Analysis
Ram Velaga, president of Broadcom's enterprise software division, told VMware Explore attendees in Las Vegas on Sept. 2 that Arm servers "won't have significant market share in the enterprise for at least three to five years," The Register reported. Velaga took the role on Jan. 1, 2026, after leading Broadcom's networking silicon group.
The claim runs against the direction of nearly every headline about Arm in the datacenter. AWS Graviton now carries a large share of EC2 instance launches, Nvidia's Grace CPUs ship inside its own systems, and Microsoft and Google have both built Arm server silicon. Velaga is drawing a distinction that gets lost in those numbers: hyperscaler-operated Arm fleets running first-party and cloud-native workloads are a different market from enterprise-owned servers running licensed commercial software with vendor certification requirements.
That second market moves on certification cycles, not on performance per watt. An enterprise running SAP, Oracle Database or a decade of internally maintained x86 binaries does not migrate because the instruction set is more efficient. It migrates when its software vendors certify Arm builds and when its virtualization layer supports them, and VMware today virtualizes only x86. Broadcom previewed Arm hypervisor technology in May 2026. Pulse has tracked Broadcom's chip and infrastructure commitments throughout the year.
“An enterprise running SAP, Oracle Database or a decade of internally maintained x86 binaries does not migrate because the instruction set is more efficient.”
Velaga's positioning statement is the strategically interesting part: "We will be the abstraction layer, for CPU and GPU, regardless of instruction set." That is VMware attempting to remain the neutral substrate through an architectural transition rather than defending x86 -- the same play VMware ran successfully during the shift to cloud, and unsuccessfully during the shift to containers, where Kubernetes took the abstraction layer instead.
The resource picture explains the caution. Broadcom's software division holds 60% of a $5 billion annual R&D budget, with most of it going to maintaining current VMware products and the remainder to new offerings including an AI factory product. A maintenance-weighted budget is not the profile of an organization racing to enable an architecture transition, and a three-to-five-year public estimate conveniently matches the pace such a budget can support.
For semiconductor investors the disagreement is worth pricing. Arm's licensing revenue thesis and the valuations attached to Arm-native server startups assume enterprise adoption follows hyperscaler adoption with a short lag. The company that controls the enterprise virtualization layer is publicly saying that lag is at least three years, and it controls a meaningful part of the gating factor.
