Nvidia’s GTC 2025 keynote: 40x AI performance leap, open-source ‘Dynamo’, and a walking Star Wars-inspired ‘Blue’ robot

San Jose, California – NVIDIA CEO of NVIDIA, the theater at the SAP center, took a healthy leather jacket without remote elements, to provide what has become one of the most expected major tools in the technology industry. GPU (GTC) 2025, which Huang described as “Super Bowl of Ai”, is a critical turn for NVIDIA and the broader artificial intelligence sector.

“What a wonderful year, and we have a lot of amazing things we are talking about,” Huang told The PlGH Arena, and he addresses a fans that grew in a great way as artificial intelligence has turned from a specialized technology to a basic power that reshapes the entire industries. The risks were especially high this year after the market disturbance caused by the launch of the Chinese Dibsic emerging for the high -efficiency R1 thinking model, which led to a decrease in NVIDIA shares earlier this year amid fears of low demand for expensive graphics processing units.

]

On this background, Huang presented a comprehensive vision for the NVIDIA future, focusing on a clear computing road map in the data center, progressing in the possibilities of thinking about artificial intelligence, and bold moves to robots and independent vehicles. The presentation showed a picture of a company that works to maintain its dominant location in the infrastructure of Amnesty International with expansion in new areas where its technology can create value. NVIDIA shares were traded during the presentation, as it closed more than 3 % per day, indicating that investors may have hoped for more dramatic ads.

But if Huang’s message is clear, this is: Amnesty International is not slowing down, nor is Nafidia. From the leading chips to pushing to material artificial intelligence, here are the five most important food from GTC 2025.

Blackwell Platform increases production with 40x performance on the voltage

The axis of the computing strategy of artificial intelligence in NVIDIA, Blackweell, is now in “full production”, according to Hanging, who emphasized that “customer request is incredible.” This is an important teacher after Huang described as “hiccups” in early production.

Huang made an amazing comparison between Blackwell and her predecessor, Huber: “Blackwell Nevlink 72 with Dynamo is 40 times the performance of the AI ​​factory from Hopper.” This performance leap is very important to the burden of inference work, which Huang has placed “one of the most important work burdens in the next decade with the expansion of artificial intelligence.”

Performance gains come in a crucial time for the industry, as artificial intelligence models such as Deepseek’s R1 require a much larger account than traditional large linguistic models. Huang explained this with a demonstration comparing the traditional LLM approach to the arrangement of wedding sitting (439 symbols, but error) versus the thinking model approach (approximately 9000 symbols, but right).

“The amount of account we must do in artificial intelligence is much greater due to the logic of intelligence and training in artificial intelligence systems and client systems,” Huang explained, addressing directly the challenge posed by more efficient models like Deepseek. Instead of developing effective models as a threat to the NVIDIA business model, HUANG frams it as it has pushed the demand for the account – about converting potential weakness into strength.

From the next generation, Rubin’s structure was detected with a clear multi -year road map

In a clearly designed step to give institutions and cloud services providers confidence in the long -term NVIDIA path, HUANG has developed a detailed road map for the infrastructure of artificial intelligence computing until 2027. This is an extraordinary level of transparency about the future products of the hardware company, but reflects the long planning lines required to reflect AI.

“We have an annual rhythm of the road maps that have been developed for you so that you can plan your Infrastructure International Infrastructure,” Huang stressed, stressing the importance of the ability to predict customers who make huge capital investments.

The upcoming Blackwele Ultra road map includes the second half of 2025, providing 1.5 times from Amnesty International from the current Blackwell chips. This will be followed by Vera Robin, which was called the name of the astronomy that discovered the dark matter, in the second half of 2026. Robin will contain the new CPU that reaches twice faster than the current GRACE CPU, along with the structure of the new networks and memory systems.

“Essentially, everything is completely new, except for the structure,” Huang explained about the Vera Robin platform.

The road map extends to the Ultra in the second half of 2027, which Huang described as a “severe scale” that offers my account 14 times more than the current systems. “You can see that Robin will lead to a significant decrease in the cost,” he indicated, indicating concerns about the economies of Amnesty International.

This detailed road map works as an answer to NVIDIA on market concerns about competition and sustaining artificial intelligence investments, as customers and investors effectively tell that the company has a clear path forward regardless of how the efficiency of the artificial intelligence model develops.

Nvidia Dynamo appears as the “operating system” of artificial intelligence factories

One of the most important ads is Nvidia Dynamo, an open source software system designed to improve artificial intelligence inference. Huang described it as “the operating system mainly for the artificial intelligence factory”, drawing parallel to how traditional data centers depend on operating systems such as VMWare to regulate institutions applications.

Dynamo deals with the complex challenge of managing the burdens of artificial intelligence work through the distributed GPU systems, dealing with tasks such as parallel pipeline, tensioner parallel, expert parallel, gathering on the plane, uncontrolled inference, and managing the work burden. These technical challenges have become of increasing importance as artificial intelligence models grow more complicated and logic -based methods require more account.

The regime gets his name from Dynamo, which Huang has noticed is “the first tool that started the last industrial revolution, the industrial energy revolution.” The comparison is placed as a foundation technique for the revolution of artificial intelligence.

By making Dynamo Open Source, NVIDIA tries to enhance its ecosystem and make sure that its devices remain the preferred platform for the burdens of artificial intelligence work, even with increasing the importance of improving programs for performance and efficiency. Partners, including confusion, already works with NVIDIA to implement Dynamo.

“We are very happy that many of our partners are working with us,” said Huang.

The open source approach is a strategic step to maintain the central status of NVIDIA in the environmental system of Amnesty International, while recognizing the importance of improving programs in addition to the performance of raw devices.

Artificial intelligence and robots at the lead center with an open source GROOT N1 model

In what may be the most amazing moments in the keyword, Huang revealed a big boost in robots and physical realism, and reached its climax with the appearance of “blue”, a robot inspired by the stars war that was walking on the stage and interacted with Huang.

Learn about Blue (Star Wars Droid) after announcing the NVIDIA Partnership with DeepMind and Disney. pic.twitter.com/ylcdouf5xc

Brianroemmele March 18, 2025

“By the end of this decade, the world will be at least 50 million workers,” Huang explained, as robots as a solution to the lack of global employment and the huge market opportunity.

The company has announced NVIDIA ISAAC GROOT N1, described as “the first open institution model in the world and is fully customized for general thinking and human skills.” Make this open source the model is a big step to accelerate the development in the field of robots, similar to how the open source LLMS has rushed to develop AI.

Besides GROOT N1, NVIDIA has announced a partnership with Google DeepMind and Disney Research to develop Newton, an open source physics engine for robots simulation. Huang explained the need for “a physics engine designed for fine grains, hardness and softness, designed to be able to train touch notes, fine motor skills and operator controls.”

The focus on robot training simulation follows the same style that has proven successful in developing independent leadership, using artificial data and learning to reinforce to train artificial intelligence models without restrictions on collecting material data.

“Use omniverse to Cosmos, and the universe to generate an endless number of environments, allows us to create and control data on which to control, and yet systematically exclusively at the same time,” Huang explained, describing how NVIDIA simulation techniques enabled robot training on a large scale.

These robot ads represent NVIDIA to the traditional computing of artificial intelligence in the material world, which may open new markets and applications for their technology.

GM partnership indicates

When distributing the NVIDIA strategy of expanding Amnesty International from databases to the material world, Huang has announced an important partnership with General Motors, “Building a Fleet in the future in the future.”

“GM has chosen NVIDIA to partnership with them to build a future self -driving fleet,” Huang announced. “The time of independent cars has arrived, and we look forward to building with GM AI in all three areas: AI to manufacture, so that they can revolutionize the way they are manufactured;

This partnership is a major vote on confidence in the staple of independent car technology in NVIDIA from the largest automotive manufacturer in America. Huang noted that NVIDIA is working on self -driving cars for more than a decade, inspired by AlexNet’s advanced performance in computer vision competitions.

“At the moment I saw Alexeant was an inspiring moment, such an exciting moment, I caused us to decide to continue building self -driving cars,” Huang recalls.

Besides General Motors’s partnership, Nvidia Halos, described as a “comprehensive safety system” for independent vehicles. Huang emphasized that safety is a priority “rarely with any interest” but requires technology “from silicon to systems, system, algorithms, and methodologies.”

Car ads extend on access to NVIDIA from databases to factories and vehicles, putting the company to capture value throughout the artificial intelligence staple and through multiple industries.

Architectural Engineer of the Artificial Intelligence law: The strategic development of NVIDIA behind the chips

GTC 2025 has unveiled NVIDIA from the GPU manufacturer to AI’s infrastructure. Through a road map from Blackwell to Rubin, Huang indicated that NVIDIA will not give in to its calculation, while its axis towards an open source program (Dynamo) and the Groot N1 models admit that the devices alone cannot secure their future.

NVIDIA has intelligently reformulated the Deepseek efficiency challenge, as it argues the most efficient models with a larger account in general with the expansion of the AI ​​scope – although investors have been skeptical, sending stocks to the decrease in shares despite the comprehensive road map.

What distinguishes NVIDIA from seeing Huang behind silicon. The robots initiative is not only related to selling chips; It comes to creating new computing models that require huge mathematical resources. Likewise, the GM partnership places NVIDIA at the AI ​​Car Intelligence Transformation Center through manufacturing, design and vehicles itself.

Huang’s message was clear: NVIDIA is competing for the vision, not just the price. Since the account extends from data centers to physical devices, NVIDIA is betting that controlling full artificial intelligence – from silicon to simulation – will know the next limits of computing. In the world of Huang, the revolution of artificial intelligence has just started, this time, coming out of the server room.