Q&A with SiFive co-founder Krste Asanović, who was part of the original team that developed RISC-V, on creating a company that builds IP around RISC-V, and more
Update: SiFive just announced today that it has partnered with NVIDIA for NVLink Fusion on RISC-V. It's a big step, and we'll be covering that separately.
Context & Ripple Effects
SiFive’s business is built around commercial IP for the open RISC-V architecture, a model that had already reached major manufacturing channels through Intel’s licensing of SiFive IP for its foundry offerings. The interview arrives alongside SiFive’s stated plan to integrate Nvidia NVLink Fusion into its RISC-V platforms, shifting the discussion from architecture origins to system-level connectivity.
The company’s earlier funding history, including its $175M Series F financing, shows the long-running effort to turn RISC-V design expertise into a scalable IP business. The Nvidia tie-up is the more immediate test of whether that IP can fit into heterogeneous compute deployments.
First-order effects
- SiFive’s RISC-V processor IP platforms gain a planned NVLink Fusion integration, enabling prospective silicon customers to connect SiFive-based designs with Nvidia and partner chips.
- Nvidia gains another processor-IP route for extending NVLink Fusion into custom silicon designs, rather than limiting the interconnect’s relevance to a single processor architecture.
Second-order effects
- Chip designers evaluating RISC-V can weigh access to an Nvidia-linked interconnect ecosystem alongside core IP choice, increasing the importance of integration support in SiFive’s offering.
- Competing processor-IP vendors and interconnect providers face pressure to show comparably practical paths for heterogeneous designs that combine CPUs, accelerators and partner silicon.
Third-order effects
- If such integrations proliferate, RISC-V’s commercial differentiation may depend less on the openness of the instruction set alone and more on access to validated interconnect and software ecosystems.
- The pattern points toward heterogeneous AI compute in which specialized IP vendors can participate in larger platforms, while platform owners retain influence through the connective technology around the chips.
The trend: Open processor IP is increasingly being positioned as a component of tightly integrated heterogeneous compute systems rather than as a standalone alternative architecture.