Broadcom's software leadership put a hard number on a debate that has simmered for years: meaningful Arm adoption in enterprise datacenters is at least three years out, and RISC-V is not a serious contender in that window at all. The framing matters because it comes from the company that now controls VMware, the virtualization layer that sits under a large share of the world's on-premises workloads.
The global read is that the bottleneck was never the processor. Arm already proved itself at hyperscale—AWS Graviton, Google Axion, Microsoft Cobalt—because those operators own their entire stack and can recompile, certify, and optimize at will. Enterprise IT cannot. It runs shrink-wrapped ISV software, legacy binaries, and certified appliance images that assume x86. Until hypervisors, management tooling, monitoring agents, and the long tail of third-party software carry first-class Arm support, the migration math doesn't close. That is a software-ecosystem problem measured in vendor roadmaps, not a fab problem. RISC-V sits a further step back: promising for embedded and specialized accelerators, but lacking the certified enterprise software base to matter for general-purpose datacenter compute soon.
The strategic tension for Intel and AMD is that this buys x86 more time in the enterprise even as it loses share in the cloud. For Arm-based server vendors, the message is that design wins alone won't move on-prem revenue without heavy investment in the boring, unglamorous certification work.
For Japan, the timing is pointed. Japanese enterprises are among the most on-premises-heavy and SIer-dependent in the developed world, and they are simultaneously absorbing Broadcom's aggressive VMware licensing overhaul—a cost shock that is already pushing firms to reevaluate their virtualization strategy. An Arm transition layered on top of that would compound migration risk, and this three-year horizon gives conservative Japanese IT departments a defensible reason to defer. That is a double-edged outcome. Japan has genuine Arm credibility through Fujitsu's A64FX and the Fugaku supercomputer, so the capability exists domestically—but HPC prestige rarely translates into mainstream enterprise procurement.
For SIers, the practical guidance is to treat Arm as a medium-term architecture play, not a near-term migration project. The value now is in advisory work: helping clients decide whether the VMware cost pressure justifies moving to alternative hypervisors or cloud-native platforms, and building Arm-readiness into those decisions rather than launching standalone porting efforts. RISC-V belongs in the R&D and edge-device conversation, not the datacenter roadmap. Firms that quietly build recompilation, testing, and dependency-audit muscle over the next two years will be positioned to lead when the ecosystem finally catches up—while those chasing the transition today risk stranding budget on software that isn't certified to run.
