Nvidia Vera CPU Arrives: Benchmarks, Customers, and the $20 Billion Bet on Agentic AI

Nvidia's push into the CPU market is no longer a rumor — the Vera processor is in customer hands and in full production. This isn't just another server chip. Vera is purpose-built for the new class of AI workloads known as agentic AI, where models don't just answer queries but execute code, use tools, and run complex simulations. Here is everything you need to know about Nvidia's Vera CPU, backed by sources and independent benchmarks.

What Is the Nvidia Vera CPU?

The Vera CPU is Nvidia's first-ever standalone data center microprocessor, announced at GTC San Jose in March 2026 . Jensen Huang called it the company's "next multibillion dollar business" . It is built around 88 custom Olympus Arm cores on a monolithic die — not a chiplet design — with 1.2 TB/s of LPDDR5X memory bandwidth and a 3.6 TB/s on-chip fabric . Nvidia claims it has the world's highest instructions per clock (IPC), executing up to 10 instructions per clock cycle .

Unlike Nvidia's Grace CPU, which was designed as a companion to its GPUs, Vera can operate as a standalone CPU for AI factory workloads such as code execution, sandboxing, analytics, data pipelines, and agent orchestration .

Timeline: From Announcement to Full Production

• March 2026 (GTC San Jose): Vera announced as Nvidia's first standalone data center CPU .
• May 18, 2026: First shipments hand-delivered to three leading AI labs — Anthropic, OpenAI, and SpaceX — followed by a delivery to Oracle Cloud Infrastructure . Nvidia VP Ian Buck personally made the deliveries to the first four customers .
• May 31–June 1, 2026: Jensen Huang confirmed full production during his Computex 2026 keynote and GTC Taipei event .
• H2 2026: The Vera Rubin platform, a full-stack system combining Vera CPU + Rubin GPU + five other chips, is targeting availability in the second half of 2026 .

Performance: How Vera Compares to Intel and AMD

Nvidia's Official Claims

Nvidia states that Vera completes agentic AI tasks 1.8x faster than x86 CPUs and delivers 80% faster sandbox environment performance than traditional CPU infrastructure . These claims are vendor-stated and should be treated as directional until more independent data emerges .

Independent Phoronix Benchmarks (May 26, 2026)

Phoronix published the first independent benchmarks of the 88-core Vera CPU :

• vs. Intel Xeon 6980P: Vera scored 1.55x faster in geometric mean across a range of workloads .
• vs. AMD EPYC 9575F: Vera was roughly 10% ahead .
• vs. Nvidia's own Grace CPU: Vera delivered a 1.6x geometric mean improvement .
• Memory bandwidth: Sustained 90% of peak in STREAM TRIAD test .
• Linux kernel compile: 20 seconds — about 2x faster per core than a 128-core x86 chip .

The Phoronix verdict, in the reviewer's own words: this is "the most formidable competition to Intel and AMD x86_64 processors ever realized" .

Perplexity's Real-World Test

Perplexity's Vice President of Infrastructure Nate Kupp said Vera processes AI agent coding tasks at about 1.5x the speed of traditional CPUs. Kupp noted that "Vera really stood out to us as just like a dead-on fit for a lot of the core work" .

Revenue Expectations: $20 Billion in Sales This Year

Nvidia expects $20 billion in Vera CPU sales by the end of its current fiscal year (end of January 2027) . This forecast is separate from Nvidia's earlier $1 trillion projection for its Blackwell and Rubin AI chips .

Nvidia CFO Colette Kress described Vera as opening a "brand new $200 billion opportunity," adding that "every major hyperscaler and system maker is partnering with us to get it deployed" . The company's overall Q2 revenue forecast of $91 billion beat Wall Street estimates of $86.84 billion .

Who Is Buying the Vera CPU?

Based on Nvidia's official confirmation of first shipments and subsequent disclosures, known early customers and partners include :

• Anthropic
• OpenAI
• SpaceX (SpaceXAI)
• Oracle Cloud Infrastructure
• ByteDance
• CoreWeave
• Dell

Nvidia has also begun making Vera sales pitches to Chinese enterprise clients, though any China-related revenue is not included in the current Q2 forecast .

Perplexity's Adoption Plans

Perplexity confirmed on July 7, 2026 that it plans to use Nvidia's Vera CPUs to support its AI agent workloads . Perplexity is already using Nvidia's Nemotron 3 Ultra for its search and Perplexity Computer product . The company did not disclose planned purchase volumes .

The Vera Rubin Platform: What Comes Next

Vera is also the centerpiece of the Vera Rubin platform, a full-stack AI supercomputer that combines the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch . The platform is designed to slash training time and inference token costs for agentic AI workloads. Production of Vera Rubin systems began in Q2 2026 across more than 350 factories in 30 countries, with systems expected to ship in the second half of 2026 .

Key Takeaways

• Vera is real and shipping: First deliveries went out in May 2026 to major AI labs, and full production was confirmed in June .
• It beats x86 on benchmarks: Independent tests show Vera outperforming Intel and AMD flagship chips by 10–55% depending on the workload .
• The financial bet is huge: $20 billion in revenue expected this fiscal year, targeting a $200 billion addressable market .
• Early customers are AI-native: The first wave of adopters are companies like OpenAI, Anthropic, and Perplexity that run agentic workloads at scale .
• The Vera Rubin platform is the long-term play: Vera is one chip in a seven-chip system designed to dominate AI data center infrastructure .