Key Highlights
- Groq 3 LPU debuts at GTC 2026, designed exclusively for AI inference applications
- New LPX server configuration features 128 Groq 3 processors, delivering 35x improved throughput per megawatt when combined with Vera Rubin NVL72
- Standalone Vera CPU rack positions Nvidia as a competitor to Intel and AMD in the data center processor market
- Vera CPU targets agentic AI applications, including web browsing automation and file data retrieval
- Data center revenue reached $193.5 billion in fiscal 2026, representing significant growth from $116.2 billion in the previous fiscal year
Nvidia presented its latest hardware innovations at GTC 2026 in San Jose this Monday, showcasing a portfolio that extends the company’s reach far beyond traditional GPU offerings.
The centerpiece announcement featured the Groq 3 language processing unit, commonly abbreviated as LPU. The chip emerged from Nvidia’s licensing agreement with Groq, which included bringing founder Jonathan Ross, president Sunny Madra, and additional team members into the fold through a $20 billion acquisition completed in December.
Groq 3 specializes in inference operations — the production phase of AI that occurs post-training. Each interaction where a user submits a query to a chatbot and receives an answer represents inference. This segment of the AI industry continues to expand rapidly, creating opportunities for specialized processors to outperform general-purpose GPUs.
Nvidia VP of hyperscale and HPC Ian Buck explained that Groq 3 features memory speed advantages over Nvidia’s GPU offerings, though GPUs maintain higher memory capacity. The strategy involves leveraging the complementary strengths of both technologies.
This approach manifests in the LPX server rack — a configuration containing 128 Groq 3 LPUs. When deployed alongside the Vera Rubin NVL72 rack, Nvidia claims customers can achieve 35x greater throughput per megawatt alongside 10x enhanced revenue potential. The system targets trillion-parameter models and million-token context windows.
Vera CPU Enters Data Center Processor Arena
The second major announcement centered on the Vera CPU rack. Previously, the Vera chip appeared in discussions about the Vera Rubin superchip — an integrated package combining one Vera CPU with two Rubin GPUs. Nvidia now offers Vera as an independent processor option.
The dedicated rack integrates 256 liquid-cooled Vera chips within a unified system. Nvidia positions it as the optimal CPU for agentic AI — autonomous systems that navigate websites, extract information from documents, or execute complex multi-step workflows without human intervention.
“We’ve designed a new kind of CPU, the Olympus core, engineered by NVIDIA for AI execution,” Buck said. Vera also plays a role in data mining, personalization, and context analysis that feeds into AI models.
This move establishes Nvidia as a direct competitor to Intel and AMD in the data center CPU segment — a market these two companies have controlled for decades.
Nvidia announced a partnership with Meta last month to implement its previous-generation Grace CPUs at unprecedented scale — marking the largest deployment of its kind. The Vera introduction amplifies that trajectory.
Additional GTC 2026 Hardware Announcements
Nvidia’s presentation included the Bluefield-4 STX storage rack and the Spectrum-6 SPX networking rack, completing a comprehensive data center hardware ecosystem.
Major hyperscalers such as Amazon, Google, Meta, and Microsoft plan to invest a combined $650 billion in AI infrastructure throughout this year.
Nvidia’s data center segment generated $193.5 billion in revenue during fiscal 2026, climbing from $116.2 billion in fiscal 2025.
Wall Street analysts tracking NVDA assign a consensus Strong Buy rating, with 38 Buy recommendations and one Hold rating issued over the past three months, alongside an average price target of $273.61.
