Nvidia might be one of the most valuable companies in the world, but according to its CEO, Jensen Huang, he's actually the "chief revenue destroyer."
Huang gave himself that tongue-in-cheek title on Tuesday during his keynote at GTC 2025, Nvidia's annual developer conference. He riffed on the company's previous-generation graphics processing unit, Hopper.
"There are circumstances where Hopper is fine... That's the best thing I can say about Hopper," Huang said during the keynote.
While Huang's comment illustrates how far the company has come since Hopper's release in September 2022, it also contains some truth.
Huang and Nvidia officially unveiled Blackwell Ultra and Rubin Ultra AI chips at GTC, which are set to come out in late 2025 and 2026, respectively, and according to Huang, they're poised to move AI infrastructure forward in a big way.
Blackwell Ultra and Vera Rubin AI chips
At the center of Nvidia's GTC keynote are two chips: the Blackwell Ultra GB300 and the Rubin Ultra.
Both "superchips" are designed to power the increasingly complex workload of AI factories — data centers that represent the backbone of large language models like ChatGPT and Perplexity.
Although the two chips' power details were sparse, Nvidia says that its Rubin Ultra chip will be 3.3 times more powerful than Blackwell Ultra.
Nvidia reps told reporters during a briefing before the keynote that its Blackwell Ultra chip delivers 20 petaflops of AI performance, the same as last year's Blackwell chip. However, it expands memory, offering 288GB instead of the previous generation's 192 GB.
Nvidia also claims that the GB300 chip brings 1.5x more AI performance than the NVIDIA GB200 and "increases Blackwell’s revenue opportunity by 50x for AI factories" compared to data centers built around Nvidia's Hopper chip.
We designed Blackwell Ultra for this moment — it’s a single versatile platform that can easily and efficiently do pretraining, post-training, and reasoning AI inference.”
Nvidia CEO Jensen Huang
That's a jargon-heavy way of saying that Nvidia is positioning Blackwell Ultra and Rubin as significant upgrades to last year's technology, which has to do with their ability to reason.
“AI has made a giant leap — reasoning and agentic AI demand orders of magnitude more computing performance,” Huang, said. “We designed Blackwell Ultra for this moment — it’s a single versatile platform that can easily and efficiently do pretraining, post-training, and reasoning AI inference.”
That's all well and good, but what kinds of things does Nvidia actually envision chips of this magnitude powering? If GTC is any indication, kind of a lot.
Self-driving, robots, and more
Among the significant innovations Nvidia's AI infrastructure enables is self-driving vehicles.
Huang unveiled a partnership with GM that aims to crack the code on truly self-driving cars, stating on stage, "The time for autonomous vehicles has arrived.”
According to Huang, GM will use Nvidia's AI models, particularly for "simulation and validation."
GM will now use Nvidia's Omniverse platform to create "digital twins" of assembly lines — digital simulacra of real-life environments used for training.
With those, GM says it will conduct virtual testing and production simulations to drive plant efficiency.
That initiative will also include training for robotics platforms that GM already uses for material handling, transport, and precision welding.
Huang also discussed Nvidia's efforts regarding robots, specifically its previously announced Groot model, which it calls a "foundational model" for humanoid robots.
Nvidia says its N1 model is "the world’s first open, fully customizable foundation model for generalized humanoid reasoning and skills" and added it's the first of many models that will be pre-trained and readied for use in humanoid robots worldwide.
Major companies in the humanoid robotics game have already jumped at the chance to incorporate Groot into their hardware, including Agility Robotics, Boston Dynamics, Mentee Robotics, and NEURA Robotics.
The incentive to use Groot is clear: Nvidia's model is pre-trained using Nvidia's hardware, which will reduce the work that robotics companies have to undertake before deploying their robots for different tasks.
Huang even appeared on stage with a robot to hammer the point home.
But wait, there's more
Believe it or not, Nvidia didn't stop at announcing Blackwell Ultra and Rubin; it also alluded to future hardware coming in 2028 named "Feynman."
What Feynman has in store is anyone's guess, but Nvidia has previously committed to year-over-year updates to its chips despite problems with producing and implementing them.
With that cadence set in stone, Nvidia will have to make more powerful and efficient chips and convince companies using its hardware that they need new chips.
In February, Nvidia's stock price was significantly hit after DeepSeek released an AI model that delivers ChatGPT-like efficiency while using much less power.
DeepSeek and its model undermined some confidence that companies will need the power-hungry AI factories of the future that Nvidia has pitched, but whether that's true remains to be seen.
For now, we can buckle up and expect more AI hardware in the coming year.
More from Laptop Mag
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
