Nvidia’s GPU Technology Conference, or GTC, ran over a week between March 17-21 in San Jose, California. On the second day of GTC, Nvidia’s chief executive Jensen Huang gave a keynote speech where he announced new AI hardware and software along with several technology partnerships.
While Huang dropped a host of big numbers around artificial intelligence’s potential, none were bigger than his prediction that Nvidia’s data center infrastructure revenue will hit $1 trillion by 2028.
Here are six takeaways from the keynote delivered by Huang, co-founder and CEO of Nvidia, at the GTC 2025 conference:
1. Nvidia market performance
Huang pointed to surging demand for Nvidia GPUs. The company shipped 3.6 million Blackwell GPUs to America’s four largest cloud service providers—Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), and Meta (META)—in its 2025 fiscal year, which ended in late January. That’s up from 1.3 million Hopper GPUs the year before.
2. Speed, scale in AI inference
Nvidia introduced its latest line of Blackwell chips that mark a “leap in inference.” These chips belong to the Blackwell Ultra GB300 family of semiconductors which will have 1.5 times the memory. Their expected launch is in the second half of 2025.
i. The GeForce 5090
Powered by Blackwell architecture, this is a new graphics card that is 30% smaller in volume and 30% better at dissipating energy than the GeForce 4090.
READ: Nvidia rises to second most valuable company after Q4 earnings surge (February 27, 2025)
ii . Vera Rubin.. to replace Blackwell?
Nvidia also introduced “Vera Rubin,” a processor named after an American astronomer who discovered dark matter which is an invisible form of matter that plays a crucial role in the formation of galaxies.
Vera is the company’s first custom CPU design, and it is expected to be twice as fast as the CPU on the Grace Blackwell chips from 2024. Vera and Rubin (a GPU), when combined, can manage 50 petaflops during inference, which is more than the 20 petaflops on the current Blackwell chips. (A petaflop, or PF, is a unit of measurement for computer processing speed. Rubin also comes with 288 gigabytes of fast memory). The expected launch is in the second half of 2026.
iii. Rubin Ultra
Following this, they plan to launch an even more advanced version called Rubin Ultra NVL576. The Rubin Ultra model is projected to deliver an impressive 15 exaflops (or 15000 petaflops) of computing power and feature a massive 4,600 terabytes per second of scale-up bandwidth. The expected launch for this would be the second half of 2027.
3. New partnerships
Nvidia also revealed a number of upcoming partnerships, including with General Motors, to train AI manufacturing models. The goal is to build the automaker’s “future self-driving car fleet,” Huang said.
Another partnership includes one with Yum brands (Taco Bell’s parent company) to bring AI to restaurants from improving back-end operations to enhancing drive through experience at popular joints like KFC and Taco bell.
Huang also announced the launch of Nvidia Spectrum X. As of now, together CISCO and Nvidia are developing a Secure AI Factory to dramatically simplify how enterprises deploy, manage, and secure AI infrastructure at any scale with help of Spectrum X.
i. Nvidia Halos
This is a safety system that integrates NVIDIA’s automotive hardware, software and AI research in the safety of autonomous vehicles.
4. Rise in Agentic AI
“The amount of computation we need as a result of agentic AI is 100x more than last year,” according to Huang.
Huang also refuted DeepSeek’s impact analysis of its R1 AI model, which suggested that future software advancements could reduce the need for chips and powerful servers. He countered this by citing a $1 trillion inflection point in AI computing demand, driven by the rise of reasoning AI and agentic AI.
The CEO also revealed that the company will be open-sourcing its cuOpt decision optimization platform for industry-wide applications such as supply chain management, last mile deliveries and more to enterprises.
i. Llama Nemotron
While discussing the future of agentic AI, the CEO also announced the open Llama Nemotron family of models with reasoning capabilities. These models have been designed to offer developers and enterprises a business-ready foundation for developing AI agents that can work autonomously or as connected teams to accomplish complex tasks. This refinement process boosts accuracy of the models by up to 20% compared with the base model.
READ: China disrupts AI market with DeepSeek: A better, cheaper version of ChatGPT? (January 27, 2025)
Additionally, Nvidia said members of the Nvidia Developer Program can access them for Llama Nemotron for development, testing, and research. Enterprises can use Nvidia AI Enterprise on accelerated data center and cloud infrastructure to run Llama Nemotron NIM microservices in production.
5. Physical AI
“The ability to understand the physical world, the three-dimensional world, is what’s going to enable a new era of AI, which we call Physical AI, and it’s going to enable robotics,” Huang said. The CEO envisions robotics as a $10 trillion industry; he also claimed a shortage of 50 million workers.
In order to accelerate robotics, Nvidia has introduced Isaac GR00T N1, the first open and customizable foundation model for humanoid reasoning. Newton open-source physics engine, developed with Google DeepMind and Disney Research will be synergised with Isaac-GR00T N1.
The company also introduced new Nvidia Cosmos world foundation models, offering greater control over AI-driven world generation.
6. Building more AI centers, Nvidia Photonics
Huang spoke about establishing a digital twin which replicates a real AI data center before it is built to control millions of variables like retrofitting, space management and energy consumption.
Then he moved on to mention the launch of Nvidia Photonics. This invention enables AI factories to connect millions of GPUs across sites while drastically reducing energy consumption and operational costs. Nvidia has achieved the fusion of electronic circuits and optical communications at massive scale.
i. Nvidia Dynamo
Nvidia also unveiled Nvidia Dynamo, an open-source inference software for accelerating and scaling AI reasoning models in AI factories at the lowest cost and with the highest efficiency.
ii. Quantum Day
On the penultimate day of the conference, Huang called the day “Quantum Day” and admitted his “mistake” back in January, that quantum computers were still 15 or 20 years away. He refuted this at the GTC conference by saying, “this is the first event in history where a company CEO invites all of the guests to explain why he was wrong.”
READ: Nvidia loses $600 billion amid DeepSeek rise (January 28, 2025)
He also announced the launch of a new quantum computing lab called The Nvidia Accelerated Quantum Research Center in Boston.
Huang even interviewed executives from quantum computing firms. Needham Analyst Quinn Bolton told clients that Huang’s insinuation about quantum’s branding was “one of the most contested portions” of the event. This might have even led to the current bearish market for Nvidia despite the new launches and long term plans.
Six takeaways from Nvidia’s GTC 2025: Was it an ‘AI Super Bowl’?
