Google and NVIDIA announce new AI computing power for large companies

Google Cloud and NVIDIA announced a new class of AI computing power for large companies. The first servers, built on NVIDIA's Vera Rubin chips, are due to become generally available next month. The news came at NVIDIA GTC Berlin 2026, an industry event.

What changed

The new servers are called A5X. They are bare-metal, which means a customer gets the physical machine instead of a shared slice. They use NVIDIA's Vera Rubin NVL72 platform. Google also says it is adding local regions in Europe, the Middle East and Africa, so companies can meet rules about where data must stay.

Built for AI agents

Google frames the launch around agentic AI. These are software agents that work through many steps on their own instead of answering a single question. Google says older infrastructure struggles with that kind of continuous work. The announcement lists several upgrades:

• Clusters can grow to 80,000 Rubin chips at one site, and up to 960,000 across several sites.
• Data can move from storage straight into the chips' memory, skipping slower steps.
• A caching layer reuses earlier context, which Google says cuts repeated work and shortens the wait for a first response.

Choice and control

Google is adding open models such as Gemma, Mistral and NVIDIA Nemotron to its Sovereign Cloud portfolio. Open models are AI systems whose core files are shared for others to run. Customers keep full custody of the model weights, the files that hold what a model has learned, along with the software that runs them. Google says this helps companies avoid lock-in.

Security gets attention too. Google describes Confidential Computing, which protects sensitive workloads with security built into the hardware. For work that must stay cut off from the internet, Google Distributed Cloud puts a managed cloud setup inside a company's own data center. NVIDIA Cosmos 3 can also generate virtual worlds where robots and self-driving vehicles practice before real-world use.

Who it affects

This is mainly for large companies. Ad giant WPP says its creative teams will use agents on the A5X servers for tasks such as frame-accurate quality checks on video. Most readers will feel the effect indirectly, through AI tools built on this kind of infrastructure. Google says the caching feature improves output per dollar.

What is not known yet

The announcement does not give prices or name the new regions. It also does not say how soon everyday AI products will get faster or cheaper. Those gains depend on how companies use the new servers.