Matt Murphy, Chairman and CEO of Marvell Technology, and Jensen Huang, founder and CEO of NVIDIA, on June 2 discussed how their companies are collaborating to provide customers with greater choice and flexibility in developing next-generation artificial intelligence (AI) infrastructure at the Taipei Nangang Exhibition Center Hall 2.
Murphy noted that tens of thousands—and eventually millions—of computer processors must work together as a single, massive compute engine, making computing at this scale fundamentally a connectivity challenge —- the next major wave of innovation and scale.
Citing tech giants Meta, Google, AWS, and Microsoft, Murphy said the industry was undergoing a major transformation. Just as it did with compute and memory, the tech sector would rally to meet this connectivity challenge, he said.
As reasoning models, mixture-of-experts architectures, and generative AI continue to evolve, larger volumes of data must move across infrastructures, demanding higher bandwidth and lower latency. Because workloads could no longer fit into a single data center, larger facilities must be built, Murphy said.
Consequently, connectivity had become a critical enabler of scaling compute, with optics widely recognized as the path forward, he said.
Highlighting a recent expansion of Marvell’s partnership with Nvidia—supported by a US$2 billion investment from the latter—Murphy said the two companies are broadening their collaboration across multiple dimensions, including optics, photonics, and NVLink Fusion.
Huang, who appeared as Murphy’s special guest, noted that AI agents utilize a highly disaggregated and distributed computing pattern, making robust connectivity essential.
When asked about the transition from copper to optics, Huang said that while copper faces limitations in bandwidth and distance, the ultimate strategy is to scale up with copper, scale up and scale out with optics, and eventually scale across using optics.
The intersection of the two technologies would persist for a long time, Huang said, predicting that the industry would use “a ton of copper and tons of optics” over the next five to ten years as data centers become core infrastructure.
During the keynote, Murphy presented Marvell’s latest optical module and announced the new T100 Ethernet Switch, which is specifically designed for AI data centers and features the industry’s lowest power consumption.
He also introduced co-packaged optics (CPO) as a solution to the fundamental challenges of density and power, calling the technology a massive shift for the industry.
Despite its high complexity, CPO is the only way to continue scaling bandwidth and overcome current hardware limitations while simultaneously reducing power consumption, Murphy said.
“This is happening now,” Murphy said. “This is where the industry is heading.”
The AI infrastructure industry is moving toward a future of “data center without distance,” where compute, memory, networking and photonics operate as one unified system, and where millions of resources across the data center can work together as if they were one machine, he said.
This would be an architecture defined by the needs of the workload, not by the limits of the connectivity, he said, “We believe this is the next era of computing infrastructure.”
