Nvidia backs RISC‑V chip play
Reports say Nvidia invested in SiFive to diversify AI infrastructure away from traditional Arm licensing, part of a broader push to broaden chip design options for high‑performance AI compute. (startupfortune.com)
Nvidia has backed SiFive in a new $400 million funding round, adding weight to RISC-V chip designs for artificial intelligence data centers. (sifive.com) SiFive said April 9 that the Series G round was oversubscribed and valued the company at $3.65 billion. Atreides Management led the round, with Apollo Global Management, Nvidia, Point72 Turion, T. Rowe Price, Prosperity7 Ventures, and Sutter Hill Ventures also participating. (sifive.com) Reuters reported the money is aimed at pushing SiFive into the market for data-center central processing unit designs, where Arm has built a strong licensing business. SiFive sells processor intellectual property rather than manufacturing chips itself. (reuters.com) RISC-V is the basic instruction set, the rulebook a processor follows to run software, and the standard is maintained by RISC-V International as an open specification. That means companies can build compatible designs without relying on a single proprietary architecture owner. (riscv.org) Nvidia’s own data-center central processor, Grace, is built on Arm Neoverse cores, not RISC-V. Nvidia says a Grace CPU Superchip packs 144 Arm Neoverse V2 cores in one module, showing how deeply its current server roadmap still sits inside the Arm ecosystem. (nvidia.com) SiFive and Nvidia had already moved closer on January 15, when SiFive said it would integrate Nvidia NVLink Fusion into high-performance RISC-V data-center products. The companies said that link would let SiFive-based processors connect coherently to Nvidia graphics processing units and other accelerators in artificial intelligence systems. (sifive.com) SiFive said the new funding will go into scalar, vector, and matrix processor designs, software work, and customer deployments for data centers. Chief executive Patrick Little said hyperscale customers are asking for customizable central processing unit designs in intellectual-property form. (businesswire.com) The bet comes as cloud companies are trying to pair Nvidia graphics processors with more specialized central processors and interconnects for artificial intelligence workloads. SiFive said the round is its final private financing before a planned initial public offering. (techcrunch.com) For now, Nvidia is not replacing Arm in its shipping server chips. It is putting money behind a second architecture that could give data-center customers another way to build systems around Nvidia accelerators. (nvidia.com)
