NVIDIA founder and CEO Jensen Huang delivered the first keynote address of COMPUTEX 2025 in a boundary-pushing presentation on how artificial intelligence will transform every global industry, with NVIDIA and Taiwan at the center.
In front of a full-capacity crowd at Taipei Music Center, Huang announced that NVIDIA is no longer “only” a technology company, but “an AI infrastructure company, one that is essential all over the world.”
Huang equated NVIDIA’s role to that of a new AI factory, whereproducing tokens, the units of data that AI models process, will be even more important as companies seek to maximize efficiency.
Huang’s keynote pushed the envelope on NVIDIA’s major innovations in recent years, from an expanding library of accelerated computing libraries and collaborations with partners around the world, to a new generation of highly advanced products that seek to make AI increasingly ubiquitous and user-friendly.
Huang announced that NVIDIA’s Grace Blackwell generationwas in full production and fully online as of earlier this year, and that in Q3 of this year, NVIDIA will release an upgraded GB-300 generation. The new chips have 1.5x the inference performance, 1.5x the HBM, and twice the networkingcapabilities, all with the same physical dimensions.
Compared with past supercomputers, Huang said that NVIDIA has achieved 4,000 times the performance in just six years, blowing Moore’s Law out of the water.
Compact supercomputers were a highlight of the keynote, as Huang introduce a whole new product category. The DGX Spark to be available in just a few weeks, while the DGX Station is slated for release later this year.
The DGX Spark, which was able to fit in the palm of Huang’s hand, boasts 1 Petaflops (PFLOPS) of computing.
On the other hand, the DGX Station is the “most performance you can get out of a wall socket,” Huang said, joking that users should not run the microwave at the same time.
Using Grace Blackwell Ultra Superchips, the DGX Station has 20 PFLOPS AI performance, 800 GB of Unified System Memory, and an 800 GB/s network connection, enabling it to run a 1 trillion parameter AI model.
Huang also presented NVIDIA’s breakthroughs in robotics, including a new platform for training robots in digital simulations that is to transform industries like self-driving cars and manufacturing.
Much of the presentation highlighted the role that Taiwan plays in NVIDIA’s innovation, as Huang highlighted breakthroughs by Taiwan-based partners including TSMC, MSI, and ASUS.
As a grand finale, Huang concluded his speech by announcing a new, state-of-the-art office in Taipei’s Beitou-Shilin area namedNVIDIA Constellation.
“You brought AI to the world,” Huang told the Taipei audience, “and now AI will transform everything you do.”
This is “an extraordinary opportunity for Taiwan,” he added.
