- Nvidia CEO Jensen Huang sets out his future vision at Computex 2025
- Nvidia looks to move into building large scale AI infrastructure
- Huang announced new products including NVLink Fusion and Grace Blackwell NVL72 chip
It’s fair to say that Nvidia CEO Jensen Huang’s Computex 2025 keynote was packed with some ambitious plans for the global tech industry.
Unsurprisingly, AI was at the forefront of not just the keynote, but the whole conference, as the world moves into what Huang called the “Age of AI”.
In his address, new Nvidia product launches, including the NVLink Fusion and the Grace Blackwell NVL72 chip, came alongside an ambitious road map for the development of AI infrastructure.
A new phase
Huang started by noting Nvidia donated DGX1 to OpenAI, underscoring the company’s place at the core of the AI Industry.
Since then, though, he argued the industry has evolved, and Nvidia is “not a technology company only anymore, in fact, we’re an essential infrastructure company”.
Huang explained that just like electricity and the internet, AI, or “intelligence” will move from being a useful tool, to an essential part of life; “In 10 years time you will look back and you will realize that AI has now integrated into everything and in fact we need AI everywhere.”
As such, Nvidia isn’t building data centres anymore. Well, not in the same way. Huang described data centres as ‘data centres of the past,’ that provide information and storage. These are similar to what he envisions in that they come from the same industry, but in the future, these will “emerge as something completely different”.
These new AI data centres will be less like the data centres we know now, and more like “AI factories”, Huang argued.
These factories will produce tokens, and they represent an AI infrastructure industry that will be measured in the trillions of dollars.
“This is the exciting future that we’re undertaking,” he explained.
Agentic AI
Huang talked about generative AI as a “universal translator”, which can, “translate from anything to anything else if we can simply tokenize it to represent the bits of information.” After the breakthrough of generative AI, we got intelligence.
“Intelligence is much more than just what you’ve learned from a lot of data that you’ve studied,” he noted. “Intelligence includes the ability to reason to be able to solve problems that you’ve not seen before to break it down step by step to maybe apply some rules and theorems to solve a problem you’ve never seen.”
This leads us to Agentic AI – a topic we’ve seen grow increasingly popular in the last year.
“Agentic AI is basically; Understand, think, and act.” Huang explained. “Agentic AI is basically a robot in a digital form. These are going to be really important in the coming years, we’re seeing enormous progress in this area.”
Moving forward, AI is going to be thinking incredibly fast, as Huang noted, “what used to be oneshot AI is now going to be thinking AI, reasoning AI, inference time scaling AI and that’s going to take a lot more computation.”
NVLink Fusion
This brings us to the Grace Blackwell NVL72, “a thinking machine” which can scale up computers, building larger computers that are “beyond the limits of semiconductor physics.”
In Q3 of 2025, and in Q3 of every year, Huang promises to increase the performance of the platform, “every single year like rhythm”.
And this year, that means an upgrade to Grace Blackwell GB300. This is the same architecture, physical footprint, and electrical mechanics, but with upgraded chips, with 1.5x more inference performance, 1.5x more HBM memory, and 2x more networking – with much higher overall system performance.
“Nvidia has been scaling computing by about a million times every 10 years and we’re still on that track,” Huang.
That’s where the world’s fastest switch, the NVLink comes in. Huang took the time in his keynote to announce a new version of NVLink tech, NVLink Fusion, which is aimed at helping other chip designers build powerful custom AI systems with multiple chips linked together.
This is significant, as so far Nvidia has only offered complete computer systems built using its own components – but with the NVLink Fusion, customized AI infrastructure can be built by organizations mixing Nvidia GPUs and custom components, allowing for much more flexible architectures.
Nvidia leads the way
Huang also announced that in partnership with the Taiwanese Government, Foxconn, and TSMC, Nvidia is going to build the first “giant AI supercomputer, here for the AI infrastructure in the AI ecosystem of Taiwan.”
Nvidia is leading the charge in AI chips, and the continued investments into infrastructure and components outlines the ambitions for the company to widen its market share of a range of technologies, he claimed.
Opening up the Nvidia ecosystem will aim to help more chip makers to take advantage of the data centre building trend and growing capabilities.
Finally, for corporations without in-house building capabilities, Nvidia is offering detailed blueprints to accelerate the process of building AI factories and infrastructure, helping remove the friction and ensure a seamless transition into the coming “Age of AI.”
