I don’t think it is an understatement to say that Nvidia wants to be the company that powers the artificial intelligence universe we may one day live in permanently.
The opening keynote at Nvidia GTC 2025colloquially referred to as the Woodstock of AI, saw CEO Jensen Huang extol the virtues of an AI ecosystem powered by hardware – and software – from the second most valuable company in the world.
We were re-introduced to the concept of AI factories, which are essentially gigantic data centers and where, Nvidia’s world, tokens generated are equivalent to revenue: the presentation mentioned a price, $1 trillion.
The cost of building data centers, like mega-farms, increases exponentially every generation as they get more complex, and require more power to house even more compute.
The lofty Stargate project has a budget of $500 billion and Apple has already committed to spending $500 billion over the next few years, a large chunk of which will be on data centres.
Depending on what analysts you talk to, the global data centre capital expenditure is expected to reach $1 trillion either in 2027 (PWC) or 2029 (Dell’Oro).
So that gives you a measure of Nvidia’s extremely ambitious targets for itself and for humanity.
A big piece of that puzzle that Nvidia introduced at the keynote is its Photonics switch systems which will allow those AI factories to “scale to millions of GPUs”, seemingly within the same physical perimeter.
Big, huge numbers
The biggest number of them all in this presentation was when Rev Lebaredian Nvidia’s VP, Omniverse and Simulation Technology, presented a slide about how Physical AI is transforming $50 trillion industries.
These span across two billion cameras, 10 million factories, two billion vehicles and, Lebareian added ‘Future Billion humanoid robots’.
Yes, it’s not just about robots, but humanoid robots, millions of them. Robots, in general, have been around for decades in many shapes and forms.
Nvidia, however, envisions a world where billions of robots are built on what it calls, its “Three Computers”: Pre and post-training (done via DGX), simulation and synthetic data generation (done via Omniverse) and Runtime (done using AGX).
The world of the future may be home to “multiple types of humanoid robots” that will be able to post-train – while on the job so to speak – with real or synthetic data, allowing them to learn new skills or enhance their knowledgebase on the fly, literally at (or near) the speed of light.
And the first of them may well be based on its new GR00T N1 humanoid robot foundation model, developed in partnership with Google Deepmind and … Disney Research.
(Ed: One can not see the tie-in with GR00T from Guardians of the Galaxy and Wall-E, a potentially prescient movie about the impact of robotics and AI on humanity).
Where does this leave us? With a clear roadmap to a world where AI becomes an indispensable part of what it means to be human, perhaps the most important utility of them all. Never, in the history of humanity, will so many depend on so few.
Oh and the other acronym (AGI) was not mentioned during the presentation, not once.