Nvidia-backed SiFive hits $3.65 billion valuation for open AI chips

SiFive, an organization based in 2015 by the UC Berkeley engineers who created an open supply chip design, has landed a $400 million oversubscribed spherical that values the corporate at $3.65 billion.

This deal is attention-grabbing for a bunch of causes. For one, SiFive’s RISC-V open chip design is predicated on the RISC processor, not Intel’s x86 or ARM, the 2 main varieties of CPUs that presently feed Nvidia’s GPU pc system AI empire.

Additionally, Nvidia was investor on this spherical, alongside a protracted record of VCs, personal fairness, and hedge funds. The spherical was led by Atreides Administration, based by former Constancy investor bigwig Gavin Baker. (Atreides was additionally an investor in Cerebras Systems $1 billion round). Different buyers within the spherical embody Apollo World Administration, D1 Capital Companions, Point72 Turion, T. Rowe Worth Sutter Hill Ventures, and others.

SiFive’s enterprise mannequin is like Arm’s was in years passed by — it licenses its chip designs to those that modify them for their very own wants and doesn’t promote the chips themselves. (In March, Arm modified its mannequin when it launched the first-ever chip it manufactured, an AI chip, developed with Meta with prospects together with OpenAI, Cerebras, and Cloudflare.)

SiFive stands in rarified air with chip designs which might be open, not proprietary, in addition to impartial, not reliant on particular prospects. The truth is, SiFive hasn’t raised since March 2022, Pitchbook estimates, when it introduced in $175 million led by Coatue Administration at a pre-money valuation of $2.33 billion. Intel Capital, Qualcomm Ventures, Aramco Ventures, have been a part of that spherical.

RISC-V has been, till lately, higher often known as a chip for smaller makes use of, like embedded programs. However with this money and Nvidia’s consideration, SiFive is transferring into CPUs for AI knowledge facilities. SiFive’s designs will work with Nvidia’s CUDA software program and its NVLink Fusion, a rack server system that lets totally different CPUs plug into Nvidia’s “AI manufacturing unit.”

In different phrases, as rivals Intel and AMD search to compete with Nvidia’s GPU, Nvidia is backing an 11-year-old startup that may design CPUs on an open and fully alternate know-how.