Jensen Huang GTC 2026 Speech Summary: From RTX Spark to Five Key Takeaways on Backplane Stocks at a Glance

NVIDIA CEO Jensen Huang announced the arrival of the agentic AI era at GTC 2026, unveiling the first laptop chip RTX Spark and Vera supercomputer. During the event, the Taiwan supply chain list was also revealed, with companies like GigaDevice and Zhen Ding entering the list for the first time.

Jensen Huang's GTC 2026 Speech Highlights

NVIDIA CEO Jensen Huang recently completed his speech at GTC 2026 Taipei, held on the eve of COMPUTEX, officially declaring a new era where computing architecture shifts from traditional "app-centric" to "agentic AI (Agentic AI)."

This article summarizes the key points of Huang’s GTC 2026 speech, covering multiple breakthrough products, including the first announcement of the RTX Spark laptop chip, Nemotron 3 Ultra, as well as recent launches like Vera CPU and Vera Rubin supercomputer, and the latest NVIDIA backplane stocks.

RTX Spark Laptop Chip

RTX Spark is a super chip designed specifically for AI agent workloads, and it is NVIDIA’s first laptop chip. The chip combines NVIDIA’s Blackwell architecture GPU, uses TSMC’s 3nm process technology, and features a new Arm architecture "N1X" CPU co-designed with MediaTek.

In terms of performance advantages, RTX Spark demonstrates several breakthroughs:

Top-tier computing power and memory configuration

RTX Spark features 6,144 CUDA cores, with AI computing performance reaching 1 petaFlop, comparable to a laptop RTX 5070 graphics card. Its CPU and GPU are integrated into a single chip via NVLink, with connection speeds up to 600GB/s, and it supports up to 128GB of unified LPDDR5X memory.

Powerful local AI and software processing capabilities

RTX Spark is also designed for the Windows platform tailored for agents, offering a new agent-based AI service that can run large models with up to 120 billion parameters locally. Additionally, NVIDIA has deep collaborations with Adobe, significantly boosting AI features within Premiere, Photoshop, and other software, doubling video editing, color grading, and special effects processing speeds.

Exceptional multimedia and gaming performance

RTX Spark supports high-end 12K 4:2:2 video editing and allows gamers to enable ray tracing and DLSS in 1440p resolution while maintaining 100fps smooth gameplay. During the demonstration, Huang showcased Microsoft’s Forza Horizon and

Slim form factor and ultra-long battery life

While delivering powerful performance, it also ensures all-day battery life, enabling laptops to be as thin as 14mm, with stunning visual effects through stacked OLED G-SYNC panels.

RTX Spark has entered production and will debut on AI PCs from major brands like Microsoft, Dell, HP, ASUS, Lenovo, and MSI.

Image source: NVIDIA GTC 2026 live broadcast GTC 2026 speech highlights: First reveal of RTX Spark laptop chip

Vera CPU and Vera Rubin

Vera CPU is the world’s first processor designed specifically for AI agents’ latency and bandwidth needs. Huang claims its performance is 1.8 times that of traditional x86 architectures, providing extremely high generation speeds with ultra-low power consumption.

Vera Rubin is an AI supercomputer system composed of multiple new chips, utilizing NVLink 72’s ultra-high-speed wirelessly interconnected technology and full water cooling design, optimized for large-scale agentic AI workloads.

Image source: NVIDIA GTC 2026 live broadcast GTC 2026 speech highlights: Vera CPU and Vera Rubin

Nemotron 3 Ultra

Nemotron is an open-source large language model tailored for enterprise customization and high-performance needs, optimized for "tool usage."

Compared to existing leading open-source AI models, Nemotron achieves up to 5 times faster speeds and reduces costs by about 30%, making it a cost-effective and high-performance open model.

The newly released Nemotron 3 series adopts a groundbreaking "hybrid latent MoE (Mixture of Experts)" architecture, effectively assisting developers in building and deploying reliable multi-agent systems.

Huang announced the new series model Nemotron 3 Ultra, which features approximately 550 billion parameters in its inference engine, utilizing NVIDIA Blackwell’s highly efficient 4-bit NVFP4 training format, suitable for complex AI applications requiring deep research and strategic planning, and purportedly reducing costs by 30%.

Image source: NVIDIA GTC 2026 live broadcast GTC 2026 speech highlights: Nemotron 3 Ultra

Additionally, NVIDIA released training datasets containing 3 trillion tokens and the "Nemotron Agentic Safety Dataset" to help teams evaluate and enhance the safety of complex agent systems. Plans for subsequent versions Nemotron 4, 5, and 6 are also underway to promote continuous model evolution.

NVIDIA Enterprise AI Toolkit

To address system integration and security challenges faced by enterprises adopting agentic AI, NVIDIA launched a comprehensive "Enterprise AI Toolkit," enabling companies to integrate AI agents into existing workflows more easily and securely.

Open Shell is a key open-source component within the toolkit, providing a sandboxed secure environment specifically designed for safely executing and orchestrating AI agents, greatly simplifying enterprise deployment.

This toolkit allows enterprises not only to easily customize and deploy the aforementioned open models but also to deeply integrate these AI agents with existing enterprise workloads and internal data sources.

To accelerate enterprise development of proprietary AI agents, the ecosystem includes open-source libraries like NeMo Gym and NeMo RL, providing essential training environments; paired with NeMo Evaluator to verify model safety and performance. These tools enable development teams to obtain powerful reinforcement learning training capabilities more quickly and easily.

DSX AI Factory Platform and Domain-specific AI

DSX is a hardware and software reference platform for designing, simulating, and operating AI data centers, capable of dynamically balancing power and cooling, solving the traditional data center’s up to 40% power waste problem.

On the application side, Huang also introduced the Isaac GR00T platform for humanoid robots, the Kosmos 3 physical AI model combining vision and language, and the Alpamayo autonomous driving system that powers 80% of global automakers.

Image source: NVIDIA GTC 2026 live broadcast GTC 2026 speech highlights: DSX AI Factory platform and domain-specific AI

NVIDIA Ecosystem and Backplane Stocks Exposure

NVIDIA is actively collaborating with many top global tech companies:

• In the international market, NVIDIA and Microsoft are jointly reshaping the PC product line
• In server and infrastructure sectors, partnering with Alibaba, ByteDance, Meta, Oracle Cloud, CoreWeave, Nebius, and other cloud and system providers
• In AI models and software applications, working with OpenAI, Anthropic, SpaceX’s xAI, ServiceNow, and Adobe among heavyweight partners.

In Taiwan, the "Taiwan supply chain key companies" shown on Huang’s backboard drew significant attention from the capital markets. According to Ledger News, the list includes over a dozen Taiwanese firms:

• Chip manufacturing and system integration: TSMC, Foxconn, Quanta, Wistron.
• Cooling and power management: Delta Electronics, Chicony, Kentech, Taisys, Lite-On.
• High-speed transmission and connectors: Chaozhou, Moxa-KY, Liangwei.
• Precision mechanical parts and slides: Chuan Lake, Fushida, Nanjun International.
• Printed circuit boards (PCBs) and materials: Unimicron, Unimicron, Taiguang, and the newly listed and surged "Zhen Ding-KY."
• Other peripherals: First-time listed and hitting limit-up, the chassis manufacturer "Keda."

Interestingly, Huang’s favorite eateries such as Fubawang Pig Trotters, Huaniang Snack Bar, and Brick Kiln also appeared on the backboard. He even stopped at their logos during the speech, seemingly eager to recommend everyone try them out.

• Related report: "Huang Huang GTC 2026" NVIDIA backplane stocks exposed, Keda hits limit-up, and a certain KY stock surges 7%

From GPU Manufacturer to AI Factory, NVIDIA’s Next Step

As the world’s most valuable company, NVIDIA is reshaping computing in the AI era. Their vision is to make AI agents ubiquitous—spanning data centers, edge computing, PCs, automobiles, and robotic factories—turning AI into practical, autonomous reasoning, planning, and action assistants.

NVIDIA’s main goal is to maximize tokens per watt, through extreme co-design of chips, systems, and infrastructure, to transform traditional human-centric computers into powerful engines optimized for AI agent operation. This aims to create unparalleled economic value and productivity multipliers in this trillion-dollar new industrial revolution.

In this wave of AI, NVIDIA is no longer just a GPU chip supplier but has upgraded to a leader providing end-to-end AI systems and infrastructure, aiming to build ultra-performance "AI factories" that convert computing power into tangible economic productivity, leading a new industrial revolution.