Nvidia says Nebius, Nscale and Verda are building AI cloud services with Vera CPUs

Nvidia said Nebius, Nscale and Verda are building AI cloud services with its Vera CPUs, and that five server makers are building Vera-based systems. The post matters to BlockWest readers because it positions the CPU, not only the GPU, as a purchase decision for AI infrastructure, but it discloses no volumes, prices or performance numbers.

• Nebius, Nscale and Verda are building with Vera to support business and developer computing needs.
• Bull, Dell Technologies, HPE, Lenovo and Supermicro are building Vera-based servers, standalone or alongside Nvidia GPUs.
• Watch for disclosed deployment volumes, pricing and availability dates, none of which the release provides.

Nvidia published the post on its newsroom. It is a corporate blog entry, not a regulatory filing, and it carries no financial terms. It lists customers and partners by category and describes their intended or early use of Vera.

Cloud providers and server makers are named, with no order sizes

The post says AI cloud providers Nebius, Nscale and Verda “are building with Vera” to support the computing needs of businesses and developers. It says the CPUs help these providers scale the data processing and software execution behind their services, alongside the GPUs that run models. Aleria, a sovereign AI infrastructure and platform provider, is described as part of the wider sovereign AI ecosystem.

On hardware, Bull, Dell Technologies, HPE, Lenovo and Supermicro are building Vera-based systems. Nvidia says Vera can power standalone CPU servers or work with Nvidia GPUs in accelerated systems. The post does not say when any of these systems ship, or how many units any provider has ordered.

Software firms report early results, not benchmarks

ClickHouse, Cloudera, Duck Labs, Elastic, Polars, Redpanda, SAP and Starburst “observed strong results” in early testing, according to Nvidia, on analytical, extract-transform-and-load and semantic search workloads. Nvidia credits Vera’s per-core performance, LPDDR5X memory bandwidth and Scalable Coherency Fabric. No speedup, latency or cost-per-task figure is given, and no baseline CPU is named.

Atlassian is “looking to” use Vera to speed the data processing that builds semantic search indexes for its AI products. Inherent has tested Vera and is looking to adopt it for agent-assisted scientific research. Nous Research uses Vera to authenticate agent identities and validate each action before it executes.

Sandbox providers Baseten-Blaxel, Daytona, E2B, Modal, Prime Intellect and Vercel are building on Vera. Nvidia says sandboxes “spin up millions of times a day,” so startup speed compounds at scale.

Each step depends on fast CPU processing.

Nvidia blog post

What changes for AI buyers, and what is left open

The argument in the post is that agent workloads, which retrieve data, run tools and execute code, put load on CPUs that GPU-centric clusters were not sized for. If that holds, operators that buy GPUs also face a host-CPU decision, and Nvidia is offering Vera in standalone servers from five manufacturers as well as in accelerated systems. The named data-processing vendors suggest the target includes analytics and search estates, not only model serving.

The post does not answer the cost question. It gives no price, no power draw despite calling the CPUs energy-efficient, and no comparison with a prior Nvidia CPU or a rival chip. The adoption language also varies: Nebius, Nscale and Verda are “building,” while Atlassian and Inherent are “looking to” adopt, so the customer list mixes commitments with intentions.

The post also lists related announcements by title only: cuDSS general availability, expanded confidential computing, an expanded Agent Toolkit and a Spark release for Europe. It gives no detail on them in this text.

The BlockWest read. We read the post as Nvidia extending its sales pitch from the accelerator to the whole server. Standalone CPU systems from five manufacturers mean Vera competes for host-processor budgets as well as riding along with GPU orders. Sandbox providers launching millions of instances daily are the first proof point, and the post offers no numbers for it. Buyers should treat the partner list as intent until volumes, prices and benchmarks appear.

Nvidia has not said when Vera systems ship or in what volume, and Atlassian and Inherent are still described as “looking to” adopt.

BlockWest is a news publication. Nothing here is investment advice. Read our disclaimer and editorial policy.