Agentic AI Pulls Compute Back On-Prem, Oxide's Cantrill Says
This summary and analysis were generated by AI from the original article at AI – SiliconANGLE and may contain errors (how Viqus works). Read the source for full details.
3
What is the Viqus Verdict?
We evaluate each news story based on its real impact versus its media hype to offer a clear and objective perspective.
AI Analysis:
The piece frames on-prem AI as a resurgent trend, but it is mostly one vendor's argument with no market figures, so the hype outpaces the evidence presented.
Article Summary
In a sponsored theCUBE interview at SC26, Oxide Computer co-founder and CTO Bryan Cantrill argued that agentic AI is reviving the rent-versus-own debate for compute. He said agentic workloads, which are largely CPU-heavy, are moving on-premises for economic, security, latency and data governance reasons. Cantrill pointed to enterprise interest in running open-weight models on hardware they control so data never leaves the building. He described Oxide's design choices, including removing the UEFI BIOS and baseboard management controller to reduce attack surface and building public cloud-grade multi-tenancy from the start, as a way to isolate adversarial tenants. The same racks can also host legacy virtualization workloads for customers unhappy with Broadcom's VMware pricing. The piece is a vendor executive's viewpoint rather than independent analysis, and it offers no new market data. It does reflect a real, if still early, trend in enterprise infrastructure strategy.
Key Points
- Cantrill argues that at sufficient scale, enterprises will prefer owning their compute instead of renting it from cloud providers.
- Agentic AI workloads are driving demand for CPU-heavy compute that enterprises increasingly want to bring in-house, with economics and model location as the main drivers.
- Oxide's design choices, such as stripping the BIOS and BMC and building multi-tenancy in from the start, are presented as security advantages for isolating agentic workloads.
Why It Matters
This is a vendor executive's perspective in a sponsored segment, not new data or a market event. It is still a useful signal that data governance and open-weight model deployment are shaping infrastructure choices, but the claims about an on-prem shift are not quantified here. Readers should treat it as routine positioning from a company that benefits from the thesis.

