Nvidia releases new AI technologies amidst design issues and changes in the sector
Nvidia CEO Jensen Huang presented the company's latest innovations in artificial intelligence hardware and software during its annual developer conference in San Jose, California, as part of efforts to strengthen its standing in a swiftly changing...
TroibNews 
Nvidia CEO Jensen Huang presented the company's latest innovations in artificial intelligence hardware and software during its annual developer conference in San Jose, California, as part of efforts to strengthen its standing in a swiftly changing market. The announcements encompassed new chips, networking solutions, and robotics frameworks, highlighting Nvidia's commitment to meeting the increasing demand for efficient AI training and inference.
Huang revealed the Blackwell Ultra graphics processing unit, which is set for release in the latter half of 2025, and stressed its increased memory capacity designed to handle larger AI models. Despite this, the current Blackwell products are experiencing manufacturing delays attributed to a design flaw, complicating their rollout as the industry shifts its focus from training AI systems to their deployment for real-world inference tasks.
Additionally, Nvidia announced the Vera Rubin computing system, which integrates a custom-designed processor with next-generation GPUs to exceed the capabilities of the Blackwell architecture. Expected to launch in late 2026, Vera Rubin will be succeeded by the Vera Rubin Ultra in 2027 and the Feynman architecture in 2028. Named after groundbreaking astronomer Vera Rubin, this system is optimized for hyperscale AI workloads and boasts enhanced chip-to-chip data transfer speeds, essential for handling complex models.
For developers, Nvidia introduced DGX personal AI computers powered by Blackwell Ultra chips, produced in collaboration with partners like Dell, HP, and Lenovo. These desktop systems aim to compete with high-end consumer devices, facilitating local inference of large models. "This is what a PC should look like," Huang stated while showcasing a motherboard during his presentation.
Software updates featured Dynamo, a free tool designed to enhance AI reasoning, alongside Isaac GR00T N1, a robotics framework that employs a dual system model for "fast and slow thinking." Developed in partnership with Google DeepMind and Disney Research, GR00T N1 incorporates Newton, an open-source physics engine aimed at improving robot simulation.
Despite Huang’s confident claims of Nvidia being "well positioned" for the shift towards inference-heavy AI workloads, lingering skepticism among investors was evident. Following the presentation, shares dropped 3.4 percent, reflecting worries about competition and production delays. Huang addressed these concerns by asserting that agentic AI and reasoning-driven tasks might necessitate "100 times more computation" than was previously expected.
Thomas Evans contributed to this article for TROIB News
Huang revealed the Blackwell Ultra graphics processing unit, which is set for release in the latter half of 2025, and stressed its increased memory capacity designed to handle larger AI models. Despite this, the current Blackwell products are experiencing manufacturing delays attributed to a design flaw, complicating their rollout as the industry shifts its focus from training AI systems to their deployment for real-world inference tasks.
Additionally, Nvidia announced the Vera Rubin computing system, which integrates a custom-designed processor with next-generation GPUs to exceed the capabilities of the Blackwell architecture. Expected to launch in late 2026, Vera Rubin will be succeeded by the Vera Rubin Ultra in 2027 and the Feynman architecture in 2028. Named after groundbreaking astronomer Vera Rubin, this system is optimized for hyperscale AI workloads and boasts enhanced chip-to-chip data transfer speeds, essential for handling complex models.
For developers, Nvidia introduced DGX personal AI computers powered by Blackwell Ultra chips, produced in collaboration with partners like Dell, HP, and Lenovo. These desktop systems aim to compete with high-end consumer devices, facilitating local inference of large models. "This is what a PC should look like," Huang stated while showcasing a motherboard during his presentation.
Software updates featured Dynamo, a free tool designed to enhance AI reasoning, alongside Isaac GR00T N1, a robotics framework that employs a dual system model for "fast and slow thinking." Developed in partnership with Google DeepMind and Disney Research, GR00T N1 incorporates Newton, an open-source physics engine aimed at improving robot simulation.
Despite Huang’s confident claims of Nvidia being "well positioned" for the shift towards inference-heavy AI workloads, lingering skepticism among investors was evident. Following the presentation, shares dropped 3.4 percent, reflecting worries about competition and production delays. Huang addressed these concerns by asserting that agentic AI and reasoning-driven tasks might necessitate "100 times more computation" than was previously expected.
Thomas Evans contributed to this article for TROIB News
Discover more Science and Technology news updates in TROIB Sci-Tech
