The company has already secured 18 custom chip design programs, which encompass both XPUs (encompassing all types of main chips, including CPUs, GPUs, and AI accelerators) and XPU-attached chips (companion chips, such as memory controllers and interconnects). Several of these chip designs are already in volume production to be deployed in major hyperscaler architectures.
Marvell expects its data center target addressable market (TAM) to reach $94 billion by 2028, representing a 26% increase from its prior estimates. Based on its existing chip design programs, the company is aiming to grow its data center market share to 20% of the TAM in 2028. Additionally, the company also has over 50 active design projects in development. These programs are estimated to generate $75 billion in lifetime revenues.
Scale-up networking
Marvell is also focusing on scale-up networking, a high-bandwidth, low-latency system that connects GPUs, CPUs, or accelerators within a server or across multiple servers in a data center. This enables them to act like a single computing system, making them suitable for training and running complex models in next-generation AI data centers.
Marvell's optical interconnect products and technologies play a crucial role in helping data centers transition from copper to fiber, enabling higher network speeds, lower latency, improved energy usage, and reduced signal loss. Marvell has started volume shipments of its 1.6T Pulse Amplitude Modulation (PAM) DSPs. This highlights its technical and commercial readiness in high-speed electro-optical interconnects.
The company is also investing in the development of custom switches aligned with the client's network protocol for scalable networking. Interconnects and switches are expected to present a significant opportunity for Marvell.
Financial performance
Marvell's balance sheet and profitability metrics highlight strong operating momentum. In the second quarter, revenue rose 57.6% year over year to over $2 billion, while operating margin was up 870 basis points to 34.8%.
The company carried cash of $1.2 billion and $4.5 billion in debt on its balance sheet. The company has also completed a $2.5 billion all-cash divestiture of its automotive Ethernet business. These proceeds provide the company with the flexibility to invest in growth initiatives and also return capital to shareholders.
Valuation
Marvell shares trade at 25.6 times forward earnings, which is more comparable to a semiconductor company than to a core AI infrastructure player. However, this valuation multiple could expand significantly if margins improve dramatically in the coming months. This is possible since increasing sales of custom AI silicon and high-speed interconnects are pushing up margins. The company also has high revenue visibility, due to its 18 design wins and over 50 projects in development.
Analysts expect Marvell's earnings per share to grow 78.2% year over year to $2.8 in fiscal 2026 (ending Jan. 31, 2026) and by 20.6% to $3.37 in fiscal 2027. Revenue is expected to grow 40.9% to $8.1 billion in fiscal 2026 and 16.6% to $9.5 billion in fiscal 2027.
Such rapid top- and bottom-line growth could translate into significant share price appreciation in future years. The stock could then begin to witness growth similar to that of Nvidia during its significant growth phase post 2020.