Tech & AI
Google Cloud to open Nvidia Vera Rubin instances to all customers next month, with plans for up to 960,000 GPUs
The A5X rental servers, built on Nvidia's newest AI chip platform, will become generally available in November, with a European rollout to follow.
HedgeWeekly Tech & AI Desk ·
Alphabet's Google Cloud and Nvidia said on Wednesday that Google's A5X instances, rented computing capacity running on Nvidia's Vera Rubin systems, will be generally available next month, meaning any customer can sign up rather than only invited testers. The companies made the announcement at Nvidia's GTC conference in Berlin. Google Cloud plans to scale its Vera Rubin infrastructure to as many as 960,000 Rubin graphics processors across multiple sites, a target first set out when the A5X line was unveiled in April.
The instances pair Vera Rubin rack-scale systems and Nvidia ConnectX-9 network cards with Google's own Virgo data-centre network and liquid cooling, and are aimed at training AI models, running them for users (inference) and reinforcement learning, a training method in which models learn by trial and reward. Google Cloud said it intends to expand availability in Europe and to bring Nvidia's confidential-computing protection, which keeps data encrypted while it is being processed, to bare-metal servers.
London-based Ineffable Intelligence, founded by AlphaGo creator David Silver, was the first customer to use A5X, for reinforcement-learning research. Advertising group WPP said it had tested its creative AI agents on the platform and that tasks which usually needed several linked servers ran on a single A5X instance. Alphabet shares were down about 1% in premarket trading after closing at $336.40 on Tuesday, while Nvidia was up nearly 2% from its $216.47 close, according to RTTNews.
Why it matters: general availability is the step at which Vera Rubin moves from early access to broad commercial rental at one of the largest cloud providers, a key test of how quickly Nvidia's next platform turns into revenue. Google is at the same time promoting its own TPU chips, so the scale of its Rubin build-out is a signal of how far hyperscalers intend to keep buying Nvidia alongside in-house silicon. Investors will look for detail on Rubin ramp timing and cloud capital spending in upcoming earnings from Alphabet and Nvidia.