iTranslated by AI
GTC 2025 Key Announcements Summary: A Simple Guide
GTC 2025 Summary: Even a Rabbit Can Understand
Hello everyone! In this article, I will introduce a summary of the major announcements from NVIDIA's annual developer conference, "GTC 2025," held in March 2025. I'll explain what's happening at the forefront of the AI revolution in a way that even a rabbit can understand, so please stick with me until the end!
What is GTC 2025? Reviewing the Basics
GTC (GPU Technology Conference) is the world's largest AI and high-performance computing technology conference hosted by NVIDIA. GTC 2025 was held from March 17 to 21 in a hybrid format, both in San Jose, USA, and online.
With the theme "AI's Future is Here," it became a large-scale event with approximately 25,000 participants on-site and about 300,000 online. There were over 1,000 lecture sessions and exhibitions from about 400 companies, showcasing the latest AI technologies and application cases.
The keynote speech by NVIDIA CEO Jensen Huang always attracts great attention, and this time, announcements regarding next-generation GPUs and new AI technologies were made.
Now, what kind of announcements were made at this GTC? Let's look at the main topics. Pyon-pyon!
Evolution of Next-Generation GPU Architecture
Details and Performance of Blackwell B300 Ultra
The highlight of GTC 2025 was the announcement of the next-generation GPU, "Blackwell Ultra (B300)." This GPU has the following characteristics:
- 30 PFLOPS of computing performance in FP4 (4-bit floating point) (approx. 1.5x compared to the current B200)
- Equipped with 288GB HBM3e memory (approx. 1.5x compared to the current B200)
- Power consumption (TDP): 1.4kW
- Adopts TSMC's N4P process
- Equipped with the latest Connect-X 8 NIV, achieving a total of 800GB InfiniBand communication
It boasts performance that significantly streamlines the development and execution of AI models such as Large Language Models (LLMs) that require large-capacity memory.
Outlook for the "Rubin" Architecture Scheduled for 2026
Furthermore, NVIDIA mentioned the next-generation GPU architecture "Rubin," which is scheduled to arrive in 2026. Rubin is named after astronomer Vera Rubin, and the following features have been announced:
- Adoption of TSMC's 3nm process
- Achieving 13TB/s memory bandwidth by migrating to HBM4 memory
- Support for 6th generation NVLink and the new Connect-X 9 networking standard
- The top-tier Rubin Ultra model is expected to achieve 100 PFLOPS performance and 1024GB memory capacity
Thanks to such high-end GPUs, the development of AI models is expected to accelerate even further. From a rabbit's perspective, that's an incredible leap!
Data Center Product Lineup and the Evolution of the DGX Series
NVIDIA also announced a new DGX series equipped with Blackwell GPUs:
- DGX Spark: A compact AI workstation for personal use that can be used at your desk.
- DGX Station: An enhanced personal AI supercomputer.
- DGX SuperPOD: A large-scale AI cluster for enterprises compatible with Blackwell Ultra.
These products will ensure that AI development environments are broadly covered, from the individual level to the enterprise level.
AI and Software-Related Announcements
Initiatives in Quantum Computing: "Quantum Day"
GTC 2025 featured the first-ever "Quantum Day," where initiatives toward the practical application of quantum computing were introduced. The event focused on quantum error correction and quantum simulation technologies, showcasing research and development cases in collaboration with industry leaders.
NVIDIA is focusing on advancing quantum computing technology by leveraging the computational power of High-Performance Computing (HPC). Applications are expected in fields such as materials science, financial modeling, and cryptography.
Evolution of AI Development Frameworks and Software Stacks
NVIDIA also announced numerous software tools to make AI development more efficient. They are focusing on providing more user-friendly and high-performance development environments, including the expansion of the CUDA-X platform and the AI-Q framework.
These tools will simplify the process from AI model development to implementation and deployment. It would be great if they become easier for beginners to handle—maybe even a rabbit will be able to understand them!
Robotics and Automotive Technology
Robotics (Physical AI) Platforms
CEO Jensen Huang stated that "the robotics industry is the next $10 trillion market," emphasizing initiatives related to robotics (Physical AI). He presented a vision where robot technology will become an important foundation supporting society amidst the deepening issues of a declining birthrate, aging population, and labor shortages.
NVIDIA is aiming to accelerate innovation in this field by providing software platforms that serve as the foundation for robot development.
Evolution of Omniverse and Application Cases
The evolution of the Omniverse platform is also a highlight. In particular, cases regarding virtual factory construction for the manufacturing industry were introduced, demonstrating the efficiency of 3D workflows using OpenUSD.
From Japan, Rikei took the stage at Japan AI Day with the theme "How to Create Next-Generation Virtual Factories with Omniverse and OpenUSD," introducing the processes and challenges of implementing digital twins in the manufacturing industry.
Overview of the AI Ecosystem
NVIDIA is evolving from a mere GPU manufacturer into a company that provides a comprehensive AI ecosystem. The figure below illustrates that ecosystem.
NVIDIA's strength lies in its comprehensive approach across diverse technological domains such as hardware, software, quantum computing, robotics, and Omniverse. This is driving AI adoption across various industries, including healthcare, automotive, finance, manufacturing, cloud, and media.
Impact on the Industry and Future Outlook
Potential for a Second Wave of the AI Boom
From the announcements at GTC 2025, it can be seen that NVIDIA is trying to create a new wave of the AI boom. With the introduction of Blackwell Ultra (B300), it will be possible to develop even larger and higher-performance AI models, and the industrial application of AI is expected to accelerate further.
In particular, by focusing on speeding up inference models, the practicality and adoption of AI will likely progress. While training has been the focus so far, the efficiency of inference is expected to become an important theme in the AI industry moving forward.
Competitor Status and Ecosystem Expansion
NVIDIA maintains approximately 82% of the GPU market share, but AMD and Intel are also accelerating their development of GPUs for AI. However, NVIDIA has gone beyond being just a hardware vendor and has built a comprehensive AI ecosystem that includes software, development tools, and industry-specific solutions.
This comprehensive approach is what strengthens NVIDIA's competitive advantage.
Collaboration with Cloud Providers and Accelerated AI Adoption
Collaboration with cloud providers is also advancing, accelerating the adoption of AI technology across various industries. Specifically, the number of use cases in fields such as healthcare, finance, manufacturing, and automotive is increasing, and the practical application of AI is moving forward.
As a result, AI is becoming a crucial element that determines a company's competitiveness.
Summary
At GTC 2025, numerous announcements were made showing how NVIDIA is accelerating the evolution of AI technology in the fields of hardware, software, and industrial applications.
Particularly noteworthy are:
- Mentions of the high-performance GPU "Blackwell B300 Ultra" and the next-generation "Rubin"
- Full-scale initiatives in quantum computing
- Focus on robotics and the evolution of Omniverse
With these technological advancements, the scope of AI applications will expand even further, and innovative solutions are expected to emerge in more industries.
True to the theme "AI's Future is Here," it can be said that NVIDIA has presented a blueprint for the next stage of the AI revolution. Pyon-pyon!
Related Events in Japan
In Japan, the "GTC 2025 Follow-up Webinar" is scheduled to be held as a summary of GTC 2025. This will feature explanations by experts from NVIDIA's Japanese subsidiary and introductions of AI application cases by domestic companies. Please check the NVIDIA Official Website for details.
Additionally, as part of GTC 2025, a special event for Japanese users called "Japan AI Day" is also being held, where case studies from Japanese companies are introduced.
Discussion