Source linked

NVIDIA GB300 Blackwell Ultra Powers Claude Agents with Hardware-Level Governance

blogs.nvidia.com@frontier_wire3 hours ago·Artificial Intelligence·2 comments

NVIDIA's Secure Agent Workspace Reference Design gives Claude agents hardware-rooted identity and network controls, turning Azure Foundry into a governed autonomous agent platform.

nvidiaanthropicclaudemicrosoft azureblackwell ultraagentic ai

Anthropic’s Claude models in Microsoft Foundry now run on NVIDIA GB300 Blackwell Ultra GPUs, and the real news isn’t just the hardware — it’s that NVIDIA integrated a hardware-level governance layer that lets enterprises run autonomous agents without handing over the keys to the kingdom.

What the GB300 NVL72 Brings to Agentic Inference

Claude in Foundry runs on NVIDIA GB300 NVL72 systems with Quantum-X800 InfiniBand networking. That cluster delivers the raw compute needed for agentic AI — think autonomous sub-agents that coordinate across business domains, not just single-turn chat. Inference performance and efficiency drive total cost of ownership, and the GB300’s architecture is built for that.

Hardware-Enforced Governance for Autonomous Agents

NVIDIA’s Secure Agent Workspace Reference Design is the centerpiece. It provides a blueprint for running autonomous agents in a governed environment where identity, network access, credentials, and runtime policy are controlled at the infrastructure level. That means an enterprise can let a Claude agent act on its behalf — query databases, hit APIs, manipulate files — without worrying about credential leaks or lateral movement. The agent gets a hardware-enforced identity, not a shared service account.

NVIDIA’s Verified Agent Skills Tie Claude to Enterprise Workflows

Through NVIDIA verified agent skills — enabled by access to NVIDIA accelerated computing — enterprises can embed Claude agents deeply into their business processes. These skills give Claude domain-specific abilities that go beyond general reasoning, turning the model into a specialized operator. And because the skills run on the same GB300 hardware with the same governance controls, they inherit the security posture.

This is the three-way partnership between NVIDIA, Anthropic, and Microsoft — announced in November — delivering a concrete, production-ready product. Enterprises can now treat AI agents as first-class, governed system components. NVIDIA just set the bar for what a hardware-accelerated agent platform looks like.

Source: Claude Meets Blackwell Ultra: Anthropic's Models Now Run on NVIDIA GB300 in Azure
Domain: blogs.nvidia.com

Read original source ->

External source stays available while the OJO article and comment thread stay local.

More in Artificial Intelligence

view topic
vLLM's Semantic Router Beats Frontier Models by Collaborating Inside the API

A single OpenAI-compatible API call now hides a team of models. vLLM's Semantic Router uses bounded looper patterns-confidence escalation, parallel fusion, and budgeted workflows-to beat frontier models without...

DiScoFormer Cuts Density Error 37x in 100 Dimensions, No Retraining Needed

A single transformer estimates both density and score from a sample, beating KDE by orders of magnitude in high dimensions and adapting to new distributions on the fly.

Ornith-1.0 9B Doubles Qwen3.5 on Agentic Coding Benchmarks

Open-source models from 9B to 397B use RL to jointly optimize scaffold and solution rollouts, scoring 69.4% on SWE-bench Verified and 43.1% on Terminal-Bench 2.1.

50,000 Queries, Zero Breaches: PAR's Three-Layer Architecture Makes LLM Security Bite

PAR Technology enforces row-level SQL security deterministically using SigV4 signing, semantic validation on Bedrock, and Split-Plane CTEs-so a jailbroken LLM still cannot see another tenant's data.

Comments load interactively on the live page.