Nvidia (NVDA) has had a busy week in Berlin. At its GTC event it announced that its next-generation Vera Rubin hardware is heading to Google Cloud, launched a blueprint for keeping AI agents on a leash, and pitched Europe on building its own AI capacity. Here is the whole thing in about three minutes.
What is Nvidia selling this week?
Three things, really. The first is hardware: Vera Rubin, Nvidia's newest rack-scale system pairing Vera CPUs with Rubin GPUs. Google Cloud's new A5X instances combine Vera Rubin NVL72 racks with Nvidia networking chips, Google's own network and liquid cooling. Google says general availability comes next month, with Europe to follow, and it plans to scale to as many as 960,000 Rubin GPUs across multiple sites. Those are plans, not delivered capacity.
The second is software: a family of open models, including Nemotron 3.5 for language, Cosmos 3.5 for robots and cars, and Alpamayo 2 Edge, a compact driving model built for Nvidia's DRIVE AGX Thor platform. The third is a safety framework for AI agents, which we look at more closely below.
What is OpenShell, and why does it matter?
An AI agent is software that takes actions on your behalf, not just answers questions. Companies like that idea until the agent touches something it shouldn't. Nvidia's OpenShell Reference Architecture is its answer: run agents in contained sandboxes, watch them, and enforce rules in hardware. BlueField chips run a security framework called Sentry that monitors agent behavior from outside the system.
Jensen Huang put it this way: "AI capability and safety must advance together". Nvidia says the architecture is available now and that it works with models from Anthropic, Google, Meta, OpenAI and others. Security firms such as CrowdStrike and Palo Alto Networks are building support, and Spanish banks BBVA and Santander are exploring it. ServiceNow says its tools can use it to revoke an agent's access across software and hardware within seconds.
There is a real-world example. Marketing group WPP ran a creative agent on A5X that turns requests into video workflows, with human approval before anything executes. In early tests, work that normally needed a multi-node cluster ran on one instance. That is WPP's own test result, reported by Nvidia.
Why is Europe the focus?
Nvidia argues European developers want compute close to their data and models they can customize. Its cloud partners, including Nscale, Crusoe and Nebius, are expanding capacity across Europe and beyond. Power is the bottleneck. Nscale says that in tests with GB300 systems, Nvidia's DSX software fit 37% more GPUs and nearly 50% more inference throughput into the same power budget. Again, that is a vendor-reported figure.
What is still unknown?
Pricing for A5X wasn't in the sources we read, nor was the exact launch date beyond "next month". Nobody has published independent tests of OpenShell, so how well it contains a determined agent is unproven. How much European capacity will actually switch on, and when, is also unclear.
What happens next?
Watch for A5X going live next month, and for whether customers beyond early adopters like WPP and Ineffable Intelligence, the London firm founded by AlphaGo creator David Silver, sign up. Would you trust an agent at work because its safety rules are baked into the chip?
The WattsUpNext take: Nvidia is no longer just selling chips for AI. It is selling the rack, the models and now the guardrails. The guardrails are the part worth testing.
That's wattsup, and here's wattsnext: Austin's electric buses can now recharge from overhead arms at the end of the line
Sources
- 1.Google Cloud and NVIDIA Announce General Availability of NVIDIA Vera Rubin Infrastructure · NVIDIA Blog
- 2.NVIDIA Open Models Advance Europe’s Sovereign Intelligence Across Industries · NVIDIA Blog
- 3.NVIDIA Launches OpenShell Reference Architecture for Safe AI Agents · NVIDIA Newsroom
- 4.NVIDIA Cloud Partners Scale Across Europe — Sovereign, Open and Built for Safe AI Agents · NVIDIA Blog
Reported by the WattsUpNext desk from the sources linked below. Spot an error? Tell us at corrections@wattsupnext.com.
The WattsUpNext Brief
Get stories like this in your inbox.
One email each weekday, only the topics you choose. Real news, sourced, no fluff. Unsubscribe in one click.
