
Why did NVIDIA invest in Mellanox Technologies? Wedbush: Aiming to enhance NVLink Fusion technology
Wedbush Securities stated that NVIDIA's $2 billion investment in Marvell Technology aims to enhance its NVLink Fusion technology. Analyst Matt Bryson pointed out that this collaboration will involve core technologies such as NVLink, AI-RAM, and optical networks. Although the criticality of specific technologies is uncertain, Marvell Technology is motivated to complete the deal, providing customized AI accelerators and networks compatible with NVLink Fusion. NVIDIA will provide technical support with its Vera CPU, ConnectX network cards, and other technologies
According to Zhitong Finance APP, Wedbush Securities stated that NVIDIA's (NVDA.US) $2 billion investment and strategic partnership with Marvell Technology (MRVL.US) is entirely for NVIDIA's NVLink Fusion technology.
Analyst Matt Bryson wrote in a report to clients: "This collaboration includes cooperation between NVIDIA and Marvell Technology on core technologies such as NVLink, AI-RAM, optical networks, and SiPh. What we are less certain about is which of these technological collaborations is crucial for NVIDIA to drive this partnership. On the other hand, we believe that given NVIDIA's $2 billion investment, Marvell Technology is motivated to complete this deal, regardless of whether collaborating with NVIDIA brings technological or operational benefits (or whether it brings benefits at all)."
NVIDIA's NVLink Fusion is the company's rack-level artificial intelligence infrastructure, allowing hyperscale data centers and custom ASIC designers (such as Marvell Technology) to integrate their custom CPUs and XPUs with NVIDIA's NVLink interconnect technology. Marvell Technology produces custom XPUs for several clients, including Microsoft (MSFT.US) and Amazon (AMZN.US).
As part of the deal, Marvell Technology will provide custom AI accelerators (XPUs) and horizontally scalable networks compatible with NVLink Fusion. NVIDIA will provide its Vera CPU, ConnectX network cards, Bluefield DPU, NVLink interconnect, and Spectrum-X switches, as well as rack-level AI computing capabilities
