SiFive targets infrastructure sidecar CPUs
SiFive
The real opening for SiFive is not replacing x86 or Arm in mainstream servers, it is winning the smaller CPUs that sit beside accelerators and keep modern systems running. These chips handle packet processing, storage tasks, system bring up, and local inference in tightly scoped environments where customers care more about customization, power, and roadmap control than broad legacy software compatibility. That makes infrastructure sidecars a much more practical path to production RISC-V volume than general purpose compute.
-
SiFive sells licensable CPU blueprints, not finished chips. Customers pay upfront for RTL, verification tools, and engineering help, then pay royalties when their own chips ship. That model fits storage controllers, networking boxes, and edge devices well, because these programs use long lived custom SoCs and can generate royalties for years after design win.
-
These entry points have narrower software demands than a general server CPU. A storage controller runs firmware and fixed data path code. A networking appliance runs a defined packet workflow. An edge inference box often runs a tuned model stack on device. In each case, the customer controls most of the software surface, which lowers the switching cost to RISC-V.
-
Competition is still intense. Arm has reduced early licensing friction through Flexible Access, which weakens the simple cost case for RISC-V. At the high end, Ventana is pushing datacenter class RISC-V chiplets and full solutions, while the biggest buyers are also internalizing design, as shown by Meta acquiring Rivos in 2025.
If RISC-V reaches production scale in infrastructure, it will likely spread from these controlled side workloads inward. A standardized win in storage, networking, or edge AI can turn into millions of units, then pull in more tools, software support, and ecosystem confidence. That is the path by which SiFive moves from experimental IP vendor to a durable part of the datacenter and edge compute stack.