This IT Home page is stamped 2026/10/22 14:03:24. Google Cloud said on the 21st, local time, that it plans to extend infrastructure based on NVIDIA's Vera Rubin platform across compute sites, with up to 960,000 Rubin GPUs. The judgment first: 960,000 is the ceiling written on this page. How many are already deployed, which sites they sit in, and from which day a customer can use them at that scale are not printed.

[1]
A cooling manifold, charcoal pipes, a muted blue header, one small oxide-red cap
A cooling manifold, charcoal pipes, a muted blue header, one small oxide-red cap., AI-generated illustration, not a news photograph

The next sentences on the same page carry two timelines, and this story deletes neither. Google Cloud also said that day that A5X instances based on the NVIDIA Vera Rubin NVL72 rack system are formally live. The page then says those instances use NVIDIA ConnectX-9 SuperNICs, Google Cloud's Virgo network, and fourth-generation advanced liquid cooling, and will be formally launched next month. Formally live comes first. Launch next month comes after. The first sentence is not rewritten as not yet open, and the second is not rewritten as already open. Keeping only one of them would misstate the supply status.

[1]

The page also names two confidential-computing offerings. Both are statements about what will be offered. Google Cloud will be among the first cloud providers to offer rack-scale NVIDIA confidential computing on bare-metal instances. Among the first is this page's wording, not a published list of vendors. Separately, Google Cloud will also formally launch confidential computing on G4 virtual machines that carry NVIDIA RTX PRO 6000 Blackwell server GPUs. The hardware named there is the Blackwell server GPU, not the same sentence as the count of Vera Rubin GPUs. How to configure it, how isolation works, and how keys are handled are not on this page, and they are not added here.

[1]

Rack-scale confidential computing on bare metal and confidential computing on G4 virtual machines are two different offerings. The first is written onto the Vera Rubin rack system. The second is written onto RTX PRO 6000 Blackwell server GPUs. The page does not say which of the two already has customers using it. The hard number this page supports is therefore the ceiling of 960,000 Rubin GPUs. The times that can be checked are this page's stamp and the local 21st it reports. Whether A5X is already open to customers, or waits until next month, is not resolved by the page itself. Calling the liquid cooling fourth-generation is this page's wording, with no separate comparison attached.

[1]