NVIDIA Vera: 88 Arm Cores and Alleged Early Customers for the Next AI Platform

Key Takeaways

• Vera CPU packs 88 Olympus Armv9.2 cores, 176 threads
• Supports up to 1.5 TB LPDDR5X memory and 1.2 TB/s bandwidth
• NVLink‑C2C links to Rubin GPUs with 1.8 TB/s coherent bandwidth
• Alleged early adopters: CoreWeave, Meta, Oracle; Alibaba remains unverified
• Vera deepens NVIDIA’s control over CPU‑GPU‑memory stack

Pulse Analysis

NVIDIA’s Vera CPU marks a decisive shift toward Arm‑based silicon in the high‑performance AI data‑center market. Built on 88 in‑house Olympus cores that comply with the Armv9.2 ISA, the processor delivers 176 hardware threads and a massive 1.5 TB of LPDDR5X memory, offering up to 1.2 TB/s of bandwidth. The design prioritises energy efficiency and density, traits essential for inference workloads that run continuously at scale. By pairing the CPU with NVLink‑C2C, Vera can move data to Rubin GPUs at 1.8 TB/s, effectively eliminating traditional bottlenecks between host and accelerator.

The Vera chip is positioned as the host processor for NVIDIA’s Rubin platform, a tightly integrated stack that also includes Spectrum‑X networking, BlueField DPUs and the latest H100‑class GPUs. While NVIDIA has publicly linked CoreWeave, Meta and Oracle to the Rubin ecosystem, a Wccftech report adds Alibaba to the early‑buyer list—an assertion that remains unconfirmed amid U.S. export controls on advanced AI hardware. If the Chinese cloud giant does secure Vera racks, it would signal a notable exception to current trade restrictions and could reshape the competitive dynamics in Asia‑Pacific AI cloud services.

From a market perspective, Vera deepens NVIDIA’s vertical integration, giving the company leverage over both compute and memory supply chains. The reliance on LPDDR5X and the proprietary SOCAMM approach could pressure traditional server‑CPU vendors that still depend on DDR5 ecosystems. Moreover, customers that adopt a single‑vendor stack may enjoy lower latency and simplified procurement, but they also increase their exposure to NVIDIA’s pricing and roadmap decisions. Analysts will watch early deployment volumes closely, as they will indicate whether the industry is ready to entrust an entire AI super‑computer tier to one supplier.