Why are RISC-V and UCIe chiplets breaking Arm's royalty tax and the x86 duopoly in custom AI silicon?
Hyperscalers and AI silicon startups are bypassing Arm's escalating per-chip royalties and the closed x86 licensing duopoly by pairing open-source RISC-V cores with Universal Chiplet Interconnect Express (UCIe) packaging. Modular die-to-die interconnects now let chip architects stitch together custom AI accelerators without paying instruction-set gatekeepers.
As monolithic reticle limits at 3nm and 2nm nodes drive tape-out costs past $500 million, the UCIe open standard enables heterogeneous chiplets—combining compute tiles, HBM stacks, and I/O dies from different foundries—to communicate with sub-2-nanosecond latency and terabit-per-millimeter shoreline bandwidth. Because x86 cannot be licensed for custom third-party chiplets and Armv9 Compute Subsystems (CSS) charge steep upfront fees plus rising royalty percentages on total chip ASP, hardware margins are migrating to open architectures.
Vendors like Tenstorrent, SiFive, and Andes are embedding royalty-free RISC-V vector and matrix extensions directly alongside neural processing engines to handle FP8 and INT4 tensor orchestration. By eliminating proprietary ISA lock-in and standardizing die-to-die physical layers, cloud providers can swap out compute chiplets across generations without rewriting their entire compiler and firmware stack.
Can Arm defend its data center licensing premium through software ecosystem lock-in, or will RISC-V and UCIe chiplets commoditize custom AI processor design?
Comments
Be respectful. Comments reported by several people are hidden until a moderator reviews them.
You do not need an account to read, only to comment. Sign in with a code sent to your email or with your Google account. We only show the name you choose, never your email.