# Oxide CTO Says Agentic AI Is Driving Demand for On-Prem Compute

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/62234
Published: 2026-11-20T17:49:35.000Z
Updated: 2026-11-20T17:49:35.000Z
Section: Technology

> Bryan Cantrill says economics is the main reason to bring workloads in-house, with model placement close behind.

Agentic AI workloads are driving an explosion in computing demand, including CPU-heavy tasks, and Oxide Computer Co. co-founder and Chief Technology Officer Bryan Cantrill says that is strengthening the case for companies to own infrastructure they can control.

Economics is the main reason to bring workloads in-house, Cantrill said, while the location of AI models is a close second. The model plane is the layer that decides where AI models run, making infrastructure choices relevant to how companies manage their data.

“There is a real enterprise drive towards open-weight models, and I think you’re going to see more and more of that as people want to run the models on-prem and allow their data to not actually leave,” Cantrill said.

Oxide raised $445 million in October to expand rack production. Cantrill said the company was built around the view that enterprises reach a scale at which owning their computers makes sense.

“Cloud computing is great, but the model shouldn’t be exclusively to rent it. You get to a certain scale, as it turns out, you want to own your own computer,” Cantrill said.

Security is another factor in infrastructure decisions. Oxide removed the UEFI BIOS and baseboard management controller from its rack to reduce its attack surface. The company also built in public cloud-grade multi-tenancy. Cantrill described tenants sharing the same computer and network while keeping one another’s information confidential.

The same racks can also run legacy virtualization workloads. Cantrill said some customers dissatisfied with pricing changes at Broadcom’s VMware business want a general-purpose platform for older and newer applications.

Companies may want elastic computing in their own data centers for economic, security or latency reasons, Cantrill said, adding that they want to control their fate.


