Nvidia and Marvell Technology tightened work on AI efforts geared at the ongoing AI-infused betterment of telecommunication radio access network (RAN) systems, developing new silicon photonics technology, and by the former investing cash into the latter.

The AI-RAN link taps Nvidia’s NVLink Fusion platform to connect Marvell into Nvidia’s AI factory and AI-RAN ecosystem. This is targeted at allowing customers greater flexibility when building on Nvidia architectures.

Nvidia launched NVLink Fusion last year, opening its decade-old NVLink interconnect technology to outside players. This allowed customers to build semi-custom AI infrastructure that links non-Nvidia accelerators with Nvidia hardware in rack-scale solutions.

Specific to Marvell, the vendor will provide customer XPUs and NVLink Fusion-compatible scale-up networking. Nvidia for its part, will supply supporting platforms like its Vera CPU, ConnectX network interface card (NIC), Bluefield digital processing unit (DPU), NVLink interconnect, and Spectrum-X switches, and its rack-scale AI compute offering.

“Our expanded partnership with Nvidia reflects the growing importance of high-speed connectivity, optical interconnect, and accelerated infrastructure in scaling AI,” Marvell CEO Matt Murphy noted on the deal. “By connecting Marvell’s leadership in high-performance analog, optical DSP, silicon photonics, and custom silicon to Nvidia’s expanding AI ecosystem through NVLink Fusion, we are enabling customers to build scalable, efficient AI infrastructure.”

This work will support Nvidia’s Aerial AI-RAN project, which the company open-sourced last October. That platform combines accelerated computing platforms, software, and services for the management of wireless networks, and is currently being piloted by a number of operators, including SoftBank in Japan.

Nvidia’s AI-RAN and Aerial work is also part of a recently unveiled Linux Foundation “open” initiative with roots in the U.S. Department of Defense (DOD) to accelerate work on streamlining AI efforts to benefit 6G technology development. That work is through the newly launched Open Centralized Unit Distributed Unit (OCUDU) Ecosystem Foundation, which involves “premier partners” such as AMD, AT&T, DeepSig, Ericsson, Nokia, Nvidia, SoftBank, SRS, and Verizon.

Those efforts are being combined with Nvidia’s previously launched 6G AI-Wireless Intelligent Network (AI-WIN) venture. That venture was initially underpinned by Nvidia’s AI Aerial platform, and late last year gained efforts from Open RAN Development Company (ODC) toward constructing an “all-American AI-RAN stack.”

“AI is redefining computing and driving the largest infrastructure buildout in human history – and telecommunications is next,” Nvidia CEO Jensen Huang boastfully claimed in a statement tied to this new endeavor. “Together with a global coalition of industry leaders, Nvidia is building AI-RAN to transform the world’s telecom networks into AI infrastructure everywhere.”

Silicon photonics and investment

Nvidia and Marvell will also collaborate on unspecified silicon photonics technology.

Marvell recently bolstered its work in this space with the $3.25 billion acquisition of Celestial AI. That deal includes Celestial’s Photonic Fabric platform and technology team that joined Marvell’s Data Center Group, a segment that’s seen sustained growth in recent years, with robust demand for its 800 Gb/s PAM4 optical products and Teralynx Ethernet switches.

"Celestial’s photonic interconnect will enable scale-up connectivity between XPUs and memory fabric connectivity to expand near-processor memory capacity," equity research firm William Blair noted in a report on Nvidia-Marvell work.

Marvell’s purchase is being somewhat offset by a new $2 billion investment from Nvidia.

"For Marvell, this investment highlights the growing significance of interconnect technology as a pivotal enabler of AI system performance across future generations," the William Blair report noted on the investment. "We also see it as a vote of confidence in Marvell’s technology portfolio, which should help quell recurring questions about the quality of its underlying IP and ability to gain share."

Direct details on Nvidia’s Marvell investment were light, but it did mirror a similar investment Nvidia made in optical and photonic developer Lumentum, which Huang touted as supporting “AI factories.”

“Together with Lumentum, Nvidia is advancing the world’s most sophisticated silicon photonics to build the next generation of gigawatt-scale AI factories,” Huang noted in a statement tied to that investment.

That Lumentum investment also added to Nvidia's aggressive efforts to shore up its supply chain amid growing industry component shortages, with it committing to securing future capacity access rights to Lumentum's laser components.

Nvidia had already moved to buttress its supply of electro-absorption modulated lasers (EML) to such a degree that reports claim rival optical module makers could face wait times for components beyond 2027.

Nvidia’s efforts to shore up its own supply chain have been linked to potential increased lead times for other transmissions over long distances – making them vital for use in high-speed optical communication systems, including in data centers to boost network performance and efficiency