NVIDIA Vera CPU to debut at Computex 2026 with 1.5x x86 performance

NVIDIA will unveil its Vera CPU at Computex 2026, promising 1.5 times the performance of competing Intel and AMD x86 servers on AI workloads [2]. The chip packs 88 custom Arm v9.2 “Olympus” cores on a monolithic die, delivering up to 1.2 TB/s memory bandwidth and using Spatial Multithreading to double thread count to 176 [2].

Analysts at GF Securities project Vera shipments to climb from roughly 1.2 million units in FY 2027 to 4.2 million in FY 2028, citing the processor’s claimed efficiency edge and a $200 billion market opportunity for Arm‑based data‑center CPUs [2]. NVIDIA’s CEO Jensen Huang is slated to present the product on June 1, a day before Computex opens, underscoring the company’s confidence that Vera will reshape server architecture [2].

AMD contests the claim, noting that its 192‑core EPYC Turin already offers nearly 2.5 × the performance‑per‑watt of Vera in modeled rack‑level tests, and that its upcoming 256‑core “Venice” EPYC could widen the lead to 330 % [1]. Those figures are based on internal modeling rather than real‑world benchmarks, and AMD warns that Vera’s Arm design may pose software‑compatibility challenges despite higher bandwidth [1].

Intel’s recent Computex remarks echo the market shift toward CPU‑heavy agentic AI workloads, with CEO Lip‑Bu Tan admitting supply constraints on Xeon 6+ chips as demand surges [3]. The company’s new 18 nm Xeon 6+ rack promises 36,864 cores but struggles to meet the influx of AI‑oriented orders, highlighting the broader industry pressure to deliver higher‑density, AI‑optimized silicon.

If Vera’s performance claims hold up under independent testing, the processor could accelerate the migration of hyperscalers from x86 to Arm, reshaping the data‑center landscape. The key test will be whether third‑party benchmarks confirm the 1.5× advantage and whether supply chains can meet the projected demand.