Marvell Technology (NASDAQ:MRVL) Chairman and CEO Matt Murphy used a COMPUTEX keynote in Taiwan to argue that connectivity is becoming the defining constraint for artificial intelligence infrastructure, as AI systems scale from thousands toward potentially millions of processors.
Murphy said AI performance is often framed around processors, process nodes and memory bandwidth, but those metrics do not fully capture the challenge of building large AI systems. “Computing at this scale is fundamentally a connectivity challenge,” Murphy said, adding that the industry is already in a transition from copper to optical connections that will “unlock new architectural possibilities.”
Marvell Highlights Shift Toward Data Infrastructure
Murphy said Marvell pursued that strategy through divestitures, internal investment and acquisitions, including Cavium, Avera, Aquantia, Inphi and Innovium. Over the past 12 months, he said Marvell divested its automotive Ethernet business and acquired Celestial AI for Photonic Fabric technology and XConn for scale-up switching.
According to Murphy, Marvell has invested about $22.5 billion through acquisitions and $18 billion organically, while divesting about $4.5 billion of assets, for roughly $36 billion of net investment in its platform.
He also cited Wall Street consensus estimates, saying Marvell is expected to reach $11.4 billion in revenue in the current year and $16.4 billion next year. Murphy said the company has been growing at about 40% annually in recent years.
NVIDIA Partnership Takes Center Stage
Murphy described Marvell as a connectivity-focused company that partners across the AI ecosystem, including with compute and memory companies. He pointed to a recently announced expansion of Marvell’s strategic partnership with NVIDIA, saying NVIDIA invested $2 billion into Marvell and that the companies are expanding work across optics, photonics and NVLink Fusion.
Jensen Huang joined Murphy onstage and said AI demand is rising because “useful AI has arrived.” He described agent-based computing as a distributed and disaggregated model that depends heavily on connectivity.
“What makes it possible is connectivity,” Huang said, referring to large AI clusters that aggregate compute, memory and bandwidth across data centers.
Huang also discussed NVIDIA’s Vera Rubin platform and NVLink Fusion, saying the partnership allows NVIDIA and Marvell technologies to be combined in heterogeneous data centers. He said customers that want to design their own ASICs can still incorporate NVIDIA technology, while Marvell can support custom silicon, interconnect, silicon photonics and optics.
Marvell Announces 100T Ethernet Switch
Murphy outlined Marvell’s connectivity portfolio across distances, from links spanning hundreds or thousands of kilometers between data centers to millimeter-scale die-to-die connections inside chip packages.
For long-distance connections, Murphy said Marvell builds coherent optical modules and is currently shipping 800 gigabit products in volume. He said the company plans to sample what he called the world’s first 1.6 terabit, 2-nanometer coherent optical solution later this year.
Inside data centers, Murphy said Marvell provides PAM4 chipsets, analog components and Ethernet switching infrastructure. He said the company began ramping 1.6 terabit, 3-nanometer PAM4 solutions last year.
At COMPUTEX, Murphy announced Marvell’s new 100T Ethernet switch, which he said is designed for AI data centers and offers the industry’s lowest power. He said Marvell’s Ethernet switching portfolio spans 12.8 terabits to 51.2 terabits.
Optics Moving Inside the Rack
A major theme of Murphy’s remarks was what he called the “copper wall,” or the limit on how far electrical signals can travel over copper as bandwidth increases. He said current 200 gigabit-per-lane systems can use copper cables of roughly 2.5 meters, but moving to 400 gigabit-per-lane speeds will make it impossible to fully connect a rack with copper.
“The wall is moving, and it’s moving now,” Murphy said. He argued that optical connections will increasingly move inside the rack, creating a significant increase in demand for optical components.
Huang said copper should be used “as much as we can for as long as we can,” but added that copper has limits in bandwidth and distance. “You use optics wherever you must. You use copper wherever you can,” he said.
Murphy said co-packaged optics, or CPO, will be one way to address density and power challenges as optics moves closer to compute and switching silicon. He showed a traditional 100T Teralynx switch board and a CPO-based 51.2T switch using 16 optical engines of 3.2T each, saying the design eliminates copper traces on the printed circuit board and sends light directly from the package.
ASE CEO Discusses Manufacturing Ecosystem
Murphy also brought ASE CEO Dr. Tien Wu onstage to discuss semiconductor manufacturing capacity and Taiwan’s role in the AI infrastructure supply chain. Wu said ASE’s relationship with Marvell developed gradually and was based on Marvell’s ability to provide insight into future architecture and technology requirements.
Wu said Taiwan’s semiconductor ecosystem is difficult to replicate because of decades of accumulated experience, scale and workforce depth. He said Taiwan has built up roughly 350,000 semiconductor employees and 1.1 million high-tech employees over 40 years of transitions from PCs to wireless, mobile computing, data centers and high-performance computing.
Murphy closed by describing a future in which optical connectivity reduces the importance of distance in data center design. He said that could allow larger scale-up domains, pooled compute and memory, and AI systems designed around workload needs rather than interconnect limits.
About Marvell Technology (NASDAQ:MRVL)
Marvell Technology Group is a global semiconductor company that designs and develops integrated circuits and related software for data infrastructure, networking, storage and connectivity markets. The company’s product portfolio includes system-on-chip (SoC) solutions, Ethernet physical-layer transceivers (PHYs), switch and switch silicon, optical interconnect components, storage controllers, and security processors. Marvell’s technology is used to enable high-performance data centers, carrier networks, enterprise and cloud storage, as well as connectivity in automotive and industrial applications.
Founded in 1995 and headquartered in Santa Clara, California, Marvell has grown through both organic development and strategic acquisitions to broaden its capabilities across networking and data interconnect.
