| Mobile Web

Search results for DPU

AI & Enterprise

Google and Meta turn back to Nvidia tie-up on AI networking hardware

Google and Meta are adopting Nvidia\'s AI networking hardware even as they develop their own AI semiconductors to reduce reliance on Nvidia. Nvidia unveiled the plan at the Open Compute Project global summit in San Jose. Meta will build open AI networking using Nvidia Spectrum-X and Spectrum-6, while Google will support Falcon Transport on Nvidia ConnectX-10 and BlueField DPU. Nvidia also shared rack designs and new data centre standards with partners.

AI & Enterprise

OpenAI missing from Nvidia AI agent safety alliance, but working on related technology

OpenAI is not participating in Nvidia’s AI agent safety alliance, but is cooperating with Nvidia on related technology, including the open-source software OpenShell, the article said. Nvidia unveiled its Open Agent Safety Platform with more than 100 companies, while OpenAI and several big tech firms were absent from the official supporters list. The platform combines software sandboxing with Nvidia hardware monitoring, including its proprietary Sentry feature.

Industry

Real agent is AI that works on its own under orchestration, Nvidia Korea executive says

Nvidia’s idea of a real agent is AI that keeps working on its own for as long as 12 hours after being given a task, Nvidia Korea executive Kim Sun-wook (김선욱) said at Digital Insight 2026 in Seoul. He said true agents run under orchestration and require linked tools. Kim described agents as an LLM paired with a harness, with LLMs running on GPUs and the harness mainly on CPUs, while DPUs handle data processing and memory supports continuity.