Hyperscalers Internalizing CPU Design Hurts SiFive

This is a distribution problem as much as a sales problem. A hyperscaler that builds its own CPU does not just stop paying SiFive a license fee, it also stops acting like a public proof point that pulls tool vendors, operating system work, and adjacent chipmakers toward the same standard. That matters because SiFive sells blueprints and enablement, so market momentum helps convert a niche architecture into repeatable design wins across many customers.

• SiFive’s revenue today is still driven mainly by upfront licensing and design services, not mature royalties. That makes large new design starts especially important. In 2023, revenue was estimated at $38.2M, with roughly 60% from IP licensing and 30% from custom design services, so losing a big in house buyer hits near term revenue quickly.
  1 sacra 2 sacra
• Hyperscalers shape the ecosystem by forcing infrastructure software to become real, not experimental. RISC-V International’s RVA23 profile was ratified to make software portable across vendors, and it explicitly adds pieces needed for rich operating systems, virtualization, and cloud workloads. When big cloud buyers build privately, less of that ecosystem work turns into a visible commercial tailwind for SiFive.
  2 sacra 3 riscv
• The substitute for SiFive is not only internal design. Ventana goes after the same cloud and infrastructure opportunity with chiplets, turnkey platforms, and IP, which lowers the amount of CPU work a customer must do itself. At the same time, Arm keeps reducing upfront adoption friction through Flexible Access, narrowing one of RISC-V’s historical advantages.
  2 sacra 4 arm 5 ventanamicro

Going forward, the biggest wins for SiFive are likely to come from companies that want custom silicon but do not want to build an Apple or Google sized CPU team. If hyperscalers keep internalizing the top end, SiFive becomes more valuable as the fast path for second tier cloud, automotive, industrial, and infrastructure buyers that need proven cores, software support, and predictable tapeout schedules.