- Nvidia has shipped 'hundreds of thousands' of Grace standalone servers and over 2.5 million Grace CPUs to data centers globally [1] - SVP of hardware engineering Andrew Bell said a dozen manufacturing partners can produce up to 1,000 Vera racks per day [2] - Meta is already deploying Grace CPUs in standalone configurations at scale for general-purpose and agentic AI workloads [3] - Nvidia hand-delivered first Vera CPU systems to Anthropic, OpenAI, Oracle Cloud Infrastructure, and SpaceX [4] - Grace delivers 2x performance per watt on backend data center workloads versus incumbent x86 alternatives, per Nvidia VP Ian Buck
[3] Nvidia has shipped "hundreds of thousands" of Grace standalone servers to data centers worldwide, the company disclosed, revealing for the first time the scale of its Arm-based CPU penetration into hyperscale infrastructure. Ian Buck, Nvidia's VP and General Manager of Hyperscale and HPC, confirmed the figure alongside a total of over 2.5 million Grace CPUs shipped to date [1] [2].
The disclosure marks a strategic pivot for Nvidia, which has long been defined by its GPU dominance. Grace — a 72-core Arm Neoverse V2 processor supporting up to 480 GB of memory — has found adoption not only as a companion chip in GPU-accelerated servers but increasingly as a standalone CPU powering backend workloads that don't require GPU acceleration [1] [3].
Nvidia is now ramping production of Vera, its 88-core successor CPU with custom Arm cores, simultaneous multi-threading, and confidential computing capabilities. Andrew Bell, Nvidia's SVP of hardware engineering, stated that approximately a dozen manufacturing partners will be able to produce up to 1,000 Vera racks per day, signaling the company's intent to compete directly in the broader data center CPU market long dominated by Intel and AMD [2].
The Shipment Milestone #
The 'hundreds of thousands' figure refers specifically to Grace-only server racks — systems built around the Grace CPU without paired Nvidia GPUs. The broader 2.5 million Grace chip figure includes Grace CPUs deployed in GPU-accelerated configurations such as the GH200 Grace Hopper Superchip and GB200/GB300 NVL systems [1] [2].
The distinction matters because standalone CPU deployments represent a new addressable market for Nvidia. Rather than selling CPUs exclusively as companions to its high-margin GPU accelerators, Nvidia is now competing for general-purpose data center compute — a market Intel and AMD have historically owned. Buck characterized Grace as 'an excellent backend datacenter CPU' that 'can actually deliver 2x the performance per watt on those backend workloads' [3].
Meta as Anchor Customer #
Meta has emerged as the most prominent customer for Grace standalone deployments, having already deployed the CPUs in CPU-only server configurations at scale. The social media giant plans to extend this approach to Vera CPUs beginning in 2027, including leveraging Vera's confidential computing capabilities for private processing and AI features within WhatsApp's encrypted messaging service [3].
Meta's adoption of Nvidia Arm CPUs is notable given the industry trend toward custom in-house Arm processors — Amazon's Graviton and Google's Axion being the most prominent examples. Meta's decision to buy rather than build its own CPU represents a significant validation of Nvidia's CPU roadmap and contributes tens of billions of dollars to Nvidia's revenue alongside Meta's broader deployment of GB300 and Vera Rubin Superchips [3].
Vera Ramp and Production Scale #
Nvidia has already begun delivering Vera CPU systems to early customers. The company publicly confirmed hand-delivering the first Vera units to Anthropic, OpenAI, Oracle Cloud Infrastructure, and SpaceX — positioning the chip as foundational infrastructure for the leading AI labs [4].
The production target of up to 1,000 Vera racks per day, cited by Bell, reflects improved manufacturability and a diversified supply chain across roughly a dozen server manufacturing partners. Nvidia has framed this capacity not as a committed daily output rate but as evidence of supply-chain readiness for what it expects to be substantial demand from agentic AI deployments [2].
Vera features 88 custom Arm cores — a step beyond Grace's use of off-the-shelf Arm Neoverse V2 cores — along with simultaneous multi-threading and hardware-level confidential computing. Nvidia has stated it is working with over 300 global partners to ensure Vera supply meets demand [3] [5].
Market Context #
The Grace and Vera disclosures come as Nvidia's data center business continues to drive record revenue quarters. Nvidia shares traded at approximately $213.72 as of July 22, 2026, within a 52-week range of $164.07 to $236.54 [6].
Nvidia's CPU push directly challenges Intel and AMD in the server processor market while also competing with the custom Arm chip efforts of hyperscalers. The company's argument centers on offering a unified software stack — CUDA, NVLink, and the broader Nvidia AI Enterprise ecosystem — that spans both GPU-accelerated and CPU-only workloads, reducing the operational complexity of running heterogeneous data center fleets.
Oracle Cloud Infrastructure has also signaled plans to deploy hundreds of thousands of Vera units beginning in 2026, adding another hyperscale anchor tenant to Nvidia's CPU customer base [5].
Companies mentioned #
Further sources #
[1] Tom's Hardware — Nvidia has shipped 'hundreds of thousands of Grace standalone … ↗
[[2] WCCFTech — NVIDIA Aiming To Produce Up To 1000 Vera Racks Per Day After Shippin… ↗](https://wccftech.com/nvidia-to-produce-up-to-1000-racks-with-vera-cpus-per-day/)
[[3] The Register — Meta already deploying Nvidia's standalone CPUs at scale ↗](https://www.theregister.com/2026/02/17/meta_nvidia_cpu/)
[[4] Nvidia on X — Ian Buck hand-delivered first Vera CPUs to Anthropic, OpenAI, Spa… ↗](https://x.com/nvidia/status/2056494241904271780)
[[5] CNBC — Meta expands Nvidia deal to use millions of AI chips in data center buil… ↗](https://www.cnbc.com/2026/02/17/meta-nvidia-deal-ai-data-center-chips.html)
[[6] Stock Analysis — NVIDIA (NVDA) Stock Price History ↗](https://stockanalysis.com/stocks/nvda/history/)
The stories that matter, in one email. Free — unsubscribe anytime.