Every AI agent that books a flight, reads a contract or touches a bank account is handling someone’s private data. As agents multiply, the question shifts from how fast a model thinks to whether anyone can see what it is doing.
That is where Intel has quietly built a lead.
Nvidia’s own flagship servers run on Xeon
Nvidia’s DGX B300 uses Intel’s Xeon 6776P as its host processor. At GTC in March, Intel was selected again as the host CPU for the next-generation DGX Rubin NVL8, with up to 8TB of system memory and 2.3 times the memory bandwidth through MRDIMM.
“In this new era, the host CPU is mission-critical,” said Intel’s Jeff McVeigh.
Intel TDX seals the whole path
Intel Trust Domain Extensions put each workload in a hardware-sealed trust domain that even the cloud operator cannot read. In the DGX Rubin systems, TDX covers the data path from the Xeon to the Nvidia GPUs, with hardware-rooted isolation and attestation. TDX Connect extends that trust to accelerators over industry-standard PCIe security.
For a bank, a hospital or a government running AI agents, that is the difference between allowed and not allowed.
Agents run on CPUs
At Computex in June, Lip-Bu Tan argued that agentic AI shifts the data center back toward CPUs. Agents spend their time calling tools, running code, querying databases and checking permissions. All of that is CPU work, and all of it touches sensitive data. Intel’s server CPU shortage in Q3 is the first sign of that demand.
On the PC, private data stays local
With Perplexity, Intel showed hybrid agents that run smaller models on an Intel laptop and reach the cloud only for heavy lifting, so private files never leave the machine.
“The future is more compute in the data center and more compute on the local machine,” Tan said.
Why it matters for INTC
Security is sold once and kept for years. Once a regulated customer qualifies its confidential AI stack on Xeon and TDX, switching means re-certifying everything. Together with Nvidia’s choice of Xeon for its own systems, that makes Intel the default trust layer of the AI data center.
As AI agents take on real work with real data, trust becomes the bottleneck. Intel is built for it.
Grogu Ledak | The Blue Team
Sources: Tom’s Hardware on DGX Rubin NVL8 · SecurityWeek on TDX Connect · Technobezz on Computex 2026 · Dow Jones on Intel’s full-stack AI strategy
Leave a Reply