infrastructure investment is reshaping the semiconductor market, with a focus shifting from raw compute performance to interconnect standards and platform integration.
The top 10 global fabless IC design houses reported total revenue of $359.4 billion in 2025, up 44% year-over-year. This growth is driven by continued investment in AI infrastructure, including GPU purchases and deployment of in-house ASICs.
NVIDIA maintained its leading position, with record revenue of $205.7 billion—a 65% increase from the previous year. The company's fourth-quarter data center revenue accounted for 90% of its total revenue, highlighting its dominance in AI computing.
A key development is NVIDIA's recent $2 billion investment in Marvell. The collaboration will focus on customized XPUs, scale-up interconnect architectures based on NVLink Fusion, and optical interconnect and silicon photonics technologies. This partnership suggests that competition in AI infrastructure is expanding beyond compute performance to include interconnect standards and platform integration capabilities.
The AI networking sector is evolving from a supporting role focused on server connectivity to a critical foundation that determines the efficiency and scalability of AI clusters. Broadcom, ranking second with $39.7 billion in revenue, benefited from growth in custom silicon and AI networking products, reflecting the increasing value of customized AI chips and broader network architectures.
AMD reported over 30% year-over-year growth in data center revenue in 2025, lifting its total revenue to $34.6 billion. This performance underscores AMD's emergence as an alternative supplier alongside NVIDIA in the AI server market, driven by growing demand for open ecosystems.
MediaTek took fifth place with strong shipments of its flagship Dimensity 9500 chipset, driving full-year revenue to a record $19.1 billion. Marvell ranked sixth, with revenue surpassing $8 billion, up 43% year-over-year. This growth was supported by the rapid adoption of AI data center connectivity, custom silicon, and interconnect technologies.
Realtek's fourth-quarter revenue declined to $847 million due to seasonality, but strong pull-in demand in the first half supported its full-year revenue at $3.9 billion. OmniVision saw a quarterly decline but achieved full-year revenue of $3.31 billion, driven by increased camera adoption in advanced driver-assistance systems (ADAS) and strong demand for action and panoramic cameras.
Novatek reported nearly $3.23 billion in revenue, up 1% year-over-year, while MPS ranked tenth with a 26% increase to $2.79 billion, supported by demand for AI and server-related power management solutions.
The partnership between NVIDIA and Marvell is a significant development, indicating that the market is moving beyond raw compute performance to focus on interconnect standards and platform integration. This shift could have profound implications for the future of AI infrastructure, potentially leading to more efficient and scalable AI clusters.
For power users and enterprise buyers, this partnership suggests a new era in AI computing, where the efficiency and scalability of AI clusters will be determined by interconnect standards and platform integration capabilities. While NVIDIA continues to dominate the market, the growing presence of Broadcom, AMD, and Marvell highlights a competitive landscape that is evolving rapidly.
