Nvidia diversifies infrastructure bets
Nvidia has broadened its AI infrastructure strategy by backing SiFive in a $400 million raise at a $3.65 billion valuation and reportedly investing $2 billion in Marvell Technology. The SiFive deal targets open RISC‑V CPU designs for AI data centres, while the Marvell investment was framed as strategically significant for networking and system-level infrastructure. Together the moves indicate Nvidia is moving beyond GPUs into adjacent compute and interconnect architecture. (parameter.io) (fool.com)
Nvidia has put fresh money behind two other parts of the artificial intelligence stack: processors from SiFive and networking gear from Marvell. (sifive.com) (nvidianews.nvidia.com) SiFive said on April 9 that it raised $400 million in an oversubscribed Series G round at a $3.65 billion valuation, with Atreides Management leading and Nvidia among the investors. The company said the money will fund high-performance data-center products based on RISC-V, an open instruction set used to design central processing units. (sifive.com) (reuters.com) On March 31, Nvidia and Marvell said Nvidia invested $2 billion in Marvell and that the two companies would work together on custom artificial intelligence systems and silicon photonics, which uses light to move data between chips. Marvell said it will provide custom processors and networking that work with Nvidia’s NVLink Fusion platform. (nvidianews.nvidia.com) (reuters.com) A central processing unit runs general-purpose instructions and manages system tasks, while a graphics processing unit handles many calculations at once for model training and inference. Networking and interconnects are the links between those chips, and they determine how fast a rack of servers can share data. (nvidianews.nvidia.com) (techcrunch.com) SiFive’s pitch is that RISC-V gives cloud and chip companies an alternative to Arm and x86 for server processors. Reuters reported SiFive is trying to enter the fast-growing market for data-center central processing units, and SiFive said agentic artificial intelligence workloads are driving demand for its processor intellectual property. (reuters.com) (sifive.com) Nvidia and SiFive had already tied their roadmaps together in January, when SiFive said its compute platforms would connect to Nvidia’s NVLink Fusion to link central processing units with Nvidia graphics processors and other accelerators. That earlier deal gave Nvidia a path to support more non-Arm host processors inside artificial intelligence servers. (sifive.com) The Marvell deal pushes on a different bottleneck: moving data across bigger systems. Nvidia said Marvell will bring scale-up networking into NVLink Fusion, while the two companies also plan to develop silicon photonics for faster, more efficient links inside artificial intelligence infrastructure. (nvidianews.nvidia.com) Investors treated the Marvell announcement as material news. CNBC reported Marvell shares jumped 13% on March 31 after Nvidia disclosed the stake and partnership. (cnbc.com) Taken together, the two moves show Nvidia spending on the parts around the graphics processor: the central processor that feeds it and the network that keeps large systems stitched together. The result is a broader bet on who controls the full design of an artificial intelligence data center, not just the chip at its center. (nvidianews.nvidia.com) (sifive.com)
