Quick Summary
- At GTC 2026, Nvidia introduced the Groq 3 LPU, a specialized processor designed exclusively for AI inference operations
- The company’s new LPX server configuration features 128 Groq 3 chips and delivers up to 35x improved throughput per megawatt when combined with Vera Rubin NVL72
- A standalone Vera CPU rack was revealed, marking Nvidia’s direct entry into competition with Intel and AMD’s data center processor market
- The Vera processor targets agentic AI applications, including autonomous web navigation and file data retrieval
- The company’s data center segment generated $193.5 billion in revenue during fiscal 2026, representing significant growth from the previous year’s $116.2 billion
At this year’s GTC conference in San Jose, Nvidia presented a comprehensive portfolio of new processors and server configurations that significantly expand the company’s reach beyond its traditional graphics processing strengths.
The star announcement was the Groq 3 language processing unit — commonly referred to as an LPU. Following a massive $20 billion acquisition completed last December, Nvidia licensed Groq’s technology and welcomed key personnel including founder Jonathan Ross and president Sunny Madra into its organization.
The Groq 3 processor focuses exclusively on inference workloads — the operational phase of AI that occurs post-training. Whenever users interact with AI chatbots and receive responses, they’re experiencing inference in action. This segment represents a rapidly expanding portion of the AI ecosystem, where purpose-built processors can outperform versatile GPUs.
According to Nvidia VP of hyperscale and HPC Ian Buck, the Groq 3’s memory architecture delivers superior speed compared to Nvidia’s GPU memory systems, though GPUs maintain an advantage in capacity. The strategy involves leveraging the benefits of both technologies.
This dual-chip approach underpins the LPX server rack — a configuration featuring 128 Groq 3 LPUs. When deployed alongside the Vera Rubin NVL72 rack, Nvidia claims customers can achieve 35x greater throughput per megawatt and unlock 10x more revenue potential. The system targets trillion-parameter models and supports million-token context windows.
Vera CPU Enters Intel and AMD Territory
Another major announcement centered on the Vera CPU rack. While the Vera processor was previously mentioned as a component of the Vera Rubin superchip — which combines one Vera CPU with dual Rubin GPUs — Nvidia has now repositioned Vera as an independent processor solution.
The newly introduced rack integrates 256 liquid-cooled Vera processors into one unified system. Nvidia positions it as the optimal CPU for agentic AI applications — autonomous systems that navigate websites, extract information from documents, or execute complex multi-step operations independently.
“We’ve designed a new kind of CPU, the Olympus core, engineered by NVIDIA for AI execution,” Buck explained. The Vera processor also handles data mining, personalization, and contextual analysis that supports AI model operations.
This move positions Nvidia as a direct challenger to Intel and AMD in the data center CPU arena — a market segment these two companies have controlled for decades.
In recent weeks, Nvidia secured an agreement with Meta for the largest-ever deployment of its previous-generation Grace CPUs. The Vera announcement extends this strategic direction.
Additional GTC Hardware Announcements
The conference also featured presentations of the Bluefield-4 STX storage rack and Spectrum-6 SPX networking rack, completing Nvidia’s comprehensive data center infrastructure offering.
Major cloud providers including Amazon, Google, Meta, and Microsoft plan to invest a combined $650 billion in AI infrastructure throughout this year.
Nvidia’s data center division reported revenue of $193.5 billion for fiscal 2026, showing substantial growth from fiscal 2025’s $116.2 billion.
Wall Street analysts maintain a consensus Strong Buy rating on NVDA stock, with 38 Buy ratings and one Hold rating issued over the last three months, alongside an average price target of $273.61.
