Nvidia launches LPU and Vera CPUs with $20 billion Groq deal

Key Takeaways

• Nvidia introduced new Language Processing Units (LPUs) and Vera CPUs at GTC 2026, with LPUs incorporating technology from a $20 billion Groq acquisition.
• Analysts view Nvidia as transitioning into a systems company, selling entire data center pods rather than just individual components.
• Senators Elizabeth Warren and Richard Blumenthal are questioning Nvidia's $20 billion Groq deal over potential antitrust concerns.
• Nvidia predicts AI agents will create a "ChatGPT moment" for robotics, enabling more autonomous robots in various industries.
• SoftBank Group plans a 10-gigawatt AI data center in Ohio, potentially the world's largest, with infrastructure costs between $30 billion and $40 billion, requiring a $33 billion natural gas power plant.
• Synopsys launched new tools on March 11 to accelerate AI chip design verification, aiming to shorten manufacturing timelines.
• Next-generation AI PCs feature specialized NPUs, LPDDR5x memory, and advanced thermal systems for enhanced local AI task performance.
• HSBC Holdings Plc is reportedly considering job cuts as part of its AI overhaul, particularly in non-client-facing roles.
• NVIDIA released Nemotron-Cascade 2, an open-weight 30B Mixture-of-Experts model, demonstrating strong reasoning and agentic capabilities.
• Benzinga launched a video newsfeed for financial platforms and AI developers, offering tagged financial video content.

Nvidia unveils new AI chips and agentic tools at GTC 2026

Nvidia's annual GTC event, called the 'Super Bowl of AI,' showcased two new chip types: the Language Processing Unit (LPU) and the Vera central processing unit (CPU). The LPU uses technology from a recent $20 billion Groq acquisition and is designed to speed up GPUs. The company also highlighted the growing importance of agentic AI tools, which are more task-oriented than simple chatbots. Nvidia CEO Jensen Huang noted that agentic AI is at an 'inflection point,' driving new computing needs. The event also featured a sneak peek of the Kyber rack-scale architecture, expected in 2027.

Analysts discuss Nvidia's shift to systems company at GTC 2026

Experts believe Nvidia is evolving from just a chip company into a systems provider, as seen at the GTC 2026 event. Analysts like Jack Gold, Anshel Sag, and Leonard Lee noted Nvidia is now selling complete CPU racks, Groq LPUs, and networking solutions together. This scale is immense, with projections of reaching a trillion dollars in revenue. Nvidia is now selling entire data center pods rather than just individual components. This strategy emphasizes the value of their offerings at a large scale.

Senators question Nvidia's $20 billion Groq deal

Senators Elizabeth Warren and Richard Blumenthal are concerned about Nvidia's $20 billion deal with Groq. They sent a letter to Nvidia CEO Jensen Huang questioning if the deal was structured to avoid antitrust review. The senators worry this acquisition could strengthen Nvidia's market dominance and slow down innovation. They are seeking more information to understand the full impact of this major transaction.

Synopsys speeds up AI chip design with new verification tools

Synopsys launched new software and hardware tools on March 11 to speed up AI chip development, especially in verification. These tools link software with hardware emulation systems, allowing earlier design validation and reducing redesign risks. This helps shorten manufacturing timelines, as verification delays can impact production schedules. The new tools aim to help customers align chip availability with rapid AI model deployment.

AI PCs get performance boost from new hardware innovations

Next-generation AI PCs feature seven key hardware innovations for better performance and efficiency. These include specialized Neural Processing Units (NPUs) for AI tasks, advanced low-power memory like LPDDR5x for faster data transfer, and improved thermal systems using vapor chambers and smart sensors. New AI PCs also have dedicated privacy chips and secure enclaves to protect user data. These advancements allow laptops to run complex AI tasks locally, enhancing user experience and battery life.

Nvidia predicts ChatGPT moment for robotics with AI agents

Nvidia's vice president of robotics, Deepu Talla, believes AI agents will create a 'ChatGPT moment' for robotics. He stated that agentic AI systems are being built for digital environments first, allowing for rapid testing before real-world deployment. Talla sees AI agents as key to making robots more capable and autonomous. Nvidia is developing robots for manufacturing, logistics, and healthcare using these AI agents, aiming for a future where robots act as intelligent partners.

SoftBank plans massive 10 GW data center in Ohio

SoftBank Group plans to build a 10-gigawatt AI data center in Ohio, potentially the world's largest, with infrastructure costs between $30 billion and $40 billion. The project also requires a $33 billion natural gas power plant to meet its energy needs. The data center will be located in Piketon, Ohio, on a 3,700-acre complex. SB Energy will deploy gas turbines for power, with initial installations within a year and full deployment by the end of the decade. This project is part of a larger $550 billion U.S.-Japan investment program.

HSBC may cut jobs due to AI overhaul

HSBC Holdings Plc is considering significant job cuts as part of its artificial intelligence overhaul, according to a Bloomberg report. The largest impact is expected in global service centers for non-client-facing roles, though these discussions are in early stages. The evaluations began before the Middle East conflict, and no final decisions have been made. Some job reductions might also result from business sales. These potential cuts are part of a broader three-to-five-year plan for the company.

Nvidia releases new open-weight AI model Nemotron-Cascade 2

NVIDIA has released Nemotron-Cascade 2, an open-weight 30B Mixture-of-Experts model with 3B active parameters. This model focuses on 'intelligence density,' offering strong reasoning and agentic capabilities at a smaller scale than other advanced models. Nemotron-Cascade 2 achieved top performance in mathematical reasoning, coding, and instruction following benchmarks, outperforming models like Qwen3.5-35B-A3B. It uses techniques like Cascade RL and Multi-domain On-Policy Distillation (MOPD) for enhanced performance and supports 'Thinking Mode' for complex tasks.

Benzinga launches video newsfeed for financial platforms

Benzinga has launched a new video newsfeed for fintech platforms and AI developers, offering direct access to its financial video content. The feed includes market opens and closes, earnings coverage, and sector analysis, with each video tagged with equity tickers and topics. Platforms can access live content or a historical archive dating back to 2020. Benzinga states that video is increasingly important for user engagement and AI training in the financial industry.

Sources

• The Tech Download: Agentic tools and chips take center stage at Nvidia's 'Super Bowl of AI'
• The Five Nine: Nvidia GTC Highlights—Top AI & GPU Announcements Explained
• Nvidia’s $20 Billion Groq Deal Queried by Warren, Blumenthal
• Synopsys expands AI chip design stack with hardware-assisted verification, tightening development timelines
• 7 Hardware-Level AI Innovations in Next-Gen AI PC Laptops
• Nvidia Robotics Chief: AI Agents Will Bring About ChatGPT Moment for Robotics
• Planned 10-gigawatt Softbank data center in Ohio might be the largest in the world — will require a $33 billion natural gas plant, equivalent to nine nuclear reactors
• HSBC weighs deep job cuts as AI overhaul unfolds: report
• NVIDIA Releases Nemotron-Cascade 2: An Open 30B MoE with 3B Active Parameters, Delivering Better Reasoning and Strong Agentic Capabilities
• Benzinga Launches Video Newsfeed for Brokerages, Trading Platforms, and AI Developers