NVIDIA Enters AI Server CPU Market! Vera Officially Ships, Directly Challenging Intel and AMD

NVIDIA (NVDA-US), propelled by surging demand for artificial intelligence (AI) accelerators, has risen to become the world's most valuable company by market capitalization. Now, the company is extending its battlefield from graphics processing units (GPUs) to central processing units (CPUs), directly challenging Intel (INTC-US) and AMD (AMD-US), which have long dominated the server market.

On Tuesday (21st), NVIDIA released detailed specifications, architecture, and performance test data for its next-generation data center CPU, 'Vera,' enabling potential customers to evaluate the product. Company representatives stated that Vera chips have already been delivered in June to clients such as OpenAI, Anthropic, and SpaceX (SPCX-US), with OpenAI planning large-scale deployment starting this quarter.

NVIDIA is pursuing a vertical integration strategy, progressively increasing its in-house developed chips and system technologies. No longer merely selling individual GPUs, the company now offers complete computational power in full rack systems. NVIDIA believes that integrating CPU, GPU, memory, and networking technologies allows engineers to further unlock GPU performance, ensuring its systems remain the top choice for leading AI labs, even as competition intensifies from AMD's accelerators and custom AI chips.

Agent AI Drives CPU Demand; NVIDIA Estimates Market Potential at $200 Billion

Before the AI boom, CPUs were the most important and expensive components in servers. When ChatGPT launched in 2022, first-generation AI servers typically paired one CPU with up to eight GPUs, shifting industry focus gradually toward NVIDIA's accelerators.

However, the rise of agent AI—capable of autonomously executing tasks in the background with minimal human intervention—has once again elevated the importance of CPUs. CPUs must manage AI agents and continuously supply data, and their speed in processing individual queries directly impacts the utilization rate of expensive GPUs.

Ian Buck, Vice President of NVIDIA's Hyperscale Computing business, stated that agent AI has made CPUs an even more indispensable part of the system, particularly emphasizing the speed at which CPUs respond to individual queries.

Financial markets are already reflecting this trend. AMD and Intel's stock prices have surged 128% and 149% respectively this year, significantly outpacing NVIDIA's 8% gain. NVIDIA estimates the total server CPU market could eventually reach $200 billion; Bernstein projects the mature server CPU market will be worth approximately $37 billion by 2025.

Currently, Intel holds about 66.8% of the server CPU market, with AMD at around 33%, though AMD continues to expand its market share and has built strong partnerships with major cloud service providers. Kevin Knox, a Gartner analyst, believes AMD is currently the key competitor to beat in the enterprise AI server CPU market.

Wolfe Research estimates the average selling price of each Vera chip at around $5,000, with NVIDIA potentially shipping about 1.3 million units this year, though NVIDIA declined to comment on pricing.

Focus on Single-Core Performance: Vera Delivers 50% Higher Efficiency in Agent AI vs. x86 Chips

Vera is NVIDIA's first server CPU designed entirely in-house from the processor core, no longer directly adopting Arm's off-the-shelf core designs. Its proprietary Olympus core targets computational bottlenecks in agent AI, emphasizing single-core speed, memory bandwidth, and low latency—distinct from Intel and AMD's recent focus on increasing core counts.

NVIDIA claims Vera delivers 50% higher performance than x86 chips used by Intel and AMD when running agent AI workloads. Hannah Coutand, Vera's product marketing lead, explained that increasing per-core speed allows AI agents to return tasks to the GPU faster, keeping expensive, high-value GPUs in data centers at higher utilization rates.

Vera can be sold individually, integrated into liquid-cooled racks housing 256 chips, or configured with two Vera chips per server. This CPU will also integrate with NVIDIA GPUs to form the next-generation Vera Rubin platform.

Vera's power consumption ranges from 250 to 450 watts and uses low-power memory commonly found in laptops and smartphones, supporting up to 1.5TB per chip. Analysts believe Vera is not intended for traditional workloads like web services but is specialized for high-intensity AI tasks, creating a new CPU category not yet directly covered by Intel and AMD.

Nonetheless, despite NVIDIA's dominance in the AI market, convincing cloud service providers to adopt Vera remains a significant challenge. The company acknowledges the product is still in its early adoption phase, with Oracle (ORCL-US) currently the only major cloud partner publicly announced. Karl Freund, founder of Cambrian AI Research, believes NVIDIA developed Vera to reduce customer reliance on Intel or AMD CPUs while capturing revenue from this segment of server infrastructure.